Illlll'llllllllllllIlllllllllllilllll'llll 3 1293 01025 3981 This is to certify that the dissertation entitled Agroecosystem Analysis in a Semi-arid, Smallscale Farming System of Zimbabwe: A Systems Approach presented by Timothy J. Lynam has been accepted towards fulfillment of the requirements for Dnrrnr' 0. degree in Ebj Igsgpb¥ M a jdr professor %J.M4/z MS U is an Affirmative Action/Equal Opportunity Institution 0-12771 - LIBRARY Michigan State University PLACE IN RETURN BOX to roman this checkout from your record. TO AVOID FINES rotum on or before date duo. DATE DUE DATE DUE DATE DUE $53 c*£§,f‘t.-g';;;:r ' [Rat u l. . .. f mg; M3 5;) NT. 2 3 2008 MSU ION! ”“ AGROECOSYSTEM ANALYSIS IN A SEMI-ARID, SMALLSCALE FARMING SYSTEM OF ZIMBABWE: A SYSTEMS APPROACH BY Timothy Jen Philip Lynam A DISSERTATION Submitted to Michigan State University In partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Crop and Soil Sciences 1 993 ABSTRACT Effective agroecosystem analysis is made difficult by the lack of suitable philosophical and methodological approaches and the difficulty of applying performance criteria, such as sustainability, to agroecosystem behavior. A research methodology for the analysis of agroecosystem behavior in data scarce environments was described. The approach was used to examine an agroecosystem in the mid-Zambezi Valley. A questionnaire survey and RRA techniques were used to develop a general understanding of the needs, resources and constraints households faced in satisfying their needs. Thereafter an extended analysis was conducted with the objective of developing a computer simulation model of the agroecosystem. Five male villagers were elected and three women volunteered to act as village representatives (VRs) in the analysis. Data collection methods were developed and then used with the VHS to identify. and weight, by relative importance, major household needs and the production enterprises used to satisfy those needs as well as all inputs to and outputs from each production enterprise. Household needs and yield probability density functions derived from the VHS were used to develop a computer simulation model of the agroecosystem. Up to 300 households were randomly placed on a raster based GIS image of the Masoka agroecosystem. Each cell of the landscape represented one acre. Households were allocated cells (fields) around their house sites and on soils adjacent to the Angwa River. The model simulated the productive activities of up to 300 households, updating variables in each cell and for each household once a season. Households could be allocated to one of four production strategies which determined the crops as well as the proportion of household land that was planted to each crop each year. Households could employ labor from other households to make up for deficits failing which their yields were reduced. A partial budget format was used to estimate returns to land, labor and initial investment. Soil erosion was modeled using the SLEMSA erosion model. The effects of changing important model inputs and parameters on model response variables were examined. Factors that had notable effects on the proportion of deficit households, yields and returns to land. labor and initial investment were rainfall, the land area available and the cash needs required by households. Copyright by Timothy Jan Philip Lynam 1994 To Limited, who never got to see a rhino and to the spirit of CAMPFIFiE. ACKNOWLEDGEMENTS In a study spanning two continents and of considerable breadth l have had to rely on the goodwill and assistance of many people and organizations. I am especially grateful to the WWF Multispecies Project, for providing me with the institutional and financial support for this research and to the Project Leader. Dr. D.H.M Cumming, whose enthusiasm, mentorship and unfailing support were crucial to every aspect of my research. I would like to thank the Rotary Foundation for the generous scholarship which made it possible for me to attend MSU. I would like to thank Dr. J.T. Ritchie, my advisor, who provided the impetus that brought me to MSU, the institutional support while I was at MSU and for leaving me to go my own way. My committee members Dr. Campbell, Dr. Crawford, Dr. Manetsch, Dr. Robertson, Dr. Paustian proved invaluable as teachers, advisors and friends. I am very grateful to them all. Like so many Zimbabwean colleagues I am indebted to Peter Frost. His remarkable stream of ideas, criticisms, support and friendship have enriched my work and my abilities. Working with Gift Chisunga, George Kanoderuka, Mai Kajengo, Mai Tekenyeka, Mai Kowe, Albert Fero. Saraoga Mbisvi and lnocent Fakero was both a delight and a pleasure. Without their enthusiasm and dedication I could not have achieved a fraction of what we did. I am grateful to Henry Elwell for his invaluable help with erosion modeling. for doing the erodibility tests for me and for the many ideas and thoughts he has so generously provided. All errors and omissions in the erosion model used here are however, my responsibility! Sarah Funkhouser made a great difference to my life and to my work, I am grateful. Bill Verboem and Katherine Verbeek were invaluable in helping with soil mapping and classifications. I am particularly grateful to Stanford Mushiri of the Chemistry and Soil Research Institute for the tremendous assistance he gave in having my soil samples analyzed. Bill Derman, Stuart Gage. Sharlene Rhines, Karen Potter-Witter, Brian Baer and Ivan Bond provided invaluable support and help over the years. I would like to thank the staff at WWF MAPS Project Office Harare, for all they have done and continue to do. To all those I have inadvertently missed, thank you. TABLE OF CONTENTS LIST OF TABLES ........................................... vii LIST OF FIGURES ............................................ x ABBREVIATIONS ........................................... xiv Chapter 1. ................................................ 1 Background and justification ................................. 1 INTRODUCTION ...................................... 1 BIBLIOGRAPHY ...................................... 10 Chapter 2. ................................................ 12 Background data on the Masoka agroecosystem ................. 12 INTRODUCTION ...................................... 12 Study site description .............................. 13 Summary ....................................... 26 BIBLIOGRAPHY ........................................... 28 Chapter 3. ................................................ 30 Towards a general understanding of the Masoka agroecosystem ..... 30 INTRODUCTION ...................................... 30 METHODS ........................................... 35 RESULTS AND DISCUSSION ............................ 40 Household needs for food and cash crops .............. 41 Production ...................................... 43 Income generating potential ......................... 47 Cropland areas required by households to meet needs ..... 50 Implications ..................................... 52 CONCLUSIONS ....................................... 54 BIBLIOGRAPHY ...................................... 57 Chapter 4. ................................................ 61 iv Needs analysis, system Identification and problem formulation In the Masoka agroecosystem ................................ 61 INTRODUCTION ...................................... 61 METHODS ........................................... 68 Selection of community representatives ................. 68 General approach to recording local knowledge .......... 70 Identification of key individuals and groups within the community ................................. 72 Identification of household needs ..................... 73 Inputs to, outputs from and measures of performance of, the Masoka agroecosystem ....................... 74 Methodological validity and data accuracy ............... 80 RESULTS ........................................... 85 Key actors and household classes .................... 85 Household needs ................................. 88 System identification ............................... 94 Maize production ................................. 95 Cotton production ................................ 99 Sorghum production .............................. 102 Summary of inputs to and losses from crop production sub- systems ................................... 102 Rainfall analysis .................................. 105 Labor requirements for each crop enterprise ............. 112 Soils and soil fertility ............................... 118 Crop yields ..................................... 122 Performance measures ............................. 130 Problem formulation ............................... 143 SUMMARY AND CONCLUSIONS ......................... 150 BIBLIOGRAPHY ...................................... 153 Chapter 5. ................................................ 160 Simulation model: development, structure and testing ............. 160 INTRODUCTION ...................................... 160 MODEL DESCRIPTION ................................. 165 Overall approach ................................. 165 Model overview .................................. 167 Rainfall generating model (RAINGEN) .................. 169 Initializing model (INITIAL) ..................................... 171 Household size generation and household field allocation ............................. 171 Cropping history ............................ 174 Vegetation cover ............................ 174 Outputs from INITIAL ......................... 176 Agroecosystem model (MASOKA1) .................... 178 The STORAGE sub-system ..................... 178 The ALLOCATIVE sub-system ................... 182 The LABOR sub—system ....................... 186 The CROP subosystem ........................ 191 The EROSION sub-system ..................... 200 The ECONOMIC sub-system ................... 205 The OUTPUT sub-system ...................... 207 METHODS ........................................... 213 Verification, validation, usefulness ..................... 213 Sensitivity analysis ................................ 215 VERIFICATION AND VALIDATION ......................... 222 Rainfall and seasons ............................... 222 Initial conditions .................................. 222 Maize yields ..................................... 227 Cotton yields .................................... 229 Sorghum yields .................................. 230 Erosion ........................................ 232 RESULTS ........................................... 233 Sensitivity analysis ................................ 233 Household needs satisfaction ................... 234 Crop yields ................................ 243 Economic response variables ................... 246 Erosion, erodibility and "fertility" .................. 253 SUMMARY AND CONCLUSIONS ......................... 257 BIBLIOGRAPHY ...................................... 262 Chapter 6. ................................................ 267 Overview and key Issues for further research .................... 267 INTRODUCTION ...................................... 267 OVERVIEW .......................................... 267 KEY ISSUES FOR FUTURE RESEARCH .................... 272 BIBLIOGRAPHY ...................................... 276 Table 2.1 Table 2.2 Table 2.3 Table 2.4 Table 3.1 Table 3.2 Table 3.3 Table 3.4 Table 3.5 Table 4.1 Table 4.2 LIST OF TABLES Major soil types of Kanyurira Ward, mapped at 1:250 000 (Anderson, 1987). .......................................... 15 Total revenues generated by wildlife in Masoka in each year since 1989, dividends paid per household and total revenues paid to households In Zimbabwe dollars (ZS). ............................... 23 Census population estimates of large herbivores in Kanyurira Ward, 1988 to 1992. .................................................. 24 Land types within the Masoka fenced area with Zimbabwean and USDA soil classifications. ..................................... 25 Household needs for maize. cotton and sorghum, by household size class. Data are mean _-I_-_ SEM, with n in parentheses. ............. 43 Maize, cotton and sorghum yields obtained by respondents for the 1990/91 season on three soil types. Data are mean .t SEM, with n in parentheses. ...................................... 45 Anticipated yields for hand cultivated maize, cotton and sorghum on three soils In good and bad rainfall seasons. Data are mean‘ i SEM, with n In parentheses. ...................................... 46 Perceived yields for tractor cultivated maize. cotton and sorghum on three soils in good and bad rainfall seasons. Data are mean 3; SEM, with n in parentheses. ................................... 47 Cropland areas (ha) required to satisfy household needs. Data are total area for household size class with the area per adult equivalent for each size class In parentheses. ...................................... 50 Subjective probabilities indicating VR perceptions of the probability of the number of rainfall events, of each rainfall type. occurring in any two week period of the rainy season ....................................... 107 Classification of ten rainfall seasons ending 1991/92 as good (G). average (A) or poor (P) by village representatives (VRs), a local informant and questionnaire survey (1988/89 to 1990/91 only). ............. 111 vii Table 4.3 Table 4.4 Table 4.5 Table 4.6 Table 4.7 Table 4.8 Table 4.9 Table 4.10 Table 4.11 Table 4.12 Table 4.13 Table 4.14 Table 4.15 Table 5.1 Table 5.2 Labor requirements (days ha“), for each activity of maize production in each month ofthegrowingseason. ........................... 113 Labor requirements (days ha") for each activity of cotton production for each month of the growing season. ........................... 114 Labor requirements (days he") for each activity of sorghum production, in each month of the growing season. ........................... 114 Relative importance weights (5 highest to 1 lowest) of tree and herb or grass species after a jecha field Is left fallow. ..................... 119 Mean (standard deviation in parentheses) yield estimates (kg ha' 1) for maize, cotton and sorghum, derived from household field maps and household respondent information. ........................ 123 Maize, cotton and sorghum yields reported for the mid-Zambezi Valley. . 123 Expected values for maize, cotton and sorghum yields (kg ha' 1) on three soils and for good, average and poor rainfall seasons. Source. possibility diagrams. ........................................ 126 Mean (standard deviation in parentheses) number of adults, number of children and years residence in Masoka for classes of household well- being. .......................................... 1 35 Mean (standard deviation in parentheses) for characteristics of households in each of four employment classes. ............... 136 Partial budgets for one hectare maize production enterprises with different input and pricing scenarios. 1991/92 season prices. ................. 138 Partial budgets for one hectare cotton production enterprises with different Input scenarios. 1991/92 season prices. ........................ 139 Partial budgets for one hectare sorghum production enterprises with different input and price scenarios. 1991/92 season price data. ...... 140 Expected returns (23 ha") to land, to labor and to Initial investment for maize, cotton and sorghum on each soil type. ................. 141 Erodibility values for cultivated and uncultivated jecha and bepe soils in Masoka. Source Elwell, personal communication. ..................... 202 Summary of output variables written to household output file (HOUSEHLDDUT). .................................. 209 viii Table 5.3 Table 5.4 Table 5.5 Table 5.6 Table 5.7 Table 5.8 Table 5.9 Table 5.10 Table 5.11 Table 5.12 Table 5.13 Summary of output variables written to SUMMARY.OUT and to MONTECAR.OUT. ................................... 212 Factorsandfactor levelsusedin sensitivity analysisofMASOKA1. . . .. 216 Goodness of fit test statistics for the distribution of numbers of children per number of adults. ................................ 23 Simulated maize yields (kg ha") for each soil and in each season. Data are means with standard deviations in parentheses. medians and the sample size. ...................................... 228 Simulated cotton yields (kg ha") for each soil and In each season. Data are means with standard deviations In parentheses, medians and the sample size. ...................................... 29 Simulated sorghum yields (kg ha") on jecha in each season. Data are means, with standard deviations in parentheses, medians and the sample size. ........................................... 231 Sensitivity of the mean of the 96 of households that were cash (ACDF), maize (AMDF) and sorghum (ASDF) deficit, each year, to changes in factor levels. ...................................... 236 Sensitivity of mean annual yields of maize (MZYLD), cotton (CTYLD) and sorghum (SRYLD) to changes in factor levels. ................. 241 The effects of factor level changes on the means of total retums (MZREI), returns to land (MZHA). labor (MZLAB) and to initial investment (MZINV) for maize. .................................. 247 The effects of factor level changes on the means of total returns (CTRET), returns to land (CTHA), labor (CTLAB) and to initial investment (CTINV) for cotton ......................................... 248 The effects of factor level changes on the means of total returns (SRRET), returns to land (SRHA), labor (SRLAB) and to initial investment (SRINV) to sorghum. ...................................... 249 ix Figure 2.1 Figure 2.2 Figure 2.3 Figure 2.4 Figure 3.1 Figure 3.2 Figure 3.3 Figure 3.4 Figure 4.1 Figure 4.2 Figure 4.3 Figure 4.4 LIST OF FIGURES Map of Zimbabwe showing the location of the study site ..... 14 Map showing the base of the Zambezi escarpment (600m). major road routes and important towns or settlements near Masoka . 16 Precipitation (mm). from Angwa Bridge (n=15) and evaporation (mm) from Muzarabani (n=6). Data are monthly means with bars showing SEM. ................................... 18 Map showing Masoka land types and game fence ......... 21 Estimated numbers of households in Masoka, 1985 to 1992 and predicted numbers of households through 2010 based on a linear and an exponential projection model. .......... 42 Household needs (ZS) and earning potential for large households over the period 1978 to 1992 ................ 49 Household needs (ZS) and earning potential for small households over the period 1978 to 1992 ................ 50 Arable land required and remaining under the population growth scenarios presented in Figure 3.1 and the land requirements presented in Table 3.5. Major actors influencing agricultural production in Masoka and their relative importance weights. Female VR perspective . 86 Major actors influencing agricultural production in Masoka and their relative importance weights. Male VR perspective . . . 86 Household classification and the relative importance weights of each household class. Male VR perspective. ............ 87 Household classification and the relative importance weights of each class. Female VR perspective. .................. 87 Figure 4.5 Figure 4.6 Figure 4.7 Figure 4.8 Figure 4.9 Figure 4.10 Figure 4.11 Figure 4.12 Figure 4.13 Figure 4.14 Figure 4.15 Figure 4.16 Figure 4.17 Household needs. with their relative Importance weights (RIW). as well as the major inputs to those needs with their RIW. Female VR perspective. ............................. 90 Household needs, with their relative importance weights (RIVV), for a large household as well as the major inputs to those needs with their RIW. Male VR perspective. .............. 91 Household needs, with their relative importance weights (RM), for a medium household as well as major inputs to those needs with their RIW. Male VR perspective. .............. 92 Household needs, with their relative importance weights (RIVV), for small households as well as major inputs to those needs with their RIW. Male VR perspective. .............. 93 Female VR perspective of what is required for, and what detracts from, the achievement of good maize yields. .............. 94 Male VR perspective on what is required for, and what detracts from, the achievement of good maize yields. .............. 95 Female VR perspective on what is required for, and what detracts from, the achievement of good cotton yields. ............ 100 Male VR perspective on what is required for, and what detracts from, the achievement of good cotton yields. ............ 101 Female VR perspective on what is required for, and what detracts from, the achievement of good sorghum yields. .......... 103 Male VR perspective on what is required for, and what detracts from, the achievement of good sorghum yields. .......... 104 Probability of a rainfall event of each of four rainfall types occurring in each week of the rainy season. ............. 108 Probability of a rainfall event lasting at least a certain time for each of the four types of rainfall. ................... 110 Increase in total tree biomass with time after field abandonment as a proportion of the maximum achievable in 15 years for major tree species growing on jecha. ....................... 120 xi Figure 4.18 Figure 5.1 Figure 5.2 Figure 5.3 Figure 5.4 Figure 5.5 Figure 5.6 Figure 5.7 Figure 5.8 Figure 5.9 Figure 5.10 Figure 5.11 Figure 5.12 Figure 5.13 Figure 5.14 VR derived performance measures to assess the liklihood of a household satliying its basic needs. ................... 131 Major components of the Zambezi Valley Agroecosystem Model (ZVAM) ................................... 166 Input and output files for the initializing model (INITIAL) ..... 175 Input and output files for the agroecosystem model (MASOKA1J77 Sequence of major components of MASOKA1 ............ 178 Flow chart illustrating sequence of events in STORAGE sub-system ..................................... 179 Flowchart of activities in the ALLOCATIVE sub-system ...... 181 The cumulative probability of leaving a jecha field fallow for households using income maximizing (MAXINC), variance minimizing (MINVAR), needs satisfying (SATIS) and environ- mentally protective (GREEN) production strategies ........ 184 Flowchart of major activities in the EMPLOY component of the LABOR sub-system. ............................... 185 Flowchart of major activities in yield correction component of the LABOR sub-system. .......................... 187 Flowchart of major activities in CROP sub-system. ......... 190 The proportion of respondents allocating each growing season from 1978/79 to 1991/92 to good, average or poor seasons plotted against the rainfall for that season measured at Angwa Bridge. Rainfall data from the Department of Meteorological Services, Harare. ................................. 192 Membership functions for good, average and poor seasons used In MASOKA1. .................................... 193 Example yield generation function from MASOKA1. Dependent variable is yield (kg ha ) and independent variable is the random variate p. ................................. 197 Flowchart of major activities in EROSION sub-system. ...... 204 xii Figure 5.15 Figure 5.16 Figure 5.17 Figure 5.18 Figure 5.19 Figure 5.20 Flowchart of major activities in ECONOMIC sub-system. . . . . 206 The cumulative proportion of a) actual (n=70) and b) simulated (n=1000) Masoka households cultivating a given area of non- riverland. ....................................... 223 The cumulative proportion of a) actual (n=70) and b) simulated (n=1000) Masoka households cultivating a given area of riverland. ....................................... 224 Causal loop diagram of the effects of increasing household cash needs on household maize and cash balances. .......... 237 Good (G), average (A) and poor (P) season membership functions used for +1 and -1 levels in sensitivity analysis. .......... 243 Total number of years each cell had been cultivated, summed over the landscape for +1 fallow practice factor level (min=0, max=10 years) and -1 factor level (min=5, max=7 years). Result of single simulation. ............................... 254 xiii ABBREVIATIONS AGRITEX ...... Department of Agricultural Technical and Extension Services CA ............................................ Communal Areas DMS .......................... Department of Meteorological Services FSRU .............................. Farming Systems Research Unit GOZ ................................... Government of Zimbabwe MAPS .................. Multispecies Animal Production Systems Project MLARR ............. Ministry of Lands, Agriculture and Rural Resettlement MZVRDP ................. Mid-Zambezi Valley Rural Development Project VR ........................................ Village Representative WWF ................................. World Wlde Fund for Nature xiv Chapter 1. Background and justification INTRODUCTION Two problems need to be overcome in analyzing the behavior of agroecosystems‘. The first of these Is the lack of a suitable and philosophically and methodologically coherent approach to the analysis of agroecosystems. The second is the problem of meaningfully applying performance criteria such as sustainabilty or stability in real world decision making. Generally speaking science is ill-prepared, both methodologically and philosophically, to solve real world problems of agroecosystems where the performance criteria are sustainability, stability or similar measures requiring a predictive understanding of agroecosystem behavior. MacRae et al. (1989) review several scientific barriers to achieving sustainable food production. They discuss the failings of the reductionist approach to scientific investigation, question Agroecosystems are defined as ecosystems in which basic biological processes are managed to obtain food and fibre as well as other goods and services. They Include social. managerial or decision making and economic as well as biophysical components. In Zimbabwe Communal Areas and at the scale of this analysis they incorporate a number of households. They include therefore, a political component as well. 2 scientific objectivity, particularly in fields of research such as agriculture, and the common belief that quantifying Is an essential precursor to rational evaluation of facts. There can be little doubt that reductionist science has been a powerful problem solving paradigm. It is limited, however, in its ability to deal with problems associated with systems and especially where multiple performance criteria are to be used (Bawden, 1991; MacRae et al., 1989). The perception of science as an objective and value free search for the truth has not been substantiated by research (MacRae et al., 1989; Busch and Lacey, 1983; Mahoney, 1979). The perception of some agricultural scientists that it is possible to remain objectively detached from social and economic processes Is a dubious assumption at best. Values influence what problems are addressed and which solutions are considered acceptable (or considered at all). These decisions have political, social and economic consequences that should be recognized. Failure to explicitly define values in agricultural research could result in the introduction of sources of bias not accounted for in subsequent analyses. Agricultural scientists rely on quantitative data to support or refute their assertions and are more likely to believe quantitative arguments (MacRae et al., 1989; Mahoney, 1976). Quantifying states and outputs in agroecosystems, with their tightly coupled bio-physical, social and economic sub-systems, is difficult. Reliance on quantified relationships could lead to emphases that would not be supported by giving equal weight to those system attributes that are difficult to 3 quantify and for which there might only be qualitative relationships. By quantifying relationships and states analysts also stand the risk of expressing a precision that is not a true reflection of their knowledge of the world. Philosophically, much of contemporary science uses the hypothetica- deductive approach in establishing whether or not assertions are false. For the most part scientists are concerned with the probability of making a Type I error (rejecting a true null hypothesis) but rarely examine the likelihood of making a Type II error (accepting a false null hypothesis). Yet, in the world of decisions, costs and benefits, these errors may have equally expensive consequences (MacRae et al., 1989; Officer and Dillon, 1968). Aversion to making Type I errors is a strong incentive to reduce problems to single variables to ensure that observed effects are due to factor changes. The assumptions imposed on observation, in order to use classical inferential statistics, constrain our ability to react to "scientific complexity" (Box and Tlao, 1992). The hypothetico-deductive methodology, as formulated by Popper (1968), also makes the assumption that, a priori, we know nothing whatsoever about the subject matter of interest (Stove, 1982). In basing decisions on this view a tremendous body of knowledge and previous experience are discarded. Given the complexity of the performance criteria used in comparing agroecosystems or agroecosystem outputs as well as the sometimes high costs of making either Type I or Type II errors, this seems a wasteful practice; in essence the tools used are dictating the nature of the problems that are addressed. 4 The systems approach is one that offers promise for dealing with the complex requirements of design and analysis in agricultural systems (Bawden, 1991). The soft systems approach of Checkland (Checkland, 1981; Checkland and Scholes. 1990) was specifically developed to deal with problems where neither the objectives nor the system boundaries are clearly defined, or perhaps even clearly definable. This may often be the case with agricultural systems. The soft systems approach does not stress the value of computer modeling as is the case in conventional or "hard" systems approaches. The systems approach is not without its critics however; Belinski (1976) is scathing in his attacks on the misuse of logic, mathematics and statistics in systems science and particularly when these methods have been applied to social and biological systems. Hoos (1972) is critical of the ability of the systems approach to adequately deal with public policy issues. She is even more damming of attempts to predict the future of social systems (Hoos, 1974). These criticisms are often valid and should not be ignored. Criticism is a necessary but, not sufficient, condition for change and improvement. Change also requires trying new techniques or taking what is deficient and improving it. So, whilst the systems approach is clearly not perfect it is a useful approach with which to start and from which to develop new approaches to agroecosystem analysis. The well developed and tested approaches to analysis, including the large body of knowledge on many aspects of modeling dynamic systems, offers a rich tool box with which to undertake the analysis of agroecosystem behavior. 5 Perhaps the most widely advocated approach to the analysis of agricultural systems in the developing world is the agroecosystem analysis approach of Conway (Conway, 1987, 1986. 1985). Agroecosystem analysis (after Conway) uses four system properties to describe system behavior. These are productivity, stability, sustainability and equitability. Grimm et al. (1992) provide a checklist of factors that should be explicitly defined when discussing ecosystem stability concepts; a checklist that appears equally appropriate for use with sustainability, productivity and equitability as it does with stability. These factors are: a) the level of description (whether individual. population or ecosystem for example); b) the variable of interest; c) the referential behavior of the variable of interest; d) the nature of the disturbance; e) the spatial scale; and f) the temporal scale. In the view of many systems’ theorists, systems exhibit what are called emergent properties; these are properties that are not deducible from observations of the individual parts of the system (Bawden and lson, 1992). If we accept these assertions then the selection of the level of analysis as well as the spatial and temporal scales of analysis are made very much more complex. Changes in factors at one level of analysis could lead to unexpected changes at others or an agroecosystem could be stable at one scale but not at another. Data suitable for a quantitative description of both the current state of the agroecosystem as well as those describing the behavior of the agroecosystem over space and in time are seldom likely to be available and particularly so in deveIOping countries. 6 These problems are compounded in practical decision making by issues of whose values need be considered and the large number of variables and performance criteria that need to be considered simultaneously. Policy makers and individual heads of household are unlikely to share the same set of values. Theoretically at least, the policy maker could be expected to act in the best interests of society whilst the household head could be expected to act in the best interests of the household. As analysts we are faced with the dilemma of selecting which set of values are to be used in the analysis. There do not seem to be any clear cut criteria for selecting one view of the world as opposed to another. In Conway’s approach (Conway, 1985. 1987) these issues are not addressed. The approach is orientated toward the analyst’s world view. Techniques that include local farmer participation are used for data collection and the farmers in the target agroecosystem appear to participate only as passive sources of data rather than active participants in an client orientated analysis. Conway (1985) admits that the properties he defines are more easily defined than measured and that "satisfactory methods of measuring sustainability still need to be found." In more recent applications of the methodology (Conway. 1987) a team of experts is used in a workshop format, to subjectively select patterns of space. time, flow and decisions, that are likely to reveal the key functional relationships that determine system properties. The agroecosystem analysis approach of Conway (1986, 1985) does not advocate predictive modeling. Evaluations of sustainability, stability and equitability 7 appear to be made subjectively by the team of experts. With scarcely any data at all, let alone time series data, this may be the only approach that is possible, yet it can lend itself to biased and erroneous interpretations. Under Zimbabwean communal area conditions, farmers and analysts are often from different cultures and backgrounds and share very different futures. Few of the analysts are likely to have spent more than a few weeks, let alone months, in a communal area farming system. Fewer still of the analysts will have futures that are more than cursorily linked to the outcome of the analysis. Given that the analysts and communal area farmers have neither values nor objectives in common and the analysts are not accountable to households for their decisions it appears unlikely that a team of professionals can reliably predict the sustainability, stability and equitability performances of an agroecosystem. It Is undeniable, that there are a number of philosophical and methodological tools available, which could be gainfully used to develop insights appropriate to the analysis of complex agricultural systems, particularly those in the developing world. There does not. however, appear to be a philosophically and methodologically coherent approach to agroecosystem analysis for use with performance criteria such as sustainability, stability, resilience or equitability, and in particular where there are no data except local or indigenous knowledge. The World Wildlife Fund Multispecies Animal Production Project (WWF MAPS) was established in 1988 with one of its major objectives being to critically examine the ecological and economic implications of multispecies and single 8 species animal production systems in Zimbabwe (Cumming. 1991). A number of the hypotheses to be tested by the project used sustainability, stability and resilience as performance criteria. As a research fellow working on the WWF MAPS project, I developed a research program designed to examine the sustainability of Masoka, an agroecosystem located in the mid-Zambezi Valley. The following chapters describe the research I conducted as part of the WWF MAPS Project. My original objective was to examine the sustainability of Masoka. The result is a first, and admittedly rather crude, methodology for conducting agroecosystem analyses where the local people were seen as equal, if not major participants in the analysis. I have not been able to address any of the philosophical problems associated with research of this kind. I have also failed to address any of the difficulties of statistically analyzing the data collected in the field as well as the simulated results from the model. Despite the obvious importance of resolving these philosphical and analytical issues I have not had the time. in this study, to do more than recognize their existence. In Chapter 2 I briefly present background data on the Masoka agroecosystem. In Chapter 3, material that was used to develop a general understanding of the agroecosystem and the constraints household face in satisfying needs are presented. Chapter 4 is a description and evaluation of the methods used for field data collection; summaries of these data are included. In Chapter 5, development of the Masoka agroecosystem model is described and 9 the results of preliminary sensitivity analyses of the model are presented. In the final chapter, major features of the research are reviewed and I make a few suggestions about key issues for future research. BIBLIOGRAPHY Bawden, R.J. and BL lson. 1992. The purposes of field-crop ecosystems: Social and economic aspects. In: Pearson, C.J. (Ed.), Ecosystems of the world. Field crop ecosystems. Elsevier, Amsterdam, Holland. 11-35. Bawden, R.J. 1991. Systems thinking and practice in agriculture. Journal of Dairy Science, 74, 2362-2373. Belinski, D.J. 1976. On systems analysis: an essay concerning the limitations of some mathematical methods in the social, political, and biological sciences. The MIT Press, Cambridge, MA, USA. Box, G.E.P. and 6.0. 1130. 1992. Bayesian inference in statistical analysis. John Wiley and Sons, New York, USA. Busch, L and W.B. Lacy. 1983. Science, agriculture, and the politics of research. Westview Press, Boulder. CO, USA. Checkland, P. and J. Scholes. 1990. Soft systems methodology in action. John Wiley & Sons, New York, USA. Checkland, P. 1981. Systems thinking, systems practice. John Vlfiley & Sons, Chichester, UK Conway, GR. 1985. Agroecosystem Analysis. Agricultural Administration, 20. 31 -55. Conway, GR. 1986. Agroecosystem analysis for research and development. Winrock International Institute. Bangkok, Thailand. Conway, GR. 1987. The Properties of Agroecosystems. Agricultural Systems, 24, 95-117. Grimm, V.E. Schmidt and C. \Mssel. 1992. On the application of stability concepts in ecology. Ecological Modelling, 63, 143- 161. 10 11 Hoos, LR. 1974. Criteria for ”good" futures research. Technological Forecasting and Social Change, 113-132. Hoos, LR. 1972. Systems analysis in public policy. University of California Press, Berkeley, CA, USA. MacRae, R.J., S.B. Hill, J. Henning and GR. Mehuys. 1989. Agricultural science and sustainable agriculture: a review of the existing scientific barriers to sustainable food production and potential solutions. Biological Agriculture and Horticulture. 6, 173-219. Mahoney, M.J. 1979. Psychology of the scientist: An evaluative review. Social Studies of Science, 9, 349-375. Mahoney, M.J. 1976. Scientist as subject: the psychological imperative. Ballinger, Cambridge, MA, USA. Marten. G.G. 1988. Productivity, stability, sustainability, equitability and autonomy as properties for agroecosystem analysis. Agricultural Systems, 26, 291 -31 6. Officer, RR. and J.L Dillon. 1968. Probability and statistics in agricultural research and extension: A pro-Bayesian view of modern developments. The Journal of the Australian Institute of Agricultural Science, September, 121-129. Popper, Sir KR. 1968. The logic of scientific discovery. Harper and Row, New York, USA. Stove. DC 1982. Popper and after. Four modern irrationalists. Pergamon Press, Oxford, UK Chapter 2. Background data on the Masoka agroecosystem INTRODUCTION In this chapter I present background data on Kanyurira ward. Kanyurira ward was selected as a study site because: a) it was one of the first CAMPFIRE projects in the Zambezi Valley; b) it has a small and relatively discrete human population; c) studies by the Centre for Applied Social Sciences (CASS) at the University of Zimbabwe (Cutshall, 1989) as well as those by the WWF Multispecies Animal Production Systems project (T aylor, 1993) provide useful background resources; d) the community has been living in the area for, at least, several generations and thus could be expected to exhibit traditional. as well as the more modern, technologies that are spreading through the mid-Zambezi valley area; 9) the absence of cattle considerably simplified analysis of the agroecosystem; f) Masoka is still relatively undeveloped. The community has therefore. a large set of development options that it could follow. 12 13 Study site description Administratively, Zimbabwe is divided into provinces, districts, wards and villages. Kanyurira Ward (Masokaz) is situated in Guruve District, Mashonaland Central Province, Zmbabwe, between 30° 10’ E and 16° 15’ S (Figure 2.1). Nearly 170 km2 (40%), of the 400 kmz, lie above the Zambezi escarpment (Taylor, In preparation). The altitude of the ward ranges from 1120 meters at the top of the escarpment to about 400 meters where the Angwa river leaves the ward in the north east (Figure 2.2). The topography of the ward ranges from steeply dissected slopes on the escarpment to gently sloping alluvial terraces along the major rivers. The geological formations of the area are largely recent to Pleistocene alluvial sands and gravel, Triassic sandstone (upper Karoo group) and micaceous sandstone, the upper portion of the latter alternating with mudstone (Broderick, 1989; Oesterlen, 1989). There is little likelihood of useable shallow aquifers in the ward (Owen. personal communication, 1991; Owen, 1989). In general, the river systems of the ward drain north and south into the Angwa River which rises about 120km to the south. above the escarpment (Figure 2.1). All rivers flow seasonally. The Angwa River is dry for most of the year (May to October), floods after the onset of the rains (November or December) and has peak flows in January and February. Masoka has a long dry season from April to November and a wet season The name Kanyurira is derived from the sub-Chief of the Ward, Headman C. Kanyurira. Masoka Is the name of the royal spirit, (Mhondoro) of the area and Is the name In common usage. It Is not used for the whole ward but is used to refer to the area of present habitation. I shall use Masoka when describing the area of current habitation and Kanyurira to refer to the Ward. 96:25 / Z\ ugm 4 and <= 7 then household = medium If # adults + 0.5 * # children > 7 then household 8 large Selection of these cutoff values was based on discussions with the three field assistants. A simple linear programming model was used to establish the optimal configuration of crops grown on each of the three soil types that would meet the needs of each household size class. The model was expected to roughly identify whether or not households, in each size class, were likely to satisfy needs for cash and food. If it were possible to satisfy needs with the available resources then the LP model should find solutions that could achieve the objective. The solutions identified by the LP model were not expected to be ideal solutions for households classes but only indicators as to whether or not households could satisfy needs in any one year. The objective function of this model was the maximization of gross household income subject to the household satisfying its needs for maize, cotton and sorghum. Households were constrained in the number of hectares they could manage. All data used in the model were derived from the questionnaire survey. Two fifths of the cropping area available to each household was allocated as shapi thougl croppl in the progra years). consid (mean- CrOp p Zlmbat maize. 39 shapi, two fifths as jecha and one fifth as bepe soils. These proportions were thought to represent the probable distribution of the different soils currently cropped in the community. To evaluate the ability of households to satisfy needs in the long run, the optimal crop and soil mixes derived from the linear programming model were used for each season from 1977/78 to 1991/92 (is. 15 years). Rainfall data from Angwa Bridge were used and the season was considered good if rainfall exceeded the 15 year mean less 20% (Le. IF rainfall > (mean-20%) THEN season = "good"). Otherwise, the season was considered bad. Crop production that was surplus or deficit to household needs was converted to Zimbabwe dollars (23) using 1990/91 season producer prices; 23264.40 ton’1 for maize, 251660 ton'1 for cotton and Z$213.35 ton'1 for sorghum. The potential of households to generate income was calculated using crop producer prices for 1978 to 1991, converted to 1991 dollars. The 1978 to 1991 Angwa Bridge rainfall data were used to determine whether a season was good or bad as described above. Yields were set as functions of soil type and season. The land allocations derived from the linear programming models were used for the optimal scenario. For the average scenario no cotton was planted. Of the average areas that different households in each size class claimed they were able to manage, one quarter was planted to sorghum and the rest to maize. Sorghum was always planted on shapi soils. The analyzed results of the survey were presented to the community using graphic techniques: colored cardboard cutouts were made of three sizes of hou: CUIO sorg sque tract BVGI’E resul type the b Svaej fairming a cone appropl ambitioL aCtUally illustrate, Mos; hotter 4o household to represent large, medium and small household classes. A number of cutouts were also made to represent bales of cotton, bags of maize and tins9 of sorghum. Other cutouts of a tractor, a man and woman hoeing, different colored squares and different sizes of cloud were used to represent the two forms of traction discussed, different soil types and whether a rainfall season was good, average or poor. These cards or icons were stuck on a large board to illustrate the results of the survey. For example, yields on each soil type in each rainfall season type could be discussed by pinning maize bag or cotton bale shaped cards on the board to represent the average number of bags or bales reported in the survey. RESULTS AND DISCUSSION The attempt during the workshop to develop a conceptual model of a farming system was inconclusive, in the sense that participants failed to agree on a conceptual model of the agroecosystem and were unable to agree on the appropriate scales of analysis. Much of this uncertainty could be attributed to the ambitiousness of the objective as well as lack of clarity in defining what was actually needed. In another sense however, this result was conclusive in that it illustrated that, as a group. the workshop participants did not have a sufficiently clear understanding of a typical farming system to derive a model or to derive 9 Most households reported sorghum yields In twenty kilogram tins or buckets. appro 1992 v increas model (populi numbe I In the 1 second, Present. village a gross in. CTOplanc SeCtl'On. needs ar Th. 41 appropriate scales of analysis. Estimated numbers of households in the community between 1985 and 1992 were provided by the village informant. These data show the average annual increase in household numbers to be ten percent (Figure 3.1). A linear regression model (population=-16291.4+8.23*year; df=6, r2=.94) and an exponential model (population(t +1) = populationt*exp(°1)) were used to extrapolate household numbers to the year 2010 (Figure 3.1). Questionnaire survey results are presented and discussed in six sections. In the first, household needs for food and cash crops are described. In the second, the areas of land that households claimed to be able to manage are presented. Production and production potential of the different soil types in the village area are presented in the third section. The fourth section deals with the gross income generating potential of households and in the fifth section, areas of cropland that households require to satisfy their needs are presented. In the last section, the implications of these results for satisfying household and societal needs are discussed. Household needs for food and cash crops The requirements for maize (n=72; r2 = 0.915; p<.001), cotton (n=56; r2 = 0.87; p<.001) and sorghum (n=13; r2 = 0.907; p<.001) increased with family size (Table 3.1). These values were derived by linear regression of kilograms of cotton, maize or sorghum (dependent variable) against the total number of adult LT PC ICU. _r\.£ cc L aw F~EsJ7§ Figure 42 700 pop(I)=pop(l-1 )*exp(0-J-)—> Population = 220 eooeeeomuomomoe N O) -I> 0'1 0\ O O O O O O O O O O i i l l i Number of households \ pop=-16291.4+8.23‘yeor Observe r d“ 0 I r T i i 1985 1990 1995 2000 2005 2010 Year Figure 3.1 Estimated numbers of households in Masoka, 1985 to 1992 and predicted numbers of households through 2010 based on a linear and an exponential projection model. Tame Ic' : cnlzr equw Maze vvere ZS35€ EQUNa 43 Table 3.1 Household needs for maize, cotton and sorghum, by household size class. Data are mean i SEM, with n in parentheses. NEED MAIZE COTTON SORGHUM “008900” Size 09) (1‘9) 0‘9) Small 720 1: s (19) 1620 :32 (13) 170 .t 14 (4) Medium 1270 1: 9 (24) 2840 i 36 (21) 230 1 40(3) Large 1610 i. a (29) 3020 -_i_ 34 (22) 470 .t 42 (6) equivalents (two children = one adult) in each household (independent variable). Maize requirements were about 180 kg per adult equivalent, cash requirements were about 23400 per adult equivalent for small and medium households and Z$350 per adult equivalent for large households. Sorghum requirements per adult equivalent were 45, 35 and 50 kg for small, medium and large households. Acreage that households can manage Respondents indicated that they would be able to manage greater areas it they had tractors for ploughing than they could manage by hand, suggesting that labor for land preparation is an important constraint. The mean areas that small households claimed to be able to manage were 1.34 ha by hand and 2 he with a tractor (t=6.26, df=18, p<.001). For medium households these means were 1.8 ha by hand and 3 he with a tractor (t=7.29, df=22, p<.001) and for large households the means were 2.2 he by hand and 3.4 ha with a tractor (t=6.48, df=24, p<.001). Produc 5 one. The Yields ii season 2 for any consider and sorg same se Har‘lzi us YIEIdS (m. yields (m,- et. a], (19' and Com: in Table 5 Fal these that. higher We make defi, 7 (10%) o: Production Seventy two percent of farmers believed the 1990/91 season was a bad one. The very low rainfall measured at Angwa Bridge (480mm) supports this belief. Yields for maize, cotton and sorghum that were obtained by farmers for this season are shown in Table 3.2. Yields were low but did not differ among soil types for any of the three crops investigated. The cotton and sorghum yields were considerably lower than the poor season (1983/84) yields for cotton (950kg ha’1) and sorghum (450kg ha") but considerably higher than the maize yields of the same season (180kg ha") reported by Harizi (1985) for the Muzarabani area. Harizi used Agritex crop yield data to estimate these yields as well as average yields (maize, 720kg ha", cotton, 1200kg ha", and sorghum 900kg ha") and best yields (maize, 1800kg ha", cotton 1600kg ha'1 and sorghum, 1600kg ha"). Brunt et. al. (1986) report yields for the mid-Zambezi Valley area of 700kg ha‘1 for maize and cotton and 600kg ha‘1 for sorghum which are similar to the yields reported in Table 3.2. Farmers who ploughed with a tractor (n=14) had yields no greater than those that used hand cultivation, nor did those who used fertilizer (n=5) achieve higher yields than those who did not. The sample size is, however, too small to make definitive conclusions about the effects of tractors or fertilizer on yields. Only 7 (10%) of the respondents had used credit facilities in the last growing season whilst 13 (18%) intended doing so in the forthcoming growing season. In general the yields that farmers stated they were likely to get on a given In LII hc 45 Table 3.2 Maize, cotton and sorghum yields obtained by respondents for the 1990/91 season on three soil types. Data are mean _-I; SEM, with n in parentheses. MAIZE COTTON SORGHUM SOIL <--------—I(g ha"--—-----—-> SHAPI 670 _-'I_-_ .18 (28) 673 i .19 (12) No data BEPE 540 .t .34 (11) 476 i .29 (4) 490(1) JECHA 482 i .19 (24) 600 i .12 (26) 198 :_I-_ .64 (6) Mean 570 i .12 (63) 600 i .1 (41) 225 i .55 (7) soil with a bad season (T able 3.3) were slightly higher than the yields they actually achieved (Table 3.2). Acceptance of the anticipated yield figures presented in Table 3.3 assumes; a) that farmers have had some experience in growing the crop and b) that they can predict the yields they are likely to obtain under a given set of conditions. To satisfy the first assumption respondents were asked only to respond to creps they had grown. As for their ability to predict yields under given conditions, a fair assumption to make is that farmers, whose families depend on their ability to produce sufficient food and cash, would be able to estimate potential yields and plant sufficient areas to meet their needs. One would in fact expect these estimates to be slightly conservative because mistakes could have serious implications for the well being of the farmer’s household. Comparisons of the yields in Tables 3.2 and 3.3 suggest that farmers tended to overestimate their poor season maize and cotton yields, on bepe soils in particular, and underestimate their sorghum yields. The anticipated yields presented in Table 3.3, however, compare favorably with the yield figures reported by Harizi (1985). The 46 Table 3.3 Anticipated yields for hand cultivated maize, cotton and sorghum on three soils in good and bad rainfall seasons. Data are mean 1 SEM, with n in parentheses. SOIL SHAPI BEPE JECHA AVERAGE CROP _1 MAIZE Good 1650.b 1 20 23306“ + 18 1610“ + 24 1870 +12 (52) (56) (39) (147) Poor 760," + 22 1080,11 22 760.cf1 24 870 1 13 (52) (56) (39) (147) COTTON Good 1780h 1 42 2510ij 1 62 1970h 1 50 2050 1 31 (37) (28) (16) (81) Poor 900 147 1150156 6105162 910132 (33) (27) (16) (81) SORGHUM Good 1260m1148 8101113 470 158 860111 (8) (6) (5) (19) Poor 5401128 160n143 170158 300156 (8) (5) (5) (13) 1. Means followed by the same letter are significantly different at a = 0.05. major differences are that Masoka farmers anticipated much higher poor season maize and good season cotton yields whereas Harizi reported much higher good season sorghum yields. In general, the anticipated yields appear to be adequate estimates for use in the current analysis. Respondents clearly perceived that yields, for all crops, on tractor cultivated fields would be higher than those on hand cultivated fields (Table 3.4; Cotton, n=324, df=323, p<.001; maize, n=585, df=585, p<.001; sorghum, n=72, df=71, p<.001). Since few households had actually used tractors in production of these crops, these yield data are to be treated with some caution. 47 Table 3.4 Perceived yields for tractor cultivated maize, cotton and sorghum on three soils in good and bad rainfall seasons. Data are mean1 1 SEM, with n In parentheses. SOIL SHAPI BEPE JECHA AVERAGE once .1 Season < "9 ha > MAIZE Good 3090_b122 4620cw119 25403125 3430113 (52) (55) (39) (147) Poor 1390“ 123 2320f 121 1330”_+_ 23 1690114 (51) (55) (33) (144) COTTON Good 2930,155 44309,,164 3230132 3452137 (37) (23) (15) (81) Poor 1380111 1 51 1860g167 1330174 1521135 (37) (28) (15) (81) SORGHUM Good 1130,,1124 17601133 920,1 55 12361 60 (3) (5) (5) (19) Poor 10201164 9701133 470k 1 36 3331 74 (7) (5) (4) (16) 1. Means followed by the same letter are significantly different at a = 0.05. Income generating potential The gross annual incomes that households could earn from the production of rainfed, hand cultivated maize, cotton and sorghum were 25430 (CV = 46%) under the average scenario and 231400 (CV = 52%) under the optimal scenario for large households. For small households these values were Z$260 (CV = 46%) under the average and 231015 (CV = 52%) under the optimal scenario. The mean values for each year 1978 to 1991 are shown for large households (Figure 3.2) and for small households (Figure 3.3). These data highlight a) how far from being able to satisfy their needs most households are likely to be; and b) the potential impel needs 2.2) ii prodL Some Cr0pl were c yields QVOWn indicate none 0- their be ”Ct at a DGeds.‘ e”°U9h manage Medium 48 importance of wildlife revenues (Table 2.2) in assisting households to satisfy their needs. Average annual household income from wildlife dividends (23150 pa, Table 2.2) could be equivalent to between 15 and 88% of incomes derived from crop production for small households and between 11 to 35% for large households. Some implications of these earning potentials are discussed later in this chapter. Cropland areas required by households to meet needs The areas required by each household class to satisfy their perceived needs were calculated by dividing the household needs for food and cash crops by the yields achieved for that crop on any of the three soil types where the crop was grown (T able 3.5). Linear programming analyses based on the acreage that households indicated that they could manage and the yield figures from Table 3.3, indicate that none of the household size classes could produce anywhere near enough to meet their basic needs in bad rainfall seasons. To do so, households of all sizes would have to increase their yields or acreage by well over 100% (Table 3.5). These figures are daunting to say the least. Even in good rainfall years it is not at all clear that production from crops is sufficient to meet basic household needs. Using the yield figures of Table 3.3 only small households could produce enough to meet their basic needs given the land areas they claimed they could manage by hand (and the allocation of soil types assumed in this analysis). Medium and large households would need to increase the areas cultivated in good vii if” I V Figure 49 A 5000 8 4500 - 4000‘ 3 . c 3500 E 3000 - f E 2500 - \ U 2000 ' - ‘ g 1500 - ,2. / / 3 1000 A . .. a 3 500 z : 2 fl. '-' 1‘ .: a i a Z 0 .1: = i i = 1 i i 1978 1982 1986 1990 1980 1984 1988 1992 Year +Needs fi-nverage stratequ -+- Optical strategu *Dptteal + midlife +flveraqe + uxldltie Figure 3.2 Household needs (23) and earning potential for large households over the period 1978 to 1992. 50 A 2500 GB '0’ 2000 r / In 0‘ C H 1500 1 ' E f‘ 8 U 1000; / '1’ C “I (U “in" g 500 ‘W I 3 Z 0 c . . . . . . 1978 1982 1986 1990 1980 1984 1988 1992 Year + Needs -8- Average strategu +0ptinal strateqtj -+- 0pt1eal+u11d11ie -)(— fiveraqe+u11dltfe Figure 3.3 Household needs (23) and earning potential for small households over the period 1978 to 1992. Tablt years I sensni areas housel rainfall the ma Size Cla with no. rairlfall l DroducE hOUSth 51 Table 3.5 Cropland areas (ha) required to satisfy household needs‘. Data are total area for household size class with the area per adult equivalent for each size class in parentheses. HOUSEHOLD SMALL MEDIUM LARGE SIZE Manageable (ha) 1.3 (.33) 1.3 (.26) 22 (.24) Good season (ha) 1.3 (.33) 2.3 (.33) 2.9 (.32) Change 0% +30% +30% Poor season (ha) 3.0 (.75) 5.2 (.74) 6.2 (.70) Change +125% +190% «113093 Weighted 2.0 (.5) 3.5 (.5) 3.8 (.4) Total (ha) 1. The total areas required during good and bad seasons are obtained by weighing the areas required In those seasons by the proportion of good (0.6) and bad (0.4) seasons at Angwa Bridge. years by up to 30% (Table 3.5). Solutions from the linear programming model were sensitive to reductions in areas of soil types and the price of maize. Using the areas that households reported as being manageable with tractor tillage, all household classes could produce enough to meet their basic needs in good rainfall seasons, but still not In poor rainfall seasons. If we accept that the yields of Table 3.3 are conservative and add 40% to the maize and sorghum yields and 50% to the cotton yields, then all household size classes are theoretically capable of meeting their needs in good rainfall years, with notable cotton surpluses. If, however, we take the long-term view and use the rainfall data from Angwa Bridge for the past 15 years, and the 1990/91 season producer prices for C grade crops, a different picture emerges. Large and medium households could show gross deficits of between 234000 and 2$6000 over the 15 52 year period whilst small households could show gross surpluses of 2510000. These figures assume everything else being held constant and optimal allocation of land to crops. The surpluses of the small household are attributable to their lower grain needs and therefore, the relatively larger areas they are able to plant tO COttOI'I. Implications When presented to members of the community there was general agreement with the results of this analysis. In seeking an answer to the question, 'Why do households not grow more cotton and use the higher returns to purchase their maize and sorghum needs?" community members responded that they wished to ensure their maize supplies - their staple food crop - and could not rely on their ability to purchase and get transport for maize supplies from Guruve. In the 1992/93 season these farmers were proved correct; the drought resulted in national maize shortages and households were unable to purchase sufficient amounts to meet their needs. A number of patterns are evident from the coarse level analysis. First, large and medium sized households are unlikely to be able to satisfy their basic needs for food and cash over the long term. They are constrained by their need to produce sufficient grain to meet household food needs. Second, households that concentrate their productive activities on cotton production appear more likely to be able to satisfy household food and cash needs than those who attempt to Iev 53 satisfy food needs from local grain production. Third, the ability of households to satisfy needs is partially attributable to the areas of each soil type to which they have access. Fourth, households would need to double the average annual wildlife dividends they received between 1989 and 1992 (is. 23400 pa) to offset the deficits the large and medium sized households could be accruing. One way for households to improve their overall yields in bad seasons is to increase the areas they cultivate so that they always plant sufficient area to satisfy household needs even in a bad season. The recent calls for a tractor are clear indicators that, in the eyes of many households, this is perhaps a more viable option than trying to improve yields by 130% or more. If this is a route that the community chooses, the choice has important implications for the carrying capacity of the arable lands in Masoka and needs to be carefully evaluated. The fine soil filth and larger field sizes that result from tractor cultivation practices could increase the erosion risks on the erodible soils of Masoka. The jump from hand cultivated technologies to tractor technologies is not always advisable (Pingali et al., 1987). The poor fence management performance of the Masoka community is illustrative of the need for caution in introducing relatively sophisticated technologies. l have not dealt with the need to allow fields to go to fallow. Discussions with farmers indicate that fertility declines significantly after three to five years of continuous cropping. With the high risks of crop failure due to low rainfall, low levels of fertilizer seems the rational choice. This means however, that the 54 community needs to plan for twice as much land for each household (assuming five years of continuous cropping and five years of fallow). Based on the hand cultivation figures households claimed to be able to manage and assuming (i) technology remains constant at current levels; (ii) the proportions of large, medium and small households within the community remain constant; (iii) two fifths of the land area is in each of shapi and jecha soil types and one fifth is bepe and (iv) all land within the fenced area is arable, then the number of households supportable within the fenced area is about 20. Given current household population growth rates the community could reach this number of households within six years (Figures 3.1, 3.4). Rapid population growth is a rational response where large families provide labor in a labor constrained environment. Population increases, the need to expand cropped areas to satisfy needs and the tendency of labor constrained households to use existing fields rather than to open new fields (as well as looming land shortages) imply the biophysical resources of Masoka will come under increasing pressure. Wildlife revenues, by buffering households from the variability in needs satisfaction that is a result of the climate, could also undermine the resilience of the agroecosystem by capitalizing the expansion of cropland areas without improving the underlying technology. 4500 , 1500 A 4000-1 1000 ‘0 A 5 3500- 500 r0 5 E 3000 - - --------0 O E, c g 2500‘ "500 E a: ' H L m 3 2000‘ “1000 E U 8 0: 1500‘ -1500 L "U U 1000- --2000 g C _.l '0 —I 500- ;-2500 0 r i 1 ' I l T -3000 1985 1990 1995 2000 2005 20 10 ’ Year + Linear need -8- Linear remain + Exp need -+- Exp remair Figure 3.4 Arable land required and remaining under the population growth scenarios presented in Figure 3.1 and the land requirements presented in Table 3.5. 56 CONCLUSIONS The majority of households in the community are unlikely to be able to produce enough to meet their basic needs from their crop production system, using current technology. Low and unreliable rainfall, a shortage of labor or traction, and inadequate information on optimal cropping patterns and technology appear major constraints to this agroecosystem being able to satisfy household needs. Smaller households and households with greater access to resources have better chances of satisfying their needs. The use of a tractor to improve tillage and the areas households can manage is one possible solution, but one that needs to be evaluated with caution. The calls from the community for a tractor have important implications both for assisting households to stabilize their needs satisfaction and also for increasing the erosion hazard in Masoka. These implications need to be more fully investigated. Average wildlife dividend payments would have to be more than doubled to ensure continuous needs satisfaction, for large and medium households, under the conditions presented in this analysis. Declining national (and global) cotton prices as well as the failure of the GMB to provide adequate food in drought years suggest the common practice of managing a portfolio of productive activities in Masoka is well advised. The fact that households have been living in the area for several generations suggests that the agroecosystem is more resilient than this coarse level analysis would have us believe. We need therefore, to investigate more completely the productive practices of households in different resource access classes. BIBLIOGRAPHY Brunt, MA, V.J. Clarke and C. de Greling. 1986. Tsetse area development and land use planning in the Zambezi Valley, Zimbabwe. Food and Agriculture Organization of the United Nations, Rome. Campbell, 3M. and M.J. Swift. 1989. Links between the environment and basic needs satisfaction. In: Campbell, B.M., R.F. Du toit and CAM. Attwell (Eds), The Save study. Relationships between the environment and basic needs satisfaction in the Save catchment, Zimbabwe. University of Zimbabwe, Harare, Embabwe. 70- 83. Central Statistics Office. 1989. fimbabwe demographic and health survey 1988. Ministry of Finance, Economic Planning and Development, Harare, Zimbabwe. Cumming, D.H.M. 1991. Multispecies systems and rural development in southern Africa: opportunities, constraints and challenges. World Wide Fund for Nature, Multispecies Animal Production Systems Project, Harare, Zimbabwe. Project Paper No. 19. Cumming, D.H.M. and l. Bond. 1991. Animal production in southern Africa: Present practice and opportunities for peasant farmers in arid lands. World Vlfide Fund for Nature, Multispecies Animal Production Systems Project, Harare, Embabwe. Project Paper No. 22. Department of Meteorological Services. 1981. Climate handbook of Zmbabwe. Government of Zimbabwe, Harare, Zimbabwe. Drinkwater, M. 1991. The state and agrarian change in Zimbabwe’s communal areas. St. Martin's Press, New York. Elwell, HA 1992. Cropland management options for the future. Proceedings of the third annual scientific conference of the SADCC-Land & Water Management Research Programme. Harare, Zimbabwe, 57 October, 1992. In press. FSRU. 1985. Annual Report of the Farming Systems Research Unit, Department of Research and Specialist Services. Government Printers, Harare, Zimbabwe. Government of Zmbabwe. 1986. First five-year national development plan 1986-1990. Volume 1. Government of Zimbabwe, Harare, Zimbabwe. 57 58 Grant, PM. 1981. The fertilization of sandy soils in peasant agriculture. Zimbabwe Agricultural Journal, 78, 169-175. Harizi, KEI. 1985. Zimbabwe. Mid-Zambezi Valley agricultural development project. Agricultural production. FAO, Rome. FAO/ADB Working paper No.2. Hopkins, M. and R. VanDerHoeven. 1983. Basic needs In development planning. Gower Publishing Company Limited, Aldershot, England. llifie, J. 1990. Famine in Zimbabwe 1890-1960. Mambo Press, Gweru, Zimbabwe. Jackson, J.C., and P. Collier. 1988. Incomes, poverty and food security in the communal lands of Zimbabwe. RUP Occasional Paper, Department of Rural and Urban Planning, University of Zimbabwe, Harare, Zimbabwe. Jansen, DJ, and A. Rukovo. 1992. Agriculture and the policy environment: Zambia and Zimbabwe. Political dreams and policy nightmares. Study prepared for OECD Development Centre. Jayne, T.S., M. Chisvo, B. Hedden-Dunkhorst, and S. Chigume. 1990. Unravelling Zimbabwe’s food insecurity paradox: Implications for grain marketing reform. Proceedings of the First National Consultative Workshop on Integrating Food, Nutrition and Agricultural Policy. Juliasdale, Zimbabwe, 15-18 July, 1990. Jordan, AM. 1986. Trypanosomiasis control and African rural development. Longman, London, England. Lynam, T.J.P., S.J. Venneulen and BM. Campbell. 1993. Contingent valuation of multipurpose tree resources in the smallholder farming sector, Zimbabwe. Accepted for publication in the Journal for Farming Systems Research-Extension. Mascarenhas, J.P., Shah, S. Joseph, R. Jayakaran, J. Devararam, V. Ramachandran, A. Fernandez, R. Chambers, and J. Pretty. 1991. Participatory rural appraisal. Proceedings of the February 1991 Bangalore PRA Trainers Workshop. RRA Notes, No. 13. International Institute for Environment and Development (IIED), London, England. Mason, E. 1990. The nutrition situation. Current strategies and plans. Consequences of malnutrition. Proceedings of the First National Consultative Workshop on Integrating Food, Nutrition and Agricultural Policy. Juliasdale, Zimbabwe, 15-18 July, 1990. Murindagomo, F. 1988. Preliminary investigations into wildlife utilization and land use in Angwa, Mid-Zambezi Valley, Zimbabwe. University of Zimbabwe, Harare, an Ndb sysié lnSi Rn. fie Pres Reh, Syste Zhnb Ruku SAIM Secu RUkc agnc Pfimd 900k CONN Shur Farnl Stag Fhsn “syn. 30b: 135991 Dow Zimbabwe. Ndlovu, L 1990. The role of ruminants in promoting food security in farming systems in the SADCC region. Proceedings of the conference on Food Security in Southern Africa. Harare, Zimbabwe, 16-18 October, 1989. Pingali, P.L, Y. Bigot and HP. Binswanger. 1987. Agricultural mechanization and the evolution of farming systems in sub-Saharan Africa. John Hopkins University Press, Baltimore, U.S.A. Reh, l. 1986. Farming systems research program and progress. Mimeo, Farming Systems Research Unit, Department of Research and Specialist Services, Ministry of Agriculture, Hmbabwe Government, Harare, Zimbabwe. Rukuni, M., T.S. Jayne, J. Tagwireyi, and L Rugube. 1990. Alleviating hunger in Zimbabwe: Towards a national food security strategy. World Bank, Washington DC, USA. Issues paper for the World Bank Agriculture Sector study for Zimbabwe. Rukuni, M. and J.B. Wyekoff, (Eds.). 1991. Market reforms, research policies and SADCC food security. University of Zimbabwe/Michigan State University Food Security Research in Southern Africa Programme. Rukovo, A.T. Takavarasha, R. Thiele and M. Wlebelt. 1991. The profile of agricultural protection in Zmbabwe. Kiel Working paper No. 457. Scoones, l., and K. Vlfilson. 1988. Socio-economic dimensions of livestock production in the communal lands of Zimbabwe. Households, lineage groups and ecological dynamics: Issues for livestock research and development in Zimbabwe’s communal lands. Harare. Shumba, E. 1989. Maize technology research in Mangwende, a high potential communal area environment in Zimbabwe. Part 1: Developing a research agenda. Farming Systems Newsletter, 34, 12-34. Stack, J.L, and C.J. Chopak. 1990. Household income patterns in Zimbabwe’s communal areas: Empirical evidence from five survey areas. Proceedings of the First National Consultative Workshop on Integrating Food, Nutrition and Agricultural Policy. Juliasdale, Zimbabwe, 15-18 July, 1990. Stilz, D., and U. Weyl. 1986. A model of an ecology-oriented agricultural development programme and its transformation into a coordinated attack on rural poverty. University of Osnabruck, Osnabruck, Germany. 60 Streeten, P. 1986. Basic needs: Some unsettled questions. The challenge of employment and basic needs in Africa. Oxford University Press, Nairobi, Kenya. Swift, M.J., P.G.H. Frost, B.M. Campbell, J.C. Hatton, and KB. Wilson. 1989. Nitrogen cycling in farming systems derived from savanna: Perspectives and challenges. In: M. Clarholm and L Bergstrom (Eds.), Ecology of arable land. Kluwer Academic Publishers, Netherlands. Whitlow, R. and BM. Campbell. 1989. Factors influencing erosion in Zimbabwe: a statistical analysis. Journal of Environmental Management, 29, 17-29. World Bank. 1991. Zimbabwe agriculture sector memorandum. World Bank, Washington DC, USA. No. 9429-Zim. To SL SYSIel IS th Bawd part c mOtiva c0W3: exam p CUTTSn minim); plans C failed S Chapter 4. Needs analysis, system Identification and problem formulation In the Masoka agroecosystem INTRODUCTION To successfully improve a system’s performance we need to first identify the system and second define what constitutes an improvement. What is not as clear is whose perspective we are to adopt in making these decisions (Bawden, 1992; Bawden and lson, 1992). A large number of actors (individuals and organizations), with an equally large set of motivations, have varying degrees of influence on a part or, the whole of, the Masoka agroecosystem. These perspectives and motivations carry with them value systems that the actors use to weight the inputs, components or outputs of the agroecosystem: the National Government, for example, might emphasize cotton production activities to maximize its foreign currency earnings. Whereas individuals might place greater weight on activities that minimize the risk of their household going without food that year. Development plans or projects that use values other than those of the target community have failed so often the credibility of this “top-down" development model has been 61 ur fin re: 50! efft inc bee ana: DBTS knov expe Work Chang diner, ahern, persp, Zimba Camp 1939; E 1985. : ThOnRD 62 undermined (Bawden and lson, 1992; Chambers 1983; Redclift, 1987). These failures often arise from inadequate understanding of household needs, household resources, household objectives and the constraints that households face and suggest that, for the purposes of this analysis, we stand a greater chance of being effective if we adapt the perspective of Masoka households. The value of incorporating local knowledge and perspectives into development planning has been receiving increasing attention (Brokensha et al., 1980; Warren, 1991). Whilst using local knowledge may be a necessary condition for successful analysis of the Masoka agro-ecosystem, assuming that adopting a household perspective would be a sufficient condition for a successful analysis is naive. Local knowledge is often restricted to what farmers can see and what is within their experience (Richards, 1980) and may be unevenly distributed in a community (IDS Workshop, 1983). If either the agroecosystem or its environment are likely to change faster than household adaptive capabilities, or in ways that are radically different from local experience or perception, then we are compelled to incorporate alternative perspectives into our analysis. The obvious source for these alternative perspectives is the large body of empirical knowledge that has been built up in Zimbabwe on many aspects of household resource use (Campbell et al., 1989; Campbell and Du Toit, 1988), crop production practices (FSRU, 1985; Reh et al., 1989; Shumba, 1984a, 1985, 1986a,b), crop yields (Akwabi-Ameyaw, 1990; FSRU, 1985; Shumba, 1984b), soils (Anderson, 1986; Grant, 1981; Thompson, 1965; Thompson and Purves, 1978), household labor use (Reynolds, 1991), household 63 economics (FSRU, 1985; Shumba, 1985; Stanning, 1987a,b; Chopak, 1991; Mudimu et al., 1988) and soil erosion (Elwell, 1980, 1985; Elwell and Stocking, 1984, 1988). Much of this information Is, unfortunately, inadequate to develop the predictive capabilities we require In our analysis. Few of these authors provide the kind of data (statistics of population distributions) needed for stochastic simulation. With the exception of the detailed study on child labor in the Zambezi valley by Reynolds (1991) and Elwell’s erosion model (Elwell, 1980), little of this information is directly relevant to the mid-Zambezi Valley. A further problem with using indigenous knowledge as the basis for an extensive analysis is the difficulty in evaluating the degree to which the information presented by local informants is an unbiased representation of the populations of interest. With few exceptions (Farrington and Martin, 1987; Gil, 1991; Swift, 1979) this issue has received little attention in the literature on indigenous knowledge or RRA. The literature on potential biases in human responses to questions is large and has been examined from several perspectives: Aler and Settle (1985) discuss sources of bias from a survey perspective and Tversky and Kahneman (1974, 1981) discuss the subject from a psychological perspective. Adelman, in describing the problems of biases in expert knowledge suggests, “At the very least, knowledge engineers and evaluators should be aware of their existence and look out for instances where elicited judgements may reflect a bias," (Adleman, 1992) which is hardly a rigorous strategy for establishing the accuracy and precision of a set of knowledgel Of particular relevance to this study is the 64 extensive treatment of bias and measurement validity by Mitchell and Carson (1989). There are two issues of concern here. The first is knowing whether one has measured that which was set out to be measured. Mitchell and Carson, (1989. p190) call this the validity of a measure and identify three indicators of a valid measure - content, criterion related, and construct. Content validity concerns establishing whether the method used to measure the parameter of interest constrains or biases the elicited response. A measurement is valid if it does not bias or constrain the response. Criterion validity concerns the degree to which other measures of the population of interest can be used as criteria to establish the validity of the elicited measure. The ideal measure here would be actual measures of the population under consideration. For example, measures of actual maize yields would be an ideal criterion to establish the validity of the yields farmers say they obtain. Direct measures are unlikely to be available (othenivise we would not be using an indirect method) but other measures that are “unequivocally closer to the theoretical construct than the measure whose validity is being assessed" (Mitchell and Carson, 1989, p192) are necessary to assess this type of validity. Construct validity concerns the degree to which other measures of the same population either converge towards the same result or are consistent with what we would predict from theory. A second issue of concern is the accuracy and precision of the measurements one makes. In evaluating sources of bias in respondent responses, 65 Mitchell and Carson (1989) note four major sources of systematic bias. The first is the use of scenarios that contain strong incentives for respondents to misrepresent their true beliefs either because they think they can influence some result of the study or because they wish to comply with the interviewer’s beliefs to please either the interviewer or the study sponsor. The second source of bias arises when respondents, unsure of their responses, rely on the information relayed to them through study questions to formulate their responses. The third and fourth of these sources of systematic bias involves the incorrect specification of the scenario about which the interviewer seeks information. This may occur either because the scenario is incorrectly specified from a theoretical perspective or the scenario is correctly specified but communicated in such a way that the respondent does not perceive it the way the researcher intended. To my knowledge no predictive analyses of community level production with differentiated households has been undertaken in Zimbabwe. There can be no doubt that analyses at the scale of the household are difficult: Households are structurally heterogenous (Cutshall, 1989); they have different labor, land and capital resource bases (Coudere and Marijse, 1991; Reynolds, 1991; Shumba, 1984b, 1985); they exhibit different production objectives (Stanning, 1987) and they engage in a diverse set of productive activities (Jackson and Collier, 1991), not all of which are agriculturally based (Helmsing, 1991). The importance of inter- household sharing, hire and employment (Shumba, 1985; Stanning, 1987; Znyama et al., 1987) as well as the major role of women and children in household 66 production (Adams, 1991; Mehretu and Mutambirwa, 1992; Reynolds, 1991) exacerbate the difficulty of analyzing agroecosystem performance from a household perspective. From this daunting complexity we must construct a model that we believe incorporates the essential elements (properties) of the agroecosystem and Is capable of accurately simulating agroecosystem functions and of predicting future states and outputs of the agroecosystem. We quite obviously have to simplify the system considerably. In this simplification process we are faced with two major methodological problems. First, how do we identify major actors, their needs and key agroecosystem components? Second, how do we collect the data we require to develop and test our predictive model? These problems are particularly difficult firstly because, Masoka household decision makers are, for the most part, illiterate and inumerate and therefore keep no records of production inputs, areas or outputs. Second, time and money constraints allow us only a single cropping season for data collection. There is little in the scientific literature to guide our search for appropriate methods. Most household analyses rely on household questionnaire surveys (Campbell et al., 1989; Jackson and Collier, 1991; MLARR, 1989; Stanning, 1987), which are questionably useful for the rapid and flexible learning recording required when no a priori model exists to guide data collection. The long term immersion approach of anthropologists often lacks the quantitative information required to predict changes and to compare alternative development or production strategies 67 over long time periods (Hasler, 1992; Reynolds, 1991). The more informal open- ended interview techniques commonly used by social anthropologists (Derman, 1990, 1992) yield qualitative information on many aspects of a production system but lack temporal breadth and are not likely to yield the general functions required for predictive simulation analyses. Recent developments in techniques classed as rapid rural appraisal (RRA) or participatory rural appraisal (PRA) show great promise in terms of their flexibility and the range of both quantitative as well as qualitative data they are able to record (Khon Kaen University, 1987; Mascarenhas et al., 1991). The depth and breadth of data that are required for predictive simulation analyses require, however, far more than the single short visit with a multi-disciplinary team of investigators, that is standard practice in RRA. Where RRA techniques have been used in Zimbabwe (Carter of al., 1993), there has been little attempt on the part of the researchers to test the validity of the methods used or the accuracy and precision of the information collected. The nature of the analysis required to develop a predictive model of the Masoka agroecosystem. from a household perspective, demands some hybrid of the in-depth, qualitative analysis of anthropologists, with the rapid and multi- disciplinary approach of RRA as well as the more formal methods of agricultural research. For the most part appropriate data collection methods have not been developed. In this chapter, the field data collection methods developed for the analysis of the Masoka agroecosystem are described. The results of applying these to: agi ecc to e is 0 asa anal nun reqm apprc 0f vill' I Ih80( Selec Wore i”form 68 methods to the Masoka agroecosystem are described and discussed with regard to: 3) identifying major actors and client groups influencing agroecosystem behavior; b) establishing the needs and objectives client groups expect the agroecosystem to satisfy; 0) Identifying the major components of the agro- ecosystem and their relationships and; d) Identifying the criteria client groups use to evaluate the performance of the agroecosystem. The focus of the data collection is on developing the ability to predict the state or behavior of the agroecosystem as a function of controls that Masoka households can impose on their resources. METHODS Having a multi-disciplinary team carry out the Masoka agro-ecosystem analysis for the period required by this analysis was not possible. Expecting large numbers of Masoka residents to give as much of their time as this analysis required was also not feasible. I decided therefore, that the most appropriate approach to knowledge collection would be to have the community elect a group of villagers to represent the community perspective. This group provided most of the community level knowledge presented in this chapter. Selection of community representatives At an open meeting held on February 2, 1992, Masoka community members were informed of the objectives of my research. Community members were informed that I would need to work intensively with a small group of people from 69 the community. No reward or bench fees for the members of the group with whom I would work were mentioned. The community members were explicitly asked if they wanted me to proceed with the analysis. Upon receiving the community’s approval, people at the meeting were asked to identify classes of household in Masoka and to proportionately weight these classes according to the number of households in each class. Community members at the meeting were asked to elect representatives from each of these classes to form the team of village or community representatives (VRs) who would work with me. The community was specifically requested to include women in this elected group but this request was refused on the grounds that the men in the community did not wish their wives or daughters to interact with unknown men. Six of the men present at the meeting were elected. All except one were young and either recently married or unmarried. At the first meeting with the VHS, n early March 1992, I was informed that the eldest VR had withdrawn. The village informant who assisted me during earlier survey work (described In Chapter 3) was elected by the remaining VHS to replace the missing man. The ward councillor was approached to have women included in the VR team. He subsequently sent three women who had agreed to join the VR team. One of these women was the councillor’s wife, one a widow and one the wife of a locally respected farmer. The women VRs were aged between 38 and 58, had between three and eight children each, and none of them had been to school. The male VRs were younger, between 22 and 26 years old and had between nine and 11 years of schooling each. Two of the men had one child each, the others 70 had none. Each of the men spoke at least some English but none of the women did. All except two of the VHS were born in Masoka and had lived there for most of their lives. In discussing the work that we were to do together, I broadly explained my objectives and the methods I would attempt to use. I volunteered to pay the VRs and to provide food for the group while we worked. A daily wage and a meal schedule were agreed upon. In this, as in all such discussions, my approach was to develop a sense of equal participation in a team effort. Each member of the team was encouraged to express his or her views and decisions were reached by consensus. In this visit I had brought with me a field assistant and translator, Mr. Limited Mukusanya, a permanent staff member of the WWF Multispecies Project. For this first visit, the VHS were split into two groups, one comprising the men and the other the women. In all other visits the VRs worked as mixed male and female groups. Altogether eight visits were made to Masoka over the period March to December 1992. Each visit lasted from seven to 14 days. After March 1992 translation during interviews and all VR activities was carried out by two of the VRs; Mr. Gift Chisunga and Mr. George Kanoderuka. General approach to recording local knowledge Field trips were generally structured as follows: I arrived in the early afternoon and used the remainder of the day to send messages to the VRs that 71 I wanted to meet the following morning. One of the VRs would find someone from the community to prepare food for the team for the duration of my visit. A morning and midday .meal were prepared for all VRs by the hired cook. By tradition men and women ate separately. I shared all meals with the VRs. The) issue of food and meals requires some clarification. Zimbabwe, at this time, was in the middle of the worst drought in memory. Many of the people in Masoka were living on drought- relief food supplied by the government. Expecting hungry people to work the long and intense hours that the VRs worked was neither realistic nor ethical. I therefore undertook to feed the VR team whilst we were working together. The sharing of meals is also an important social process in the rural areas of Zimbabwe, and facilitated an open and trusting relationship among the VRs and me. At the first meeting the VHS were asked to describe events and issues that had occurred or were being debated in the community since my previous visit. Thereafter, the objectives of the visit were outlined and work sChedules were proposed, discussed and arranged. All work times and arrangements were arranged to suit VR needs. New concepts and techniques were illustrated with simple, and locally relevant examples. VRs would then carry out a task using the new technique or concept. Once a task was complete the entire group would collect to discuss and seek consensus on the outcome of the task. When the basic data collection procedures and methods had been used for several field trips, VRs were asked to prepare projects that they would like to see 72 developed in Masoka. The VRs were asked to do this in my absence with the objective of evaluating whether the techniques that had been developed were likely to be useful outside of the framework of my own research. Identification of key Individuals and groups within the community Perhaps the most important data collection method developed during this analysis was what is called a spidergram: a graph with a central node and arms or branches leading off this node. Each branch may end in a sub-node which may also have branches leading from it to further sub-sub-nodes and so on (Figure 4.1). Three points are allocated for each branch emanating from a node. These points are allocated to each branch of a node to weight the relative importance (RIW) of that branch. The central node is designated as the main concept or question about which we are seeking information. For example, Figure 4.1 shows the spidergram developed to answer the question 'Which individuals or groups in Masoka have most influence on agricultural production?" Each branch emanating from the core node represents an individual or a group. The numeric values are the weights attached to these individuals or groups in terms of their degree of influence on agricultural production. RIWs were normalized by dividing each branch score by the total number of points allocated to that node (i.e., three times the number of branches). VRs drew these spidergrams in the sand and used stones or sticks to indicate the RIW of each branch. VRs were encouraged to use appropriate 73 symbols to represent each node so that even those VRs who could not read could understand the representation. When a spidergram was complete the group would gather around it and the VHS who had drawn the diagram would carefully explain what each node and branch represented and then have to defend, to the entire group, the branches they had included or left out and the RiWs they had allocated to each branch. After consensus was reached that all relevant branches were present and that the Rle were correct, the spidergram was copied into a field notebook. Exceptions to the consensus rule were the actors and needs spidergrams developed separately by male and female VRs. Consensus was reached within these groups but not among the entire VR team. Identification of household needs VRs were asked to develop spidergrams to express the needs that households required to live a basic, but adequate life. During a subsequent field trip the identification of these needs was developed in greater detail using needs calendars or matrices. VRs were asked to draw columns in the soil to represent each month of the year. Then, for each of their major needs they made a row and filled in the amount of that need required for each month. Separate matrices were developed for the needs of an average man, an average woman and an average child. in a thirteenth column the VFls were asked to allocate a number of points between one and five to indicate the degree to which they could not live without that need (five points) or if it was more of a luxury than a necessity (one point). A 74 row was added to the matrix and once again the VHS were asked to allocate points between one and five, but this time to indicate the degree to which they experienced needs shortages (five being none and zero being extreme) in each month of the year. When a group of VRs had completed a matrix the whole team gathered around it and the authors were asked to explain the matrix and defend their inclusion or exclusion of items, the amounts they claimed were required in each month and the relative weights they gave to each need and to each month. Once consensus was reached by the VHS on the diagram and its content, it was copied into a field notebook. Inputs to, outputs from and measures of performance of, the Masoka agroecosystem Identification of the major inputs and outputs from the agroecosystem built upon the spidergrams developed for needs identification. Once the needs nodes were identified. they became central nodes for more detailed input/output spidergrams. The question posed in the central node then became, for example, "What is needed in order to produce a good yield of maize?" Branches from this node indicated the major inputs to, and losses from. the production of maize. These branches were weighted and sub-branches were added where necessary. All spidergrams were discussed and defended before final acceptance. A third major tool for recording local knowledge was the possibility diagram. The objective of these diagrams was to enable VHS or other village informants to 75 express the probabilities that they believed described the relationships under consideration. A possibility diagram for maize yields is described for illustrative purposes. Specific maize production scenarios were identified from the production spidergrams and from discussion with VRs. Soil type and rainfall, for example, might be identified as key determinants of maize yields. For each soil type and rainfall season combination a possibility diagram was developed. A single stick or ‘, was placed at the base stone. representing a maize yield of one 91 kg bag acre' of a column drawn in the sand. VRs were then asked to indicate the possibility of obtaining at least that yield, under the specified conditions, by placing between zero and 10 sticks or stones in the column above the symbol representing the single bag maize yield. A score of 10 indicated that households could definitely obtain at least that yield whilst a score of zero indicated they would definitely not be able to achieve that yield. Scores between 10 and zero indicated the probability that a household could achieve at least the yield specified at the base of the column, under the given circumstances. A column was then added with stones representing the number of bags yielded incremented by one. The scoring process was repeated as before. Columns were added (and possible yields steadily increased) until a score of zero was given to a particular yield to indicate that it could definitely not be achieved. Possibility diagrams were developed for maize, cotton and sorghum yields on each of the soil types on which they were cultivated (bepe, jecha and river lands) and for a good, an average and a poor season. Possibility diagrams were 76 also developed for maize, cotton and sorghum in each season and grown on virgin jecha (i.e., never cultivated) and jecha that had been continuously cultivated for five years. The need for this differentiation was realized following discussions with VHS and other community members as yields on the jecha soils decline with continuous cultivation. Possibility scores were normalized to lie in the range of zero to one. These diagrams represent the cumulative probability function of achieving at least a given yield. The probability of achieving a given yield can be calculated by subtracting the normalized probability of achieving a particular yield from the probability of achieving the next highest yield. The mean of a yield distribution (E[x]) was calculated as: Eixl ‘3 2 PDQ"? The variance of a yield distribution (V[x]) was calculated as: Vin = 2: (mix?) - 5x12 Where: P(xi) = the probability of achieving yield level xi xi = the ith yield level. Where weighted means and variances were used they were calculated according to the following: Weighted an = 23 Wt, (E[x]) Weighted M = )3 (Wt, Mm + wr, (Eixlizll - am: Where: Wti = the probability of a particular event occurring, such as the probability of a particular season or soil type. Possibility scores were also used to analyze the rainfall of Masoka. VRs were asked to identify the types of rainfall that occurred in Masoka. For each of these rainfall types possibility scores were developed to indicate a) the probability of a rainfall event of each type occurring in any week of the rainy season; b) the probability of a given number of these rainfall events occurring in any two week period of the rainy season; and c) the probability of a rainfall type lasting for a given period of time. VRs were asked to map the soils of Masoka, map the placement of households in Masoka, as well as the areas that were unsuitable for crop production due to soil, slope or cultural reasons. These maps were drawn in the sand or on canvas, using locally available materials for differentiation. The maps were described and defended by their authors and then copied onto paper. A paper map was also drawn by two of the VHS to indicate the historical distribution and movement of households within the Ward. To elicit information on cropping decisions, yields and farmer perceptions of soil condition, a combination of field mapping and semi-structured interviews 78 was used. When used with VRs as respondents these were combined with what were called field interviews. A hypothetical example is described for illustrative purposes. The respondent was asked to point to where he or she lived and was asked to draw a circle on the grbund to represent their household. Respondents were asked to stand within that circle and were orientated towards the Angwa river, a feature all could point to from any place within Masoka. Then for each year since their arrival in Masoka, respondents were asked to draw the fields they had opened or used. Respondents were asked to indicate the size (acres) and soil type(s) of these fields as well as the crops they had grown, the yields they had achieved and any problems they had encountered. When this exercise was carried out with VRs as respondents the VR was asked to stand in particular fields and to pretend that they were the field. They were then interviewed by me and other VRs as if they were the field. in some instances, a second VR was asked to pretend that he or she was the farmer so that the field interview went through the "farmer" to the "field". Questions related to weed populations, yields, fertility and general soil condition were asked. To obtain a clearer picture of the politically sensitive issue of land allocation to new settlers or sons of local residents VFls were asked to adopt the roles of the various actors involved in land allocation and to present a short play of the proceedings. Once a play was completed the actors were switched around so that different perspectives and emphases could be observed. Plays were observed for a would be new settler and a long-term resident seeking land. The dialogue in 79 these plays was translated for my benefit and discussed after each performance, which lasted only a few minutes. Spidergrams were used to identify the attributes that VRs considered useful in determining a household's ability to continuously satisfy needs. The central node asked the question 'What do you look for to tell you that a household is able to satisfy its needs?" Major factors, and sub-factors that could be used to measure these major factors were identified, added to the branched nodes and weighted in the usual way. Well-being ranking (based on wealth ranking methods, Grandin, 1988) of all households in Masoka was used to identify classes of household based on their ability to satisfy needs. VRs were asked to define a household and to define classes of well-being in Masoka. The name of each household head was then written on a card and the cards shuffled together. The VRs, working as a group, selected each card and discussed the attributes of the household before placing it into one of the well-being classes. Decisions were reached by consensus. Thereafter, VRs ranked the cards in each well-being class according to the household’s ability to satisfy needs. The name of each household was read out to the group of VRs. VRs discussed whether or not that household was better or less able to satisfy needs than each card that had already been discussed. The card under discussion would be moved through the stack, from bottom to top. until the household was no longer considered to be better able to satisfy household needs than the household represented by the next card on the stack. VRs worked in 80 smaller groups to identify the number of children and adults in each household as well as the number of acres of river-land that each household possessed and whether or not the household was an employer or an employee of other households, had a family member in other forms of employment or received remittances. Using an open-ended interview (format, VRs were asked to recall the expenses they had incurred and the prices they had received in producing each crop (maize, cotton and sorghum) in the previous season (1990/91). VFts also developed labor requirement calendars (matrices) for each crop that indicated the number of labor days (per acre) required for each task, in each month of the growing season. Possibility diagrams were developed to indicate the possibility of one adult completing each activity on one acre in at least a given number of days. These data (expenses, prices and labor budgets) were used to develop cash and labor budgets for each crop. Methodological validity and data accuracy The issue of the validity of the measures made in this study as well as their accuracy and precision is of obvious importance and is perhaps the most difficult issue to logically and statistically resolve. As discussed previously there are two broad areas of concern. The first of these is the degree to which the methods employed bias the results obtained and the second is the accuracy and precision of the measures made. 81 All interactions with VHS and other community informants required some measure of translation; translation of the question from English to ChiSezuru (Shona) and then the translation of the responses from Shona to English. To avoid gathering biased information then we need to be confident that: a) the translation of English symbolism into Shona symbolism was accurate; b) how the respondent perceived the translated symbols was consistent with how the analyst perceived these symbols; c) the respondent was accurately portraying their knowledge of the phenomena under consideration; d) the translation of Shona symbolism into English symbolism was accurate; and e) how the analyst perceived these translated symbols was consistent with how the respondent perceived these symbols. This process alone, had enormous potential for introducing error. The approach to dealing with these problems in this study had been to use, as far as possible, graphic representations of questions, situations and responses. The assumption being made was that graphic representations provide a common set of symbols that minimize the translations required as well as forcing both analyst and respondent to simplify all communications. Wherever possible specific, real world symbols were used to illustrate important phenomena; an old maize cob might be used to represent maize in a spidergram, for example. Content validity (Mitchell and Carson, 1989) was assumed to be minimized by the following circumstances. if there was a distortion in the translation process we might expect answers to either be inconsistent with expectations or for divergent answers to be expressed. All data were discussed almost as soon as 82 they were generated. There was therefore, opportunity for immediate discussion if the data did not conform to expectations or appeared inconsistent. Consensus was sought on all data developed by the VRs and in most cases groups of VRs independently developed data representations and then had to defend their representations to the entire VR team. This worked much like the scientific peer review system. In most situations, the VHS themselves defined the specific information they presented. The analyst posed general questions or concepts but allowed VHS to define the specifics of each issue. in some instances, the use of different elicitation methods enabled us to identify whether or not there was divergence in the recorded information. The second test of the validity of the methods used was the extent to which they generated results that were consistent with other measures of the same population. For most data collected in this study, the requirement that these alternative criteria be "unequivocally closer to the theoretical construct than the measure whose validity is being tested“ (Mitchell and Carson, 1989) was impossible to satisfy. There just are no reliable data for most of the factors of interest in Masoka or in similar situations. The best that can be done is to use published results that are most likely to have populations, of the factors of interest, that are similar to those under scrutiny in Masoka. The final test of the validity of the methods used was the degree to which they produce results that were either consistent with what might be predicted by theory or were correlated to measures from the same theoretical population. For 83 example, the Department of Meteorological Services (DMS) literature on rainfall in Zimbabwe (DMS, 1975) provides a distribution of storm duration that we can use to test the information provided by VRs. We might also expect different correlations between yield and rainfall on different soil types - expectations that we can test. The issue of data accuracy and precision was very difficult to completely resolve. To some extent the tests of validity discussed above will enable us to, at least qualitatively, establish the accuracy of the results we observed. if the methods are shown to be valid, then if the same populations are measured using different methods and each method yields the same result then we may feel confident in the accuracy of these results. in addition, if the sample statistics we observe are close to those published in the literature then we may feel confident in the accuracy and precision of these results. Problems obviously arise when different methods produce different results or the results we obtain are inconsistent with the literature or there are no comparable results. The approach adopted in these cases was firstly, to examine the validity of the measures using the three types of validity tests discussed by Mitchell and Carson (1989). We recognize criterion validity as being the strongest test and content validity as being the weakest. if the measure is shown to be valid and we find no reason to believe that the responses are Systematically biased, then we will accept the information. Quite clearly, the degree of confidence we hold in particular information will be related to the strengths of the tests that information has withstood. The theoretical model (logical hypothesis) which underpins our data 34 collection is as follows: Masoka residents have a long history of depending on the biophysical environment to provide most of their food and other material needs. Household well-being has been, and is, dependent on their knowledge of the structure and patterns of their environment as well as their ability to manipulate their environment and the resources at their disposal to satisfy their needs. Two sets of predictions are deducible from this model. The first of these is that Masoka residents (as represented by the V83) can accurately describe the structure and patterns of key environmental inputs to household needs satisfaction. The second of these is that VRs can correctly describe the results they are likely to achieve using commonly used combinations of controllable inputs with different resources and under commonly experienced environmental situations. The grounds for accepting or rejecting a prediction in this study are difficult, because in most cases, there are insufficient data or their form is such that statistical analyses are impractical. The decision to accept or reject a prediction will therefore, be based on careful evaluation of the validity of the measure and comparison of the measure to comparable results published in the literature. These decisions will be made as transparent as possible so readers are able to reach their own conclusions. 85 RESULTS Key actors and household classes in response to the question 'Which individuals or groups have greatest influence on agriculture in Masoka?" both male and female VFI groups identified the spirit medium, headman and ward councillor as key actors (Figures 4.1 and 4.2). These data also illustrate the power split between traditional leaders (Headman and Kraalheads) and the more recent elected leaders (ward councillor and Village development committee (VIDCO) chairmen). The importance of spirit mediums in community decision making within Zambezi Valley communities has been noted by Lan (1985) and by more recent anthropological research (Hasler, 1992; Derman, 1992). The split of power between traditional leadership and the post 1980 elected leadership appears common through much of Zimbabwe (Drinkwater, 1991). The male and female VRs differ most in their perceptions of the importance of VIDCO Chairmen, with the women perceiving them to be the third most important group whilst the men considered them least important. ‘ Both male and female VRs used size as a primary factor for classifying households within the community (Figures 4.3 and 4.4). Male VRs extended their classification to include wealth as a second factor to differentiate household classes (Figure 4.3). Male VRs perceived most households to be in the large and wealthy category. Using the household size classification rules of Chapter 3 to allocate female VR classes to large medium or small, their analysis indicates that 86 Spirit medium .28 Chairmen\-05\. ' Major Actors 11 .17\C ll - ounce or / ZANU PF .17 Branch Chairman Households Figure 4.1. Major actors influencing agricultural production in Masoka and their relative importance weights. Female VB perspective. Spirit Medium Headman Households \03 ' .17/ ZANU PF \ v / VIDCO Branch Chairman \ 07 17/ Chairmen ° \9 Major /' NJ o School .———-—-' Chairman / \ Councellor .07 .07 .07 Health / / \ Workers Wildlife Committee Kraal Chairman Heads Figure 4.2. Major actors influencing agricultural production in Masoka and their relative importance weights. Male VFl perspective. 87 Wealth Poor y‘.67 .33’ Wealthy W l \ ea thy .83 .22) / b/ .17 .33 / \ Poor Poor ND = No data Figure 4.3. Household classification and the relative importance weights of each household class. Male VB perspective. MFG-4* M3F0-17 - \20 l / MFS e Household classes M2F4-14 MF7-12 * M-> number of male adults F-> number of female adults 0-4->number of children Figure 4.4. Household classification and the relative importance weights of each class. Female VR perspective. 88 medium sized households dominate (Figure 4.4). For all household size classes male VBs indicated wealthy households dominate. This latter result is in contradiction to the result presented by the community members present at the meeting when the VBs were elected. At this meeting households were classed as poor (BlW=.8) and not so poor (BIW=.2) indicating a predominance of poorer households. This contradiction suggests that the male VB spidergram did not adequately representthe question of interest - 'What types of household occur in Masoka and which are the most common?" it appears that the male VBs were perhaps answering the question 'What types of household occur in Masoka and which have the greatest influence?" The male VB spidergram does however, provide useful information. Household needs In answering the question 'What do households need in order to live a satisfactory life in Masoka?" female and male VB groups developed the Spidergrams shown in Figures 4.5 through 4.8. Both female and male VBs identified food and cash as most important household needs as has been reported elsewhere in Zimbabwe (Shumba, 1988). The female VBs also indicated the importance of trees and water, land and a clinic (Figure 4.5). From their perspective, major inputs to satisfying household food needs were maize. cash, Chickens, pumpkins and finger millet. Major inputs to household cash needs satisfaction were maize, cotton, chickens and brewing beer. 89 in the male VB perspective. household needs and their BIVV were indicated for each household class (Figures 4.6 to 4.8). For most household classes male VBs identified maize, chickens, sorghum, pumpkins and vegetables as major inputs to household food needs satisfaction. Male VBs also noted the greater relative importance of food to poor households compared to wealthy households. They perceived cotton and maize as major inputs to household cash needs satisfaction for wealthy households whilst cotton, craft and sales of fish or honey were identified as major inputs to household cash needs satisfaction for poorer households (Figures 4.6 to 4.8). These data indicate a decrease in the number of production activities with decreasing household size and to a lesser extent with wealth. Similarly, Jackson and Collier (1991) found an increase in the per capita incomes of households with increasing number of productive activities. Also notable from Figures 4.6 to 4.8 is the shift from relying almost totally on crop production, for wealthy households, to increased reliance on indigenous resources by the poor households of each size class. This latter observation is in accord with the coping strategies of households in deficit conditions described by Zi nyama et al., (1987) in which households make increasing use of indigenous (Wild) resources as food scarcity intensifies. These authors also suggest that it is the women who tend to guide the household through these difficult times. The sensitivity of women, as expressed by the female VHS, to household needs for indigenous resources (Figure 4.5) supports this proposition. Women are also more directly concerned with the daily collection of water and fuelwood and are 90 828828 m> 295.... .>>_m :9: 5? atom: 9.05 2 £35 BEE 9: 3 =95 mm 92$ £299; mocmtoag $26.9 :9: 53, «com: 20:33: 2m V 9:9“. .353”. EapoE / =25 5.8 cases: 2228 a. N z \ /mN. I ‘2 5N. 5N. we 6.8-5.8.... 52m \8. 412/5.535 , // v\:.\ 2.x 085 5330588 852 «new... @3521.“ 1:. 38 Extra. «”6 t. «poo: : @aiillit. ...l.ll..l .o 20:09.0: .8. IlllcoamZomcoo sumo \h d/mM. / 50¢“; \.NN\ bogkm c.3822 V/NN/ 9.....iil :80 8. r .mfiI'OCOv—nucc EoE>o_aEm\ \8.\\1.‘\.\@/.fl/ . t. FN/NEZ at 5. =80\ // 3. B. .25: 0 Bl I $33oao> \8 .580 B .35... mo 8 mo \ o==2 _ a G @5390 />E=On_ 5:298 / tawn— ficE 91 mm 228:0; moi m be 23.530 830 I or. :r t mcidEad ISTIV \ B n . ‘ Eacmaow\ p &\t 8. \ \ >25: u__>>\ &\ 9. 628828 ¢> 9m 2 .>>_m :9: 5? name: $0.: 2 235 SEE 9: mm :95 555 2:995 wocmtoaé 9:29 :9... 5:5 .mp8: 20:832.. .3. 9:9“. «mos Eageom/ .80 \ m: QE: 29.0.50 2. 9. \ .3 d . or. /t. IQSwScucaew cums. lth /flo¢omo.lmo.nflomo> . / 8. 8/mco__oE-._2m>> B. 8. _ mzogaoao i=2 cozoo ON. NEE K 9.\o \ sKa/ _ 5:00 520 T . 5. .1 @d/ 91600 wo_fi.omo> \ 8.\t \ /1/ or. . 5. /. mco=mE¢2m>>\ \8 9. m 9.020 /. Manuaow 830 9:: u: 20 628898 m> was. .BE .65 55, mp8: 805 o. 985 BEE 9: 8 =95 8 20:83: 6:68. a .2 55.5 £5995 8:2..an 8:39 .65 £3, .88: 20582.. .5. 9:9“. $388; E3590 goofing mco__oE.._2a>>/ m’o. w/mw. 3\ m: QE: 2.15 t 89.020 /2. .b:o=oE.._2£$ .2 .. lwflfi gel... 9 a \ 8. EEQMMono\t. “a / mm. /mc_xaE:d . . one: 8. B. «2.22 92 29.820: \2. 8 \éc/ . 6% :90 B. 8 \ d/ . \ 9. / . 8 .68: was \ 8 R 8:00 2 3 / 5.80 not _ / \ t. / 28 >30 “no” cums. >801 628828 m> was. .>>_m .65 53> mp8: 805 9 98:. SEE 05 mm :95 mm 8.88:0; =me .2 .925 9:995 85:08:: 8:39 :8. 2:; .88: 29.8.8... .m.v 9:9“. menses; 89.0.8 2 29.35 \ mcflaEsd/ B. / or .7. . £1.49 2095030 \ ON. 9. mo_§omm>\ \ / one: 8. «Sc 9265 205830: % \el :95 ollé/ \BF. 8. 9» /6/ «N\ a 3 / a / B \ A 8:8 N. / :88 _.P. .. _ / 94 therefore, more likely to be aware of the importance of these resources than are men. The greater reliance of poorer households than wealthy households, on craft production for the generation of cash incomes (Figures 4.6 to 4.8) suggests that these households could be important initiators of informal seetor production enterprises. Boserup’s (1965) model of agricultural development implies these groups play an important role in rural development. What is surprising from these analyses, is that only the female VRs indicated the importance of land to household needs satisfaction. (Figure 4.5). This is particularly noteworthy in that women do not have the same rights to land as do men. Their rights of use are normally derived from their husbands or fathers. and sometimes even their children (Reynolds, 1991). Perhaps because of this relative landlessness, women have a greater sensitivity to land scarcity than do'men. System identification Spidergrams were developed to illustrate the inputs to. and losses from, each component of household needs. Only the spidergrams for maize, cotton and sorghum are developed and discussed in this chapter. The importance of these crops as inputs to both household food and cash needs (Figures 4.5 to 4.8 and Tables 4.16 to 4.18) suggests their selection is appropriate. 95 Malze production In answering the question 'What is needed to produce a good yield of maize?" both the male and female VRs identified rainfall, soil, crop variety and labor as key inputs (Figures 4.9 and 4.10). Female VRs noted the importance of land but male VRs did not. Major losses from the maize production sub-system were identified as being caused by wild animal damage (particularly from elephant. buffalo and wild pigs) by both male and female groups of VRs. Insect pests were not identified as major loss causing agents although the male VRs identified termites as causing loss. Female VRs developed a more detailed breakdown of labor inputs to each task than did male VRs. Gender differentiation of tasks is evident as is the importance of the labor of children, findings that are in accord with those of Reynolds (1991) for the Tonga people of the Zambezi Valley. Female VRs also indicated a greater appreciation of natural resources than did male VRs through their inclusion of trees and grass as inputs to maize production; These inputs are important in restoring field fertility. Male and female VR perceptions of the relative importance of different soil types for maize production differed considerably. The female VFi group stressed the importance of katondo and bepe with shapi and jecha being similarly weighted. Very few households in Masoka have access to katondo soils. These soils are red soils with a high clay content. Discussion with the VFls indicated that they were identifying these soils with the red soils of the major maize growing 96 .mp8; mums. boom do EoEo>oEom oz. .80: $09.me .23 can g2 8.559 m. 8:; so 95328 m> p.95“. .m.v 9:9”. moot. 5965 896 .28; mo. .o5___tmu_ / 8. mo. \ 03:5 mo mo. 82 . 2. / an . m . . mw. /» Euro \ \3 \ax $2. 5320 \ e 9 ma 25> SE; _mcoaushtmmué x1 > m&m\ 8 8 95K 5%. Ag a.) must \ \. . 8.1 9:235 =£Emm .833 came boom do EoEo>mEom on. .80: $09.8 .9? can ..2 .8569 w_ “an; :o 9:83on m> was. .94. 9:9“. 383 80: roam—BF 9.0mm 98 areas of Zimbabwe. Thus the women were being true to the question in identifying these soils as being best for achieving good maize yields. Of the soils that are generally available to households, both male and female VR groups agreed on the relative importance of bepe over jecha. There appears to have been some confusion as to the identification of shapi soils. Only one household in the community was observed to grow crops on shapi, suggesting that few households had any experience with this soil type. it is also important to recognize that at this early stage of the analysis the VR groups were not distinguishing between the riverine sandy soils (river jecha) and the alluvial terrace sandy soils (jecha). These distinctions only became apparent later in the analysis when dealing with the yield potential of different soils. Female VRs identified maize variety R201 as being the best which agrees with the varietal recommendations made by Whingwiri and Harahwa (1985) for NR Ill and with those of Shumba (1984a). The selection by male VRs, of variety R215, a long season variety. in preference to R201 and R200, both short season varieties is more difficult to understand. Whingwiri and Harahwa (1985) suggest that R215 is a poor performer in a low yielding environment but can respond significantly to an improvement in the environment. The interaction of soil and variety was not investigated in this study but could be a fruitful subject for future research. Cotton production In contrast to the maize production sub-system, female and male VR group perceptions of the relative importance of inputs to cotton production were quite different. In answering the question 'What is needed to achieve good cotton yields?", both groups indicated the importance of soil and chemicals (Figures 4.11 and 4.12). Female VRs considered rainfall and crop variety to be less important than labor, whereas male VRs perceived rainfall to be the most important input and both rainfall and crop variety were perceived as being more important than labor. The Rle ascribed to women’s labor by both male and female VRs (Figures 4.11 and 4.12) suggests why female VRs perceived labor to be of such great importance. Women are the dominant source of labor for all of the major activities except land clearing. Cotton production adds appreciably to the demands made on household labor and particularly that of women and children. Male and female VRs differed also in what they perceived to be the major losses from the cotton production sub-system. Male VRs identified insect pests as being less important than monkeys and baboons whilst the female VRs identified red bollworm as the major loss and did not note losses due to wild animals. Both male and female VR groups identified jecha as the most important soil type for cotton production but differed greatly on the relative importance of the other soil types. 100 .mu_o_> cozoo boom .0 EoEm>mEom 9: .80: Eugen 8...? new .8. 09309 m_ .23 :0 93898 «S 295“. .P 3. 9:9“. 5.4.. a»... 8 x 832 \. Ezoz,__oo .x 528:3 Euros B. \ 9.an 58; a o. . 583 0:0 o cozoo 9.520 23 E82509 \m. com 8 /mc_>maw 8. m_mo_Emno\ \ P «P. /bo:m> was. n. 2.8% 8mm 8 Macon Barman 8:23. 101 .333 8:8 boom so 59:99:08 on. .80: «88:8 .853 new :8 8.509 m_ .23 :0 026828 m> 28.2 d; 959“. 29:25 53882 .25... SEQEm ofisc 28:50.98 cops; .4 /ON. \agw_o 5.0 /8 :. 822328 B. 8. mean 2.8””. 532.3 580$ éf beam 8.1 \ c238\8..1 - _ £9.85. Ken 2 a 8.88 _ Q, . mm. insists? 9. /3 x 8\ x {2‘2 Siol:.t.\ c2823.. . .om.Emc 8/ C0300 no.0 4.8 ofl/ 9.280 .80.. \ mp. 82 53;. bo:m>\ x3928; ma/cmcaom :Emm w_mo_EonO\ 8. @Mgofl. _amnw oo/om 1 02 Sorghum production In answering the question 'What is needed to achieve good sorghum yields?" both male and female VFi groups identified rainfall, labor and seed variety as major inputs (Figures 4.13 and 4.14). The most notable differences in the female and male VR perceptions were the inclusion of land by the female VHS and the much higher importance of soil type in the male VR analysis than in the female VR analysis. As with the maize and cotton production spidergrams, the female VRs developed more detailed labor components to their spidergrams than did the male VRs. The importance of male labor in so many of the activities of sorghum production (Figure 4.13), the identification of a male householder as the local expert on sorghum production, as well as the greater detail in the variety component of the male VFi spidergram (Figure 4.14) suggested that sorghum is less of a women’s crop than is generally believed. The relative importance of local seed sources compared to commercially available hybrids and the relatively high weights given to bepe soils by both male and female VRs are consistent with Reynolds (1991) findings elsewhere in the Zambezi valley. Summary of Inputs to and losses from crop production sub-systems For maize, cotton and sorghum rainfall, soil type, seed variety and labor were identified as the major determinants of good yields by both male and female 103 .mp6; 53:98 noon be .5592ch on. .89. £858 “on; new c2 8.53. m_ “an; :0 93828 «S macho“. .96 2:9". magneto 3%: t. 8.8 88 .535 cozoaooa Eacmoom /. /® Tamilaom m. w... .323”. . / . 52m. / 02.93. 9.3... 104 .333 £398 boom so 298538 we .89. 38:8 an; new c2 8562 m_ “as; :0 920898 m> 292 .36 059.... Magnum 8. K madcap no.0 \ mp. 8. /...E_m oozes—boa Eacmoow om. _Esam §mfinm\¢. 8. W goals. .. I 8836\8. V8. om/. 28.. 80m 105 VR groups. Wild animals were held responsible for a large proportion of the perceived losses from maize, cotton and sorghum production whilst insect and mite pests were important loss causing agents only in cotton. Female VRs consistently expressed their perception of the importance of land and, in the case of maize, their perception of the importance of trees and grass for fertility restoration. Women appeared to be the dominant source of labor in cotton and perhaps in maize although the trend was not as clear as for cotton. Men seemed to have a greater role in sorghum production than is generally recognized. Children contributed importantly to the labor requirements of the household in most major crop production activities. Commercially available seed varieties (hybrids) are most important for maize and cotton but traditional (open pollinated) varieties are most important in sorghum production. More detailed analyses of the rainfall, soil and labor inputs to crop production were conducted and the results are presented below. Despite the stated importance of crop variety in achievement of household needs satisfaction, this factor was not investigated further. Rainfall analysis The VRs identified four distinct rainfall types: The first, named Mhepo (called thunder in subsequent discussions), occurs throughout the season and is characterized by wind and thunder with only a light rainfall over short periods of time. The second rainfall type, Mhunurukwa (called heavy rain in subsequent 106 discussions) is characterized by heavy rainfall with storms lasting longer than several hours. This rainfall type occurs mostly in January and February. The third rainfall type Mvuramabwe (called hail in subsequent discussions), is characterized by hail, occurs infrequeme and usually lasts for less than 30 minutes. The fourth rainfall type, Pfunambuya (called light rain in subsequent discussions), is a light rain that continues for several hours and sometimes even a whole day. The VR data indicated a high probability of the rainfall season beginning in the first two weeks of November and certainly beginning in the last two weeks of November (Table 4.1). The Department of Meteorological Services (DMS) indicated November 22"d as the mean date of the first rainy pentad for the eastern mid- Zambezi valley and the end of the rainy season occurring in the first two weeks of March (DMS, 1981 ). The VR data suggested the last certain rains occur in the first two weeks of March (Table 4.1) but recognized a high probability of light rain occurring through until the first two weeks in April. The DMS also indicated the mean number of raindays of 1mm or more as being between 50 and 60 days for that region of the Zambezi Valley (DMS. 1981). if we take the sum of events with possibility scores greater than zero as an upper bound on the number of rain events and those with a possibility score of 0.5 or greater as a lower bound (Table 4.1) then the VHS indicated a range of between 41 and 76 rain events a season. We should expect these figures to be slightly higher than the DMS values due to the possibility of multiple rain events in one rain-day. The DMS identified the major hail season as occurring in October, November and then declining sharply in 107 Table 4.1 Subjective probabilities indicating VR perceptions of the probability of the number of rainfall events, of each rainfall type, occurring in any two week period of the rainy season. Month Tact Nov Bee Jan Feb Mar Apr Week Rain type 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2 3/4 1/2 # Events 1 .2 .a o o o o o o o o o o 2 o o o o o o o o o o o o 3 o o o o o o o o o o o o 4 o o o o o o o o o o o o ”a" 5 o o o o o o o o o o o o 5 o o o o o o o o o o o o 1 o .s 1 1 1 1 1 1 1 .5 o o 2 o .2 1 1 1 1 1 1 .7 .2 o o 3 o o .5 .5 .5 .5 1 .6 .3 .1 o o 4 o o .3 .2 .2 .3 .4 .5 .1 o o 0 H3131? 5 o o .1 o o o .1 .2 o o o o 6 o o o o o o o o o o o o 1 o o o o 1 1 .4 .4 .3 .2 o o 2 o o o o 1 1 .2 .2 .1 .1 o o 3 o o o o .a .5 o .1 o o o 0 mm“ 4 o o o o .5 .3 o o o o o o 5 o o o o .2 .1 o o o o o o s o o o o o o o o o o o o 1 o o o o o o o 1 1 1 .5 .s 2 o o o o o o o 1 1 .7 .3 .2 3 o o o o o o o .s .9 .2 .2 0 ”gm rain 4 o o o o o o o .4 .4 o .1 o 5 o o o o o o o .1 .1 o o o 5 o o o o o o o o o o o o December (DMS, 1981). These expectations were consistent with the VR perceptions of a low possibility of hail in late October and a high probability in early November (Table 4.1, Figure 4.15). The VRs did not recognize any possibility of hail occurring in December. VRs indicated a high probability of thunder activity in late December through to the beginning of March (Table 4.1, Figure 4.15) which Probabllttg 000053.099 Ox l .3"1 u * I I ‘ I as: . E \ 0. 1 a ’5 a. ‘, ‘. t 0" Iifo :.:.':—_:'II::IZIIII 06/09 26/ 10 15/ 12 03/02 25/03 14/05 Date (dag/month) “.O- 253‘. '.:|-. i: 1 I 1 1 0.1- .. l —B— Hall ............. Heavg W Thunder “’0' quht Figure 4.15. Probability of a rainfall event of each of four rainfall types occurring in each week of the rainy season. 109 is consistent with the DMS reports of high thunder activity in November through to the end of February and then dropping rapidly in March (DMS, 1981). The DMS reported the frequencies of rainfall storms lasting for particular periods of time (DMS, 1975). These data may be used to compare the rainfall event duration probabilities developed by the VRs for each rainfall type (Figure 4.16). The DMS indicated a 45% probability of a storm lasting for one hour or less, 26% probability of a storm lasting for two hours, a 14% probability of lasting three hours, an 8% probability of lasting four hours, a 5% probability of lasting five hours and a 1% probability of lasting 6 hours. The VRs indicated a 30% probability of a heavy rainfall event lasting one hour or less, a 30% probability of it lasting two hours, a 10% probability of it lasting three hours a 20% probability of it lasting four hours, a 10% probability of it lasting five hours and no probability of it lasting as long as six hours (Figure 4.16). VRs indicated that light rain events would, in general last, longer than the heavy storms (Figure 4.16). Masoka residents, VHS and a local expert, identified by the VHS, classified previous rainfall seasons as being either good, average or poor. The VRs, working as a group, and the local expert classified the previous ten seasons whilst in the questionnaire survey presented in Chapter 3 all respondents were asked to classify three seasons ending 1990/91. The results of these classifications, together with the rainfall measured at Angwa Bridge are shown in Table 4.2. Drobab111tg 110 0. 0. 0. " 0. 0. .. 0. 1- x 0 fl I T I I .“‘* I 0 1 2 3 4 5 6 7 8 Event duratlon (hrs) + Hall "'8'" Thunder W Heavq W quht Figure 4.16. Probability of a rainfall event lasting at least a certain time for each of the four types of rainfall. 111 Table 4.2 Classification of ten rainfall seasons ending 1991 [92 as good (G), average (A) or poor (P) by village representatives (VRs), a local informant and questionnaire survey (1988/89 to 1990/91 only). Rainfall season Measured Source of season classifications1 rainfall (mm yr") , , at Angwa Bridge VRs Local Questionnaire expert survey % Respondents G A P 1 982/83 561 P P 1 983/84 529 G A 1 984/ 85 1 303 G P 1985/86 ' 1057 A e 1 986/87 558 G G 1 987/88 579 A P 1938/89 91 1 A P 76 2 22 1 989/90 803 P nd 41 25 34 1 990/91 478 P nd 1 4 1 4 72 1 991 / 92 431 P Dd 1.Seasonclasses:G=good,A=averageandP=poor.nd=nodata. These data show an inconsistent pattern of agreement between rainfall amount at Angwa Bridge and the season classification. Some of this inconsistency is attributable to the spatial variability of rainfall in the Zambezi Valley area; quite possibly the rainfall at Masoka differed significantly from that at Angwa Bridge in any one season. We could expect some overlap in seasons classed as good with average or poor with average but we would not expect a consistent classification system to confuse good and poor. We note one such conflict (for the 1984/85 112 season) between the VR classification and that of the local expert (Table 4.2). The VR classification, based on a consensus of eight people, was consistent with the rainfall data. if we accept that the VR classification is correct then the proportions of good, average and poor seasons are .3, .3 and .4. Consistency in results of the rainfall analysis that were obtained using different methods (Table 4.1 and Figure 4.15) gave us confidence that the results were not biased by the method of data collection. The good agreement between the published records of rainfall patterns and the observations of the VHS gave us further confidence in the validity of the methods used as well as in the accuracy and precision of the VR observations. The positively skewed distributions of rainfall event durations provided by the DMS (DMS, 1975) provide the theoretical expectation with which to compare VR data. As with previous tests the VR data were consistent with the theoretical distribution providing further support to the validity of the methods as well as the accuracy and reliability of the results. In each of these cases the results do not support the prediction that VRs are unable to provide accurate descriptions of the structure or patterns of key elements in their environment. Labor requirements for each crop enterprise Cotton has the highest labor demand per hectare (160 person days) of the three crops under consideration and sorghum the lowest with 84 person days (Tables 4.3 to 4.5). The most labor demanding activities were weeding, harvesting 113 Table 4.3 Labor requirements (days ha"), for each activity of maize production in each month of the growing season. Labor days ha'1 morith'1 Activity NOV Dec Jan Feb Mar Total Digging holes 10 10 Planting 5 5 Weeding 12 25 37 12 86 Harvesting 17 17 Collecting maize 7 7 TOTAL 27 25 37 12 24 125 and land preparation in all three crops with pest management (scouting and spraying) being demanding on cotton. Peak labor demands occurred in November, December and January which are similar to the results reported by Shumba (1985. 1988) for Mangwende CA. The Farm Management Research Section of the Economics and Markets Branch of the Ministry of Lands, Agriculture and Rural Resettlement provided labor input data input for maize, cotton and sorghum collected by questionnaire survey in several CAs of Zimbabwe (MLARR, 1989, 1990). Not all of the categories of activity in the MLARR report were comparable to those used by the VHS for Masoka but the general trends may be observed. The hourly labor requirements presented in the MLARR report have been divided by six to develop daily figures for comparison with the Masoka data. Reynolds (1991) reported working days of between four and six hours and Shumba (1986a) also uses a six hour working day in his analyses. 1 14 Table 4.4 Labor requirements (days ha") for each activity of cotton production for each month of the growing season. Labor days ha'1 month" Activity Dec Jan Feb Mar Apr Total Digging holes 12 12 Planting 5 5 Weeding 20 20 20 1o 70 Scouting 2 5 14 Spraying chemicals 5 10 7 2 24 Picking 35 35 TOTAL 44 35 32 14 35 160 Table 4.5 Labor requirements (days ha") for each activity of sorghum production, in each month of the growing season. Labor days ha" month" Activity Oct Nov Dec Jan Feb Mar Total Clearing 5 5 Digging holes 12 12 Planting 5 5 Weeding 20 17 10 47 Harvest 10 10 Carrying 5 5 TOTAL 5 1 7 20 17 1O 1 5 84 For maize production the MLARR reported labor inputs as being between 63 and 127 labor days per hectare for their sample sites in Natural Regions Ill and IV. Due to the non-availability of animal draft power in Masoka and the limited ability of households to hire a tractor for ploughing operations we might expect 1 15 Masoka labor inputs to be higher than those in the areas reported by the MLARR. For the most part the VR data indicated a greater proportion of the labor allocated to weeding than is reported by the MLARR and less on harvesting than reported by the MLARR. Ascertaining from the MLARR report what activities are included in their harvesting category, which accounts for between 50 and 70% of the labor inputs reported for maize was not possible. Shumba (1986a) reported half the harvest labor, at approximately the same yield level as do the MLARR. These great differences suggest the MLARR harvest data included activities not used by the VR nor were in the data reported by Shumba. The MLARR data on weeding labor ranged from 15 to 27 days per hectare in 1988/89. Possibility diagrams were developed by the VR depicting the probability of one adult completing a task in a given number of days. For weeding one hectare of maize the expected value (standard deviation in parentheses) for task completion was 16.3 (3.4) days which is close to the 17.5 days implied by Shumba (1986b). Shumba indicated that farmers weed their maize fields twice in a season. The VR data indicate six weeding rounds per season for maize which must be an ideal situation. if we assume two weeding rounds in Masoka then the total labor requirements for maize are 72 days ha", much the same as Shumba’s 73 days. Mangwende farmers used slightly more labor for planting (7.4 days ha") as well as for harvesting (30 days ha") (Shumba, 1986) than do farmers in Masoka. The lack of draft power in Masoka increased the labor demand by 15% for maize. 116 The MLARR data presented for cotton labor requirements ranged from 107 labor days for NR III to 172 labor days for NR N (MLARR, 1989). Once again these overall totals were similar to those presented by the Masoka VRs (Table 4.4). The weeding labor figures for one of the MLARR sites (Nyjena, NR N), 76 labor days, were similar to those expressed by the VR for Masoka (7O labor days, Table 4.4). The mean number of days for all sites was however, 36 days ha'1 which suggested about two weeding rounds a season. The harvest labor depicted by MLARR for cotton, 42 to 91 labor days, (MLARR, 1989) was higher than the 35 labor days expressed for Masoka by the VRs. The VRs did not develop a possibility scoring diagram for cotton picking but indicated that three picking rounds would be required; the first requiring about 14 labor days, the second 10 and the last 6. These data were consistent with the 35 days depicted in Table 4.4. It seems likely that VRs were representing an ideal situation in identifying four weeding rounds on cotton. As with maize it seemed likely that only two would be done. If this was the case then the weeding labor would be 40 days ha'1 and make the total labor requirement for cotton, 130 labor days ha". The MLARR labor input data for sorghum production indicate about 65 labor days were required per hectare (MLARR, 1989) which were lower than the 84 days indicated for Masoka by the VHS (Table 4.5). As with cotton and maize production we might expect Masoka labor requirements to be higher because of hand tillage requirements and the need to guard crops from'wild animals. Guarding labor data were not included in the labor budgets and the labor required for digging holes 117 was 12 labor days per hectare. With these days removed the Masoka labor data were much closer to those from the MLARR. As with cotton and maize however, the distribution of labor among tasks differed greatly: weeding labor in the MLARR sites (between 21 and 33 labor days ha") was less than in Masoka and that for harvesting (29 labor days) much higher. As with the previous two crops we might assume that households weed their sorghum fields only twice (Chiduza et al., 1992; Reynolds, 1991) which would make the Masoka data (37 labor days ha") comparable to the MLARR data. When the data from the labor matrices (T ables 4.3 to 4.5) were compared to those derived from possibility diagrams there was general agreement giving us confidence that the methods were not distorting the results. The results for maize presented by the VRs were very close to those described for Mangwende by Shumba (1986a) as well as being close to the more general patterns of labor input described by Reynolds (1991). The VRs appeared to be indicating optimal labor inputs in order to answer the question posed. We might expect therefore, the actual labor inputs to be more in line with those suggested by Shumba - particularly with regard to weeding. The MLARR data exhibit great variability in labor input data between sites and except for the harvest data, the Masoka labor input values were well within the ranges reported by the MLARR. Given the close agreement between VR harvest data for Masoka and that of Shumba (1986a) for MangWende we may be reasonably confident that the labor input data reported by the VHS were valid and accurate. These results do not support the prediction 118 that VRs were incapable of providing accurate descriptions of the structure or patterns of key elements in their environment. Soils and soil fertility VRs were asked to define and then map the soils of Masoka and then to select sites, representative of those soils, in which to dig soil pits and collect soil samples for soil analysis. A pedologist from the University of Zimbabwe, Dr. W. Verboem, was brought to Masoka to assist in mapping and classifying the local soils. He identified essentially the same soil classes as had the VRs. VRs were asked to describe the indicators they used to determine how fertile or infertile a particular field was and how they would know when a fallowed field's fertility had been restored. Loss of fertility was indicated by yield decreases and the increasing cover of annual weed species commonly called Warimani and Kambumbu1o (Table 4.6). They said this occurred after three to seven years on jecha soils and was consistent with the findings of Reynolds (1991). Fertility was indicated with the presence of tree and grass species, particularly Acacia tortillas (Table 4.6). They believed full fertility was restored on jecha soils after 15 years of being left fallow at which time A. tortillas reached a height of nine to 10 meters and a maximum biomass (Figure 4.17) although few farmers leave their fields fallow for that long. Grant (1981), in discussing the fertility of granitic sands in the CA5 of Zimbabwe, suggested these soils could be cropped for three years and then left 10 Botanical names for these plants were not available at the time of writing. 119 Table 4.6 Relative importance weights (5 highest to 1 lowed) of tree and herb or grass speciesafterajechafieldlsleftfallow. Local species name Years after cultivation ceases incl1 1 2 3 4 5 5 7 5+ Trees Mzungu (Acacia F 1 2 3 4 5 5 5 5 tortillus) Marowe F 1 2 Mhangara (Dalbergi'a 1 2 2 melanoxylon) Mutohwe (Azanza F 1 2 2 2 2 garckaena) Mupakasa 1 2 2 2 2 2 2 (Lonchocarpus capassa) Herbs and grasses Warimani l 5 6050 (T richodesma 1 2 3 3 3 3 3 3 zeylancia) Mhunhuniwa 1 2 3 Tsikinya F Kambumbu l 4 3 2 Tsine (Heteropogon F 1 2 3 3 contortus) Bande (Urochloa F 1 2 3 3 tribhopus) 1. ind = Fertility indicator. F = Indicator of fertile soil. l = Indicator of infertile soil. fallow for 15 or more years until legumes of the genus Croteli'aria indicated that the soil was again fertile. Attempts to establish normal fallowing practices in Masoka proved fruitless. No consistent pattern was discemable either from discussions with the VHS or from the people within the community that the VHS indicated were particularly knowledgeable about Masoka. Many farmers stated that they 1mportance uelqht Qelatlve .—. 120 10=Mox biomass 1=M|n blomdss _;;§;§g§g§g§g 0 O=Does noioccur 7 iii???3253555353333??? 5 .:i§i§§§§§§§§é§§§3" 4 2§§§§§§§§§§§§3§3§* 3 Mzunqa 2 Hpakasa Marowe 1 .t..........«....... Mpanqara I\I\I\/\/\/\/\/\/\/\l\I\I\/\I\/\/\/\I\I\ 0 III/Ill/l/l/l/IIIII/ HUtOhue 11213 4 5 6 7 8 9 N31112131415 Years alter jecha field abandonment Figure 4.17. Increase in total tree biomass with time after field abandonment as a proportion of the maximum achievable in 15 years for major tree species growing on jecha. 121 would leave a field fallow after five years of continuous cultivation but when asked about the actual history of their fields few actually did what they said they should do. The period that fields were left fallow was equally difficult to discern clearly. One local expert, Mr. Dishon, who had been cropping in Masoka since 1964, indicated that he cropped his jecha fields for four consecutive years and expected slight yield reductions each year. In the fifth year he would plough the field with a tractor and the yields would increase to the same as those expected from a virgin field. Thereafter, they would steadily decline until the ninth year when he would leave the field fallow. He left the field fallow for five years and then restarted the cycle. Most Masoka residents that were questioned on this subject held approximately this view of the fallow cycle however, their actions were more diverse than this pattern indicates. This nebulous fallow practice was also reported by Reynolds (1991) for areas further west in the Zambezi Valley. The VRs indicated that households seldom, if ever, left bepe or river jecha fields fallow. They knew of no one in the community who did so. Bepe soils were considered to be endlessly fertile and the periodic flooding of the Angwa river was considered to restore the fertility of the river jecha fields. None of the households in the community had had sufficient experience with cultivation of either katondo or shapi soils to know when fertility began to decline or how long to leave them fallow for fertility restoration. The understanding of soil fertility by VRs was consistent with the trends reported in the literature but the practices of households differed considerably from 122 what was considered good practice. Some farmers, for example, followed more or less regular fallowing cycles whilst in at least one case, a farmer had continuously cultivated the same fields with sorghum for at least 20 years and perhaps as much as 30 years. There was general agreement on the importance of trees, particularly A. tortillas, in restoring soil fertility and several tree and grass species were used as indicators of fertile soils. A. albida is considered a sacred tree in Masoka and residents may not cut one down. No attempts were observed or described by the VRs to enhance the nutrient restoration potential of these trees by silvicultural or management practices. The clear descriptive understanding VRs had of their soils, as well as the good relationship between the knowledge of VRs on soil fertility and that in the published literature, do not support the prediction that VRs were unable to provide accurate descriptions of the structure or patterns of key elements in their environment. Crop yields The yield estimates derived from asking household respondents to map household fields and asking them for their estimates of acreages and total yields generated the yields shown in Table 4.7. Analysis of variance indicated yields were different among seasons for maize (p<.05, F=4.468, n=52), for cotton (p< .05, F=5.375, n=28) and for the two classes of season (average and poor) for which there were sufficient sorghum yield data to make comparisons (p<.01, F=8.439, n=26). These data were similar to the average yields reported for the 123 Table 4.7 Mean (standard deviation in parentheses) yield estimates (kg ha“ 1) for maize, cottonandsorghum,derivedfromhouseholdfieldmapsandhousehold respondent inforn'iation. Season Maize Cotton Sorghum < kg l‘la'1 > Good 1334 (755) 755 (35) ND n=2 n=2 Average 804 (515) 915 (458) 631 (445) n=14 n =1 1 n: 8 Poor 511 (429) 459 (209) 235 (252) n-36 n=15 n=18 Weighted Mean1 e45 (575) 700 (337) -- n=52 n=28 1. Weighted mean. Season averages weighted by the probability of that season occurring. Good = .3. average = .3 and poor = .4. ND = No data. Table 4.8 Maize. cotton and sorghum yields reported for the mid-Zambezi Valley. Brunt‘ Harlziz Hawkins’ Jessat‘ Avg. Dry East Best Avg. 1983] 84 Maize 700 200 300- 1800 720 180 846 1340 1000 (793) Cotton 700 nd 500- 1600 1200 950 981 1600 1800 (512) Sorghum 600 450 400 1600 900 450 540 910 (364) 1. Brunt et al. (1986). Average, drought year and eastern Zambezi valley yields 1984/85 season. nd = No data. 2. Harizi (1985). Best, average and 1983/84 (drought) season yields for Muzarabani. 3. Hawkins Associates (1982). Mean (standard deviation in parentheses) yields for multiple sites in the eastern midlambezi Valley area. 1981/82 season (low rainfall). 4. Jassat and Chakaodza (1986). Average yields, based on Agritex crop forecasts, for Rushinga District (NR N and ill), some of which lies in the eastern Zambezi Valley. 124 mid-Zambezi valley by Brunt et al. (1986) and to the maize yields reported by HaWkins Associates (1982) and Harizi (1985) as shown in Table 4.8. Cotton yields reported by Hawkins Associates (1982) and Harizi (1985) were higher than those presented in Table 4.7. Without good season yields for sorghum estimating the weighted mean and hence comparing these results to published yield figures was difficult. The values for an average season were however. reasonably close to the mean yields reported by Hawkins Associates (1982) and Brunt et al. (1986), but were much lower than those reported by Harizi (1985) as shown in Table 4.8. None of these authors reported yields on different soil types. The data collected from field maps show reasonable agreement with published yields for this part of Zimbabwe and were within the range one could expect from national data sources. There may, however, be problems with the repeatability of these measurements; The mapping of household fields and requests of yields was repeated twice each for two VRs (one male, one female). This facilitated examining the repeatability of yield estimates derived in this manner. In both cases the responses differed quite markedly from one mapping exercise to another. Differences occurred in the statements of the acreages as well as in the total yields obtained. These anomalies, as well as the notable variation in yields reported for the Zambezi Valley (Table 4.8), prompted development of the possibility scoring ’method for estimating yields. The expected yields of each soil type (Table 4.9) were weighted by that soil type expressed as a proportion of the total of cultivatable soils (bepe = .13, jecha 125 = .77 and river jecha = .1, when calculating cotton and maize values; bepe = .144 and jecha = .856 when calculating sorghum values). For jecha soils the average of virgin and five year cultivated jecha expected yields were used to derive the weighted means. These values are shown in the weighted mean column of Table 4.9. The weighted means across seasons were calculated by weighting the expected yields of each season by the proportion of those seasons calculated from Table 4.2 (good=.3, average=.3 and poor=.4). The yields of all crops, on all soil types and for any season show considerable variability (1' able 4.9) with co-efflcients of variation between 10 and 133%. Variability among seasons appeared greater than that among soil types. VRs reported highest maize yields on river jecha soils in any season. The low per household proportion11 of river jecha fields (p=.46 of households have river jecha and p = 0.1 of the lands are river jecha) reduced the impact of river jecha yields on household needs satisfaction. Maize yields on both bepe and river jecha were slightly higher in an average season than in a good season. This is consistent with what we expect because the high clay content of the bepe soils makes them waterlogged in wet years, thus reducing yields (Mittra and Stickler, 1961). On river jecha soils the flooding of the Angwa river in wet years reduces yields through waterlogging, lodging or being washed away. Poor season maize yields were higher than what we might expect on reading the literature; Yields on 11 The proportion of river jecha fields was estimated using a digitized soil map. The procedure is described in Chapter 5. 1 26 Expected values for maize, cotton and sorghum yields (kg ha“) on three soils and for good, average and poor rainfall seasons. Source, possibility diagrams. Table 4.8 Crop Season Bepe Virgin 5yr River Weighted means class' jecha2 cult jecha 1 2 G 1800 2000 1100 3300 1750 1600 (475) (530) (351) (427) (797) (621) Maize A 2000 1350 1000 3500 1500 1300 (629) (571) (424) (330) (885) (614) P 700 550 250 1350 550 450 (218) (178) (109) (209) (362) (236) Wtd. 1450 1200 750 2600 1200 1050 Mean‘5 (745) (748) (504) (1055) (883) (709) G 850 1200 650 1200 950 900 (385) (395) (545) (395) (526) (531) Cotton A 1100 1050 700 1400 950 900 (484) (466) (446) (370) (506) (494) P 250 250 150 400 250 200 (129) (151) (174) (272) (139) (163) Wtd. 650 750 450 950 650 650 Mean (508) (547) (467) (559) (524) (527) e 500 1100 300 NA‘ NA 700 (149) (1 127) (181) (834) Sorghum A 1000 850 250 NA NA 600 (285) (259) (323) (422) P 50 150 50 NA NA 100 (26) (98) (25) (as) Wtd. 550 650 200 NA NA 450 Mean (439) (752) (233) (583) 1. Season class: G = good; A = Average; P a Poor. 2. Virgin jecha Jecha soils that had not previously been cultivated. 3. 5yr cult jecha. Jecha soils that had been cultivated for five years continuously. 4. NA - not applicable. Sorghum was not grown on this soil type. 5. Weighted mean across soil types. The expected value for each soil type was weighted by that soil type expressed as a proportion of all cultivatable soil. For jecha the mean of virgin and 5 year cultivated expected values were used. 1 = All soil types including river jecha. 2 = All soil types excluding river jecha 6. Weighted mean across seasons. The proportion of seasons in each class, G=.3, A=.3 and P=.4, were used to weight the expected values for each season. 127 bepe were comparable with national averages for the CA5, 695kg ha". (T attersfield, 1982) and the yields on river jecha were well above the all-season averages reported for the Zambezi Valley (Table 4.8). In contrast, the poor season yields of cotton and sorghum were very low. These figures add support to the argument that sorghum is not as resistant to drought as is often believed (Shumba, 1990) and suggests why farmers continue to select maize as- their major food crop (Johnson, 1992) even when they are considered to be outside the region of dryland farming, as Masoka is (Whitlow, 1980). The very noticeable drop in expected yields from an average to a poor season in each of these crops was remarkable. This implies a steeply sloping sigmoid curve rising rapidly between a poor season and an average season and then leveling out on (jecha soils) or declining slightly (on bepe and river jecha) in good seasons, trends that have been shown in the literature for jecha (Phia, 1992). There appeared to be little difference in expected yields among soil types for cotton and for sorghum. Expected yields were lower on cropped jecha soils than on virgin jecha soils (Table 4.9) a trend that were consistent with the findings of Reynolds (1991). Cotton yields also appeared to be more variable on cultivated soils than on virgin. The most striking feature of the data derived from VR probabilities were the great variability in yields, among soil types and among seasons (Table 4.9). This variability was reflected in the literature for the eastern mid-Zambezi Valley (Table 4.8). Mean maize yields derived from VR knowledge were similar to those reported 128 in the literature (Table 4.8). If we take the cultivated jecha yields as a yardstick, (these soils being the most comparable to those of the eastern mid-Zambezi Valley as a whole) then the VR derived yields were entirely consistent with published yield records (Table 4.8). VR derived mean yields were also similar to the yields households achieved (Tables 3.2, 4.7). The average maize yields reported for Masoka appear to be slightly higher than the averages reported for the eastern mid-Zambezi Valley and were also slightly higher than the national average, 900kg ha", for the period 1952 to 1999 (Ashworth, 1990). Perhaps the most useful data with which to compare the yields derived from VR probability estimates was the frequency distribution of maize yields reported by the Farm Management Research Section of the Economics and Markets Branch, Ministry of Lands, Agriculture and Rural Resettlement (MLARR, 1989). This positively skewed distribution, with a mean (for NR III, N and V sites) of 1100 and a standard deviation of 1060kg ha'1 was close to the VR estimated distribution, in terms of the mean and spread of the data. The MLARR data for the 1989/90 season showed a similar distribution. The published data for average cotton yields were also highly variable with average yields between 500 and 1800kg ha'1 being reported (Table 4.8). The VR derived yield estimates were similar to what households achieved in good and average seasons (Table 4.7). The VR derived poor season cotton yields were much lower than what households achieved in the drought season of 1990/91 (Table 3.2) and less than the drought season yields reported by Harizi (Table 4.8). 129 In general, the average cotton yields predicted by the VRs were lower than the trends reported for the eastern mid-Zambezi valley (Table 4.8) but were close to the national average of 700kg ha“, for the period 1982 to 1989 (Ashworth, 1990). The MLARR (1989) data for cotton yield distributions are from a smaller sample size than the maize data (n=46 vs. n=255) so the distribution shape was not as clear. The mean of 630 and standard deviation of 509kg ha'1 were, however, very similar to the VR derived yields. The published data for average sorghum yields also showed considerable variation (Table 4.8) with average yields ranging from 540 to 900kg ha". The achieved, poor season yields (Tables 4.7 and 3.2) were higher than those predicted by the VHS (Table 4.9), but the great variability made detecting true trends difficult. The poor season yields achieved by Masoka households (Table 4.7) were closer to those predicted by the VRs (Table 4.9) than were the mean yields published in the literature (Table 4.8). The average season sorghum yields achieved by households and those predicted by VRs were very similar. In general. the sorghum yields households were likely to achieve in Masoka were lower than yields reported in the literature but were higher than the national average of 350kg ha" for the period 1982 to 1999 (Ashworth, 1990). Deriving statistically satisfying tests of our hypotheses from these yield data was difficult. Inconsistencies in VR responses to field mapping yield estimates indicated that this method may bias the results. The consistency in yield estimates derived from two separate data collection exercises suggested that this method 1 30 did not bias the results. Whilst collecting these yield data, I brought to the attention of the VHS the inconsistencies in yield estimates among methods. After much debate and reflection, the VHS re-examined each of their yield possibility diagrams and confirmed that these results represented their best estimates of yield relationships. These factors made us confident that the method was unbiased and provides reasonably accurate and precise information on yields under the specified conditions. Over the range of conditions investigated, these results are likely to be at least as accurate and precise as published data. Performance measures VRs were asked to develop a spidergram that would indicate what factors determine a household’s ability to satisfy its needs (Figure 4.17). Knowledge was considered the most important factor (RlW=.42) followed by the availability of labor (RlW=.25) and access to good soils (RlW=.25). Wealth (RIW= .08) was considered least important. Each of these factors was then further defined in secondary, and in some cases tertiary, levels of indicators. Factors were included if households could be observed to posses or not posses that factor. Whilst developing Figure 4.17 VRs focused much of their discussion on the importance of, and their ability to observe or measure, a household’s knowledge. The harvest component of knowledge involved establishing whether households had good harvests or not. Households consistently identified as the community’s best farmers were identified as such due to their high total production. When, as a group exercise, the yields 131 .mooo: 9mg mg 9.35:8 b.2525: o .o 5852. 55 mmmmmm 2 mosmamE 3:25.55 32.50 m> N 5. 2:9". 582» .8... .80.. ._ so“. >_ so“. Mm. \ unis“. BIS“. # mm. 8838 / \ K 9358 mm. NO bogub 0cm \ £905 1 Lone. meg: . 52m . cocom bopcooom / c_ 5.5.20 N F . 259 52a / / m w. m__0m 88 its. 583 . mm. 9.52. \ .5282. cozooamzom mode: .3525 895 8 .0 382m new . \ 5522 3.55: ll: % mm. 8.3 cm. . baa maps—.5 Eco. 8.8 558.95 _ 8582 55¢ 5 658.39 EoEmmocoE 132 per hectare of some of these respected farmers were evaluated, their per hectare production was found to be much lower than many other producers. Most of the VHS however, retained their belief that these were still the best farmers in Masoka. Management of yields involved allocating points to a household for each month it had grain in its granary - up to a maximum of 15 points. If households had sufficient labor to weed their fields in good time, guard their fields from animals and birds as well as harvest and store crop yields, then these factors contributed importantly to that household’s ability to satisfy its basic needs. The VRs considered a household having access to jecha soils as being more important than access to other soils types in determining the household’s ability to satisfy its basic needs. Education, hiring of labor and hiring a tractor for ploughing as well as the qualities of the homestead building, were important indicators of household wealth. VRs were asked to assign each household in the community to a class according to the household’s ability to satisfy its basic needs. Three classes were defined: households that always satisfied their needs, those that sometimes satisfied their needs and those that rarely satisfied their needs. Of the households classified (n=98) 19% were classed as needs satisfiers, 45% were classed as households who sometimes satisfied needs and 36% were classed as households who seldom satisfied needs (Table 4.10). The numbers of adults differed among these three classes (p< .01, n=98, F =6.349) as did the number of children (p< .01 , n=98, F =5.68) but the relationship with class was not clear. Households that 133 sometimes satisfied needs generally had fewer adults and children than either of the other classes which had similar numbers of both adults and children. The number of years households had been resident in Masoka was not statistically different among household needs satisfaction classes. Households in the higher well-being classes were more likely to employ local labor than households in the lower well-being classes (p<.001, n=98, Spearman’s rank correlation r = -.271). Households in the lower well-being classes were more likely to work for other households than households in the higher well-being classes (p<.001, n=98, Spearman’s rank correlation r = .35). The VRs developed a list of all households in the community and identified those who regularly employed members of other households as agricultural workers, as well as which households were regularly employed. Twenty one households (15%) were identified as regular employers of either school children or members- of other households. This value is similar to the 13% of households employing labor in Mangwende CA, reported by Shumba (1985). Thirty two households (22%) were identified as being regularly employed by other households within the community. Thirty seven households (26%) were considered to be regular receivers of incomes from working for other households, from formal employment within or outside Masoka or from both of these sources. Only nine households (6%) were thought to receive remittances from relatives living outside of Kanyurira Ward. The proportion of households regularly employed by other households was almost double that reported by Jackson and Collier (1991) in their 134 Table 4.10 Mean (standard deviation in parentheses) number of adults, number of children andyears residenceln Masokaforclassesofhouseholdwell-being. Household characteristic Household well-being classes Always Sometimes Seldom satisfy satisfy satisfy needs needs needs (Class 3) (Class 1) (Class 2) Number of adults 2.4 1.8 2.2 (0.61) (0.48) (0.79) n=19 "=44 "=35 Number of children 5.4 2.8 4.0 (3.89) (1.82) (3.51) n=19 n=44 n=35 Family size (adults + children) 7.8 4.6 6.2 (4.22) (2.02) (3.69) n=19 n=44 "=35 Years resident in Masoka 15.9 10.5 9.6 (8.53) (10.91) (9.52) n=16 n=41 n=35 Dependency ratio (children / adults) 2.2 1.5 2.1 (1.3) (1.07) (219) n=19 n=44 n=35 Number employing local labor 10 6 5 Number employed as local labor 0 7 17‘ 1. The total number of employed reported here are less than 32 households described inthetextasthesetabledvaluesdo notinclude VaDomahouseholdsworkingfor Masoka households but the totals reported in the text do. national survey, whilst the proportion of households receiving remittances was very much less than the 37% of households reported by these authors. Stanning (1989) also reported much higher proportions of households receiving remittances in her study in Hurungwe District (88% of households, Table 5, p12) as well as 64% of households receiving local off-farm wages. Stanning, as well as Jackson and Collier, reported that local off-farm incomes contributed little to total household incomes. The 15% of regular employers reported by the VR was similar to the 135 findings of Jackson and Collier (1991) that around 10% of communal area households dominated production and of Starining (1989) who reported 30% of households in Hurungwe accounted for more than 75% of marketed maize production. Households whose members were employed as agricultural workers by other households were resident in Masoka for a shorter period of time than households that were not regularly employed and had a higher proportion of children to adults (dependency ratio) than households that were not regularly employed (Table 4.11). Employing households had a slightly larger number of Table 4.1 1 Mean (standard deviation in parentheses) for characteristics of households in each of four employment classes. Household employment classes ””8”“ °' “3mm" Employing Not- Employed Not- emplovine employed Number of adults 2.5a 2.09 2.0 2.1 (0.75) (0.58) (0.62) (0.68) n=21 n=77 n=24 n=74 Number of children 4.0 3.6 4.7 3.4 (4.14) (2.78) (3.53) (29) =21 n=77 n=24 n=74 Family size (adults + 6.6 5.6 6.8 5.5 children) (4.37) (3.05) (3.8) (3.18) n=21 n=77 =24 n=74 Years resident in Masoka 11.3 11.0 7.6b 12.3b (7.20) (10.84) (8.0) (10.60) n=19 n=73 n=24 =68 Dependency ratio (children/ 1.6 1.9 2.5c 1.7c adults) (1.5) (1.59) (2.16) (1.24) n=21 n=77 n=24 n=74 1. Values followed by the same letter were significantly different (p<.05): a) t=-3.766, DF=96 b) t=2.003, DF=90 c) t=-2.256, DF=96 1 36 adults per household than non-employing households (Table 4.11). These results reconfirm the need to differentiate among households and incorporate household interactions in our predictive analysis. The cash and labor budget data provided by the VRs were used to develop partial budgets for cotton, maize and sorghum (Tables 4.12 to 4.14). These tables give a clear indication that households faced complex decision making processes in selecting crop type, crop variety, soil and input mixes to satisfy their objectives. In each of these budgets, the B scenario approximates what the wealthier households might actually do. Scenario C showed the effect of changing producer prices on this scenario for maize and sorghum. Scenario D for maize and sorghum and C for cotton showed the effects of poor season yields. Scenario A was an all factors base case. Using the government stipulated producer price in the maize budget (Scenario C, Table 4.12) indicated quite clearly why no households sold maize to the Grain Marketing Board (GMB) in the past few seasons; The returns to. household labor were well below the local daily wage rate (232.19 day") and households stand to make substantial losses in poor seasons. With poor season yields and the Masoka maize price, household labor earned a little less than the local daily wage (Scenario D, Table 4.12). The returns to labor and to land for cotton were substantially higher than those for maize. The returns to labor and land for cotton in a poor season were however, very much less than those for maize. What was interesting to notice was the relatively high return on investment for maize, in an average and poor season, 137 Table 4.12 Partial budgets for one hectare maize production enterprises with different input and pricing scenarios. 1991/92 season prices. Costs and returns ha’1 Scenario Scenario Scenario Scenario A‘ 9 c o FIXED COSTS Transport to Guruve 22.00 22.00 22.00 22.00 Food 15.00 15.00 15.00 15.00 SUB-TOTAL 37.00 37.00 37.00 37.00 VARIABLE COSTS Seed (25kg ha") 45.55 45.55 45.55 45.55 Seed transport (231.00 I bag) 2.47 247 2.47 2.47 Ploughing (23149.20 ha“) 149.20 0.00 0.00 0.00 Fertilizer (25197.50 ha") 197.50 0.00 0.00 0.00 Weeding labor (25219 day") 51.45 51.45 51.45 51.45 SUB-TOTAL 445.25 100.45 100.45 100.49 HARVEST COSTS Bags (@3200 ea) 23.07 23.07 23.07 10.03 Total Costs 505.35 150.55 150.55 147.51 Returns 515.75 519.75 25242 355.90 NET MARGIN 31240 559.20 101.97 209.39 RETURNS TO OWN LABOR 3.22 5.15 0.95 1.95 RETURNS PER HECTARE 312.40 559.20 101.97 209.39 RETURNS TO iNVESTiiliENTz 0.52 4.10 0.09 1.41 1. Scenario A: Yield level 1050kg ha". hired labor for two weeding rounds, tractor ploughed, 124 kg ha'1 each of Compound D and Ammonium Nitrate fertilizers, household uses own labor for all activities on 0.405ha. Local Masoka maize price used (230.78 kg"). Scenario B: As for Scenario A except hand cultivated instead of tractor cultivated, no fertilizer and hired labor used for two weeding rounds. Scenario C: As for Scenario 8 except producer price of maize used (230.25 kg"). Scenario D: AsforscenarioBexceptpoorseasonyield levelused(450kg ha' ). 2. Returns to investment = Gross returns / total costs. 138 Table 4.13 Partial budgets for one hectare cotton production enterprises with different input scenarios. 1991/92 season prices Costs and returns ha'1 Scenario Scenario Scenario A B c FIXED COSTS Transport to Mahuwe 44.00 44.00 44.00 Food 30.00 30.00 30.00 Transport 8. cost of empty bales 15.00 . 15.00 15.00 Knapsack spray hire (for season) 15.00 15.00 15.00 SUB-TOTAL 104.00 104.00 104.00 VARIABLE COSTS Seed (25kg ha") 30.95 30.55 30.99 Seed transport (231.00 I 10 kg bag) 247 2.47 2.47 Ploughing (25149.20 ha") 149.20 0.00 0.00 Fertilizer & chemicals (Agritex 1 acre packs 345.80 345.80 345.80 @23140.00/pack) Weeding labor(@2$2.19 day") 10290 51.45 51.45 SUB-TOTAL 530.25 430.50 430.50 HARVEST COSTS Harvest labor (@251900 bale") 59.13 59.13 19.23 Transport (@zsazbo bale") 105.12 105.12 3241 SUB-TOTAL 154.25 154.25 50.54 Total costs 947.05 599.95 555.23 Returns 1919.50 1919.50 591.42 NET MARGIN 1071.45 1219.55 5.19 RETURNS TO OWN LABOR 10.51 10.57 0.05 RETURNS PER HECTARE 1071.45 1219.55 5.19 RETURNS TO INVESTMENT 1.25 1.74 0.01 1. Scenario A: Yield level 650kg ha". Hired labor used for weeding two rounds, tractor ploughed, cotton producer price 25292 ltg". Household labor used exclusively for all activities on 0.405ha. As for Scenario A except hand cultivated rather than tractor ploughed. As for Scenario 8 except poor season yield used (200kg ha"). Scenario 8: Scenario C: 139 compared to cotton. This suggested, that for poor households that are short of money to invest in crop inputs, cotton was not as attractive an investment as maize. Sorghum however, showed substantially higher returns to investment in an Table 4.14 Partial budgets for one hectare sorghum production enterprises with different input and price scenarios. 1991/92 season price data. Costs and returns ha" Scenario Scenario Scenario Scenario A1 5 c 0 VARIABLE COSTS Seed (7.41 kg ha") 3.71 3.71 3.71 3.71 Harvest labor (@25219 day") 27.05 27.05 27.05 27.05 Weeding labor (@25219 day") 51.45 0.00 0.00 0.00 Total costs 9250 30.75 30.75 30.75 Returns 247.50 247.50 100.39 55.53 NET MARGIN 159.75 220.20 59.53 24.55 RETURNS TO OWN LABOR 3.55 4.77 1.51 0.54 RETURNS PER HECTARE 159.75 220.20 59.53 24.99 RETURNS TO lNVESTMENT 2.05 7.15 225 0.91 1. Scenario A: Yield level 450kg ha". Hired labor used for harvest and two weeding rounds, Masoka price (230.55 kg' 1) used. Household labor used exclusively for all activities, except harvest, on 0.405ha. Scenario 8: As for Scenario A except hired labor only used for harvesting. Scenario C: As for scenario 8 except producer price (230.22 kg") used. Scenario D: AsforScenarioB exceptpoorseasonyield (120kg he“) used. average season than either maize or cotton and greater returns than cotton in a poor season. We can clearly see from scenario C of sorghum why few households sell grain sorghum to commercial buyers - their returns to labor, land and their initial investment would be just a fraction of what they earn by trading within 140 Masoka. Returns to labor for sorghum were double the local daily wage in an average season but only one quarter the local wage in a poor season. If we take the expected yield for each season on a each soil type, multiply it by the probability of a season type and then sum these for each season we obtain the expected returns households fees for each crop on each soil (Table 4.15). Looking at the returns to land and labor, cotton was clearly a more attractive Table 4.15 Expected returns (23 ha") to land, to labor and to initial investment for maize, cotton and sorghum on each soil type. Retums to land Returns to Returns to (23 ha") labor (ZS investmertt Crop Soil ha") (25 25") Bepe 951.25 8.90 5.45 Maize Virgin jecha 802.34 7.51 4.67 Cultivated jecha 422.55 3.95 254 River jecha 1540.30 17.22 9.09 Bepe 1314.35 11.50 1.70 Cotton Virgin jecha 1559.68 13.64 1.93 Cultivated jecha 719.49 5.30 1.00 River jecha 2005.55 17.55 241 Bepe 250.79 5.09 9.14 Sorghum Virgin jecha 195.43 7.11 10.67 Cultivated jecha 72.15 1.57 2.35 crop on any soil type than either maize or sorghum and sorghum was by far the poorest performer. The returns to labor for maize and sorghum were close for comparable soils except maize out-performs sorghum on cultivated jecha. When 141 we look at the returns to initial investment however, a remarkably different picture emerges: Overall, cotton was easily the worst performer and the least attractive option. Returns to dollars invested in sorghum were far higher than maize except on cultivated jecha where maize returns to investment were a little higher. These findings had obvious importance for developing predictions as to what we might expect different classes of household to grow and on what soils. We might expect the poorest households to focus on sorghum production because they may still achieve a respectable return on their investment. We also might expect strongly risk averse households to focus on maize, because even in poor seasons on the poorest soil, maize yields positive returns to land, labor and to initial investment. Both cotton and sorghum however, might yield negative returns in poor seasons on cultivated jecha soils - the soil type that comprised most of the average household’s land holdings. Households that try to maximize their incomes and are not labor constrained might focus on cotton as the returns to land and labor for this crop are far higher than for either maize or sorghum. To summarize, the performance criteria developed by the VR and extended in the partial budget analysis, the VRs identified household knowledge, the availability of labor to a household, a household’s access to good soils and the wealth of a household as being key determinants of a household’s ability to satisfy its basic needs. Of these factors, knowledge was considered almost twice as important as any of the other factors. Households in the community were classed into one of three well-being classes; those who always satisfied their needs, those 142 who sometimes satisfied their needs and those who seldom satisfied their needs. Households in the first of these classes were more likely to be employers of local labor whilst households in the last of these classes were more likely to be local farm wage employees. Employed households had generally lived in Masoka for fewer years and had a greater number of dependents than employing households. The data presented in Table 4.15 provide support to the VR assessment of the importance of soils and labor; households that only had access to cultivated jecha fields were less likely to be able to satisfy their needs than those who had access to all soil types, particularly river jecha and bepe. Households that were labor constrained might not be able to either clear new fields so that exhausted jecha soils could be rested or they may not be able to meet the higher labor demands of cotton or maize. Both factors would reduce their ability to satisfy needs. Problem formulation In this section, the needs that VRs indicated households attempted to satisfy were quantitatively presented. These data were important in that they were used to set the household objectives to be incorporated in latter simulation analyses. The detailed needs for an average woman, an average man and an average child are presented in Tables 4.14, 4.15 and 4.16. The importance weights placed on these needs indicated which needs a person cannot live without (weight = 5). 143 to which needs are essentially luxury (weight = 1). The weights assigned to each month reflected the periods of greatest scarcity (weight = 1) to least scarcity (weight = 5). For both groups maize was the principle food with chickens and pumpkins also being important. There were some notable inconsistencies when comparing these data and the RIW associated with needs identified in Figures 4.5 to 4.8. In their needs calendar, female VRs identified both fresh vegetables and sorghum as being more important than pumpkins (Table 4.14) but in their needs spidergram, female VRs identified vegetables and sorghum as being less important than pumpkins (Figure 4.6). The spidergrams were developed in mid-March, a period towards the end of the growing season but where pumpkins and pumpkin 12 leaves were readily available and fresh vegetables are becoming available. The needs calendars were developed in late June when all forms of fresh food were becoming scarce. The shifts in perceived relative importance of these items is therefore, not surprising. This suggests that the importance weights that VRs associated with needs may be seasonally biased, particularly in regard to seasonally available inputs to needs satisfaction. Female VRs suggested that women needed (used) a greater diversity of food types and needed more of each food item. Males required about one third more cash each year than women (Table 4.16). Children had approximately the same food needs as men but required only about half quantity adult males Pumpkin leaves are an important vegetable relish; a side dish used to garnish the staple food of maize meal porridge. 144 3§§§39 2555855253551..o§§E255585umEmas85§E_d 55255925» iconocoumfimasaoaom; OZ. no. 0.. 5 8' mp 559:9: 9"“7 99* have as 88 0.. or V 0.. 9. mw NM 88 BIDNIO N cw ON V 88 5101'- 0.. 80 “’8 O P '88 ase~~v 9. Op on m N mw ON mp mp m Ni!) 88 mmIOIOQVVVVVV‘COGCONNv-v-v-Pv- a... 5...: a... moofiEo... 2095.530 25.0.2552, a... 95.5 85a mmmm a... 8998 503m maidEbm .85.. 89m .3. 8590 .5... E398 030E .520 a... seem a... fies. .923. 85.5.5 a... 8.3559, been a... 5m .52... do) 28.“. ad :80 a... 555.2 45.0... 5:05 (It) 2|!) LL filo E .5803 0935.6 :3 .0. 92903 3.238 memo: 3.55:. ecu $550.03 0052.038. so... .800: 3.5.5.2 meooz A- Is 339.com m We 03o... 145 .Sozgzoncobucubaxa. u 9.965835 222.03: um £92535:an d :9.» 22.2.3 .1. P .5305 u m .3903 323$ .. 5 mp mu m m 9.8 m h h: IDIDQMNNNCOVNPPO c O O O ON ON ON ON ON ON ON ON ON ON mp mp m.. m— a... 5.“. a... 5...: 38 .9. 8988 $95 .8. 850.0 .9: Suzuki 88:. :35 .9: was. 36:3. 89.0.5 a... 8.859.. 8.5 .295. a? 58.“. .8. :80 a... gas. 45.0... (’3 o O m v z 0 m < q. __. 2 < '1; O v 2m LLID '50) 5:0: .:m:. 3926 cm .6. “3:92., 3.28» 88: 25:08 new «@5599: 3:2..an :35 .88: 25:05. 382 4:5. 2.28m h ...v 030.. .52.; 8 c8 05 233. n F .20.. .8: 205.800 a 0 20.03 85:85. .0 :90 20525 n p .585 u 0 £0.03 E0000 .— 146 3000. 0:00 .0030 0:0..0:..0.0>> 0.005.030 6... 05:05.90 .0... 5...: mmmw .0... 08.08.. .0030 .8. 09.8.0 a... .002 0.2;. 08.0.5 .005. 00> :8... .8. :80 .0... 3.0.2 O O O O». on V800 0 0. 0. IDIOVCONO‘JNNNNPNP 00002 0 0 0 0 0 0 0 0 v 0 0 c A-..>>2.280 “1; .0... 0.00.. 0" v. 0.30 000.05 :0 .0. 05903 3.0.000 0000: 25:0... 0:0 00:55.03 00:00.09... .0000: 25:22 09.0 0.00... 147 required (Table 4.18). The RIW given to needs in the male VR spidergrams (Figures 4.7 to 4.9) appeared more consistent with the weights given to needs identified in Table 4.17 than those of the women. Maize, chickens. vegetables and sorghum were consistently rated as important needs. For the most part, the data presented in Tables 4.16 to 4.18 cannot be validated by comparative reports as. with the exception of Stanning’s work (1987a) there are no published accounts of Zimbabwean communal area household needs or objectives. The maize needs identified in Tables 4.16 to 4.18 were similar to the data published elsewhere. Jayne and Chisvo (1991) reported per capita grain requirements of 219kg while Akwabi-Ameyaw (1990) reported mean per capita maize retentions in Mufurudzi resettlement scheme for 1984/85 season of 232kg. Stanning (1987a) reported per capita maize grain retentions, in Hurungwe, of about 190kg for the 1983/84 and 1984/85 seasons but only 130kg per capita in the 1985/86 season (Stanning, 1987b). The sorghum retentions reported by Stanning (1987a) however. were very much lower than those reported by Masoka VRs. Periods of greatest food scarcity were the months of September through January - the period immediately prior to the growing season through to the time of the first harvests. The male VR needs matrix showed November to be the month of greatest scarcity. Children have a greater proportion of months with high scarcity weights than either adult males of females (Tables 4.18. 4.16, 4.17). 148 SUMMARY AND coucwsrous An account of the methods of identifying key actors and groups, of establishing the needs of client groups, of identifying the components of the system that households use to satisfy needs as well as the criteria they use for evaluating needs satisfaction is presented. The results of applying these methods to an analysis of the Masoka agroecosystem are described and discussed. This study differs from reports in the literature by using local farmers to state their objectives, identify their production system and the controls they use to achieve their objectives. Households attempt to satisfy cash and food needs from a broad range of agriculturally based, as well as some non-agricultural production activities. The relative importance of each of these needs. as well as the inputs to satisfy these needs, were identified. Major inputs to achieving good yields of maize, cotton and sorghum were identified as rainfall, soil, crop variety and labor. Each of these factors, except crop variety, were analyzed in greater detail. Household differentiation. as well the interactions among households, were shown to be important factors to include in our predictive analysis. VRs used possibility diagrams to indicate the probability of households achieving specific yield levels of each of three crops, on particular soil types in any season class. Expected yield values were derived from these possibility diagrams and these yields compared favorably to published yield levels under similar conditions. Despite the great variability in the yields reported by the VHS, as well 149 _ as those reported in the literature, the yield functions developed from VR information are considered to be reasonably accurate representations of the yields households achieve. These yields were combined with labor input and price data to develop partial budgets for maize, cotton and sorghum. These budgets indicated the far greater profitability of cotton when considered from the perspective of returns to land and labor but when evaluated from the perspective of returns to initial investment cotton was a poor performer. From this latter perspective, sorghum was easily the best performer, except on cultivated jecha. Soil type had a notable effect on the returns to crops; maize generated the best returns on river jecha, followed by bepe; cotton generated best returns on river jecha and then virgin jecha; and sorghum returns were best on virgin jecha. Declines in soil fertility on jecha soils generally more than halved the returns from a particular crop. VRs indicated that farmer knowledge, access to labor, access to good soils and wealth were key determinants of a household’s ability to satisfy needs. The measures used to record the VR responses to examination of their knowledge were shown to be unbiased for all cases except in collecting yield information. These biases were due to the field mapping method of eliciting yield information. The possibility scoring method provided yield distributions that were repeatable as well as being similar to published yield distributions. The pattern of yields by season, on different soil types, were also found to conform to theoretical expectations which increased our confidence in the methods and in the results. 150 Despite the difficulty of quantitatively testing the predictions posed earlier in this chapter, the results of these analyses provide little justification for us to accept the alternatives to the predictions: 1) That households are able to accurately describe the patterns or their environment. 2) That households can correctly describe the results they are likely to achieve using commonly used combinations of controllable inputs with different resources and under commonly experienced environmental situations. We may therefore, conclude that our model of accurate and reliable local knowledge being essential for survival in Masoka can, for the moment at least, be accepted. We may then feel confident in using this information to develop a simulation model with which to investigate various household, community and policy level questions on the ability of households to satisfy their basic needs in the future. BIBLIOGRAPHY Adams. J.M. 1991. Female wage labor in rural Zimbabwe. World Development, 19, 1 63-1 77. Adelman, L 1992. Evaluating decision support and expert systems. John Wiley & Sons, Inc., New York. Akwabi-Ameyaw, K 1990. The political economy of agricultural resettlement and rural development in Zimbabwe: The performance of family farms and producer cooperatives. Human Organization. 49, 320-338. Alreck. P.L and RB Settle. 1985. The survey research handbook. Richard D. Irwin Inc., Homewood, IL, USA. Anderson, I. 1986. Communal Land Physical Resource Inventory. Chemistry and Soil Research Institute, Harare, Zimbabwe. Ashworth, VA. 1990. In: Agricultural technology and the communal farm sector. Background paper to the Main Report, Zimbabwe agriculture sector memorandum. The World Bank, Washington. DC. USA. Bawden. R.J. 1992. Systems approaches to agricultural development: The Hawkesbury experience. Agricultural Systems. 40, 153-176. Bawden. R.J. and R.L. lson. 1992. The purposes of field-crop ecosystems: Social and economic aspects. In: C.J. Pearson (Ed.), Ecosystems of the world. Field crop ecosystems. Elsevier, Amsterdam. Holland. Boserup. E. 1965. In: The conditions of agricultural growth, George Allen & Unwin Ltd, London, UK Brokensha, D.W., D.M. Warren and 0. Werner. 1980. Indigenous knowledge systems and development. University Press of America, Lanham, MD. USA. Brunt, M.A., V.J. Clarke and C. de Greling. 1986. Tsetse area development and land use planning in the Zambezi Valley, Zimbabwe. Food and Agriculture Organization of the United Nations, Rome. Campbell, 8M. and R.F. Du Toit. 1988. Relationships between wood resources and use of species for fuel and construction in the communal lands of Zimbabwe. Monographs in Systematic Botany of Missouri Botanic Gardens, 25, 331-341. 151 152 Campbell, B.M., R.F. Du Toit and CAM. Attwell. 1989. In: The Save Study. Relationships between the environment and basic needs satisfaction in the Save catchment, Zimbabwe. Supplement to Zambezia, University of Zimbabwe. Harare, Zimbabwe. Carter. S.A. Chidiamassamba, P. Jeranyama, B. Mafukidze, G. Malakela, Z. Mvena, M. Muduhara, N. Nabane, S. Van Oosterhout-Campbell, l. Price and N. Sithole. 1993. Some observations on wealth ranking after an RRA looking at soil fertility management in northeastern Zimbabwe. RRA Notes, 18, 47-59. Chambers, R. 1983. Rural Development: Putting the Last First. Longman Scientific and Technical, Harlow, England. Chiduza, C.S.R. Waddington and M. Rukuni. 1992. On-farm evaluation of short season sorghum and fertilizer for smallholder farmers in a semi-arid region of Zimbabwe. Farming Systems Bulletin, 11, 25-33. Chopak, C.J. 1991. Determinants of rural incomes in communal areas of Zimbabwe: Household food security implications. PhD dissertation. Michigan State University, East Lansing, Michigan. Coudere. H. and S. Marijse. 1991. Rich and poor in Mutoko communal area. In: Mutizwa-Mangiza, ND. and A.H.J. Helmsing (Eds). Rural development and planning in Zimbabwe, Avesbury, Aldershot, UK. Cutshall. CR. 1989. In: Masoka / Kanyurira Ward. A socio-economic baseline survey of community households. Centre for Applied Social Sciences, University of Zimbabwe, Harare, Zimbabwe. Department of Meteorological Services. 1981. Climate handbook of Zimbabwe. Government of Zimbabwe, Harare, Zimbabwe. Department of Meteorological Services. 1975. Rainfall intensity in Rhodesia. Government of Zimbabwe, Harare, Zimbabwe. Derman, B. 1992. Careless development: The mid-Zambezi rural development project. Culture and Agriculture. 44, 2-9. Derman, B. 1990. In: The unsettling of the Zambezi Valley: An examination of the mid-Zambezi Rural Development Project. Centre for Applied Social Sciences, University of Zmbabwe, Harare, Zimbabwe. Drinkwater, M. 1991. The state and agrarian change in Zimbabwe’s communal areas. St. Martin’s Press. New York. 153 Elwell, HA 1985. An assessment of soil erosion in Zimbabwe. The Zimbabwe Science News, 19, 27-31. Elwell. HA. 1980. In: Design of safe rotational systems. Department of Conservation and Extension (AGRITEX). Harare, Zimbabwe. Elwell. HA. and MA Stocking. 1988. Loss of soil nutrients by sheet erosion is a major hidden farming cost. The Zimbabwe Science News, 22, 79-82. Elwell, HA and MA. Stocking. 1984. Estimating soil life-span for conservation planning. Tropical Agriculture, 61, 148-150. Farrington, J. and A. Martin. 1988. Farmer participation in agricultural research: a review of concepts and practices. Agricultural Administration Occasional Paper No 9. Overseas Development Institute, London, UK FSRU. 1985. Annual Report of the Farming Systems Research Unit. Department of Research and Specialist Services, Government Printers, Harare, Zimbabwe. Gil, G.J. 1991. But how does it compare with the real data? RRA Notes, 14. Grandin, BE. 1988. Wealth ranking in smallholder communities: A field manual. Intermediate Technology Publications Ltd. Rugby, UK. Grant, PM. 1981. The fertilization of sandy soils in peasant agriculture. Zimbabwe Agricultural Journal, 78. 169-175. Harizi, KB. 1985. Zimbabwe. Mid-Zambezi Valley agricultural development project. Agricultural production, FAO/ ADB Working Paper No. 2. FAQ, Rome. Hasler. A.R.P. 1992. The cultural and political dynamics of Zimbabwean wildlife resource use in the Zambezi Valley: A case study of Chapoto Ward. PhD dissertation. Michigan State University, East Lansing. Hawkins Associates. 1982. Mid-Zambezi Development Study. Agricultural Development Authority, Harare. Zimbabwe. Helmsing, A.H.J. 1991. Non-agricultural enterprises in the communal lands of Zimbabwe. In: Mutizwa-Mangiza, ND. and A.H.J. Helmsing (Eds.), Rural development and planning in Zimbabwe. Avesbury, Aldershot, UK. IDS Workshop. 1989. Farmer’s knowledge, innovations, and relation to science. In: Chambers, RA Pacey and LA. Thrupp (Eds), Farmer first: Farmer innovation and agricultural research. Intermediate Technology Publications. London, UK 154 Jackson, J.C. and P. Collier. 1991. Incomes, poverty and food security in the communal lands of Zimbabwe. In: Mutizwa-Mangiza, ND. and A.H.J. Helmsing (Eds.) Rural development and planning in Zimbabwe. Avesbury, Aldershot, UK Jayne, TS. and M. Chisvo. 1991. Unravelling Zimbabwe’s food insecurity paradox. Implications for grain market reform in Southern Africa. Food Policy, August, 319-329. Johnson, P.H.S. 1992. Why farmers grow maize in Natural Region V of Zimbabwe. In: Proceedings of the second annual scientific conference of the SADCC-Land 8: Water Management Research Programme. Mbabane, Swaziland, October 7-9, 155-163. Khon Kaen University. 1987. Proceedings of the 1985 International Conference on Rapid Rural Appraisal. Khon Kaen, Thailand, September 2-5, 1985. Lan, D. 1985. Guns & rain. Guerrillas & Spirit Mediums in Zimbabwe. James Currey Ltd, Islington, UK Mascarenhas, J.P. Shah, S. Joseph, R. Jayakaran, J. Devararam, V. Ramachandran, A. Fernandez, R. Chambers and J. Pretty. 1991. Participatory Rural Appraisal. Proceedings of the February 1991 Bangalore PRA Trainers Workshop, RRA Notes, No. 13. lntemational Institute for Environment and Development (IIED), London, England. Mehretu, A. and C. Mutambirwa. 1992. Gender differences in time and energy costs of distance for regular domestic chores in rural Zimbabwe: A case study from Chiduku Communal Area. World Development. 20, 1675-1683. Ministry of Lands, Agriculture and Rural Resettlement (MLARR), Farm Management Section of the Economics and Markets Branch. 1990. In: The second annual report of farm management data for communal area farm units 1989/90 farming season. MLARR, Harare, Zimbabwe. Ministry of Lands, Agriculture and Rural Resettlement (MLARR), Farm Management Section of the Economics and Markets Branch. 1989. The first annual report of farm management data for communal area farm units 1988/89 farming season. MLARR, Harare, Zimbabwe. Mitchell, RC. and Carson, LT. 1989. Using surveys to value public goods: The contingent valuation method. Resources for the Future, Washington, 0.0., USA. Mittra, MK and PC Stickler. 1961. Excess water effects on crops. Transactions of the Kansas Academy of Science, 64, 275-286. 155 Mudimu, G.G. Mbwanda, S. Chigume and J. Govere. 1988. Household food insecurity in low rainfall areas of Zimbabwe: Initial findings in Mudzi, Mutoko and Buhera communal areas. In: Southern Africa: Food Security Policy Options. Harare, Zimbabwe, 1-5 November, 1987. 217-233. Phia, MI. 1992. Efficient fertilizer and water use under variable rainfall conditions. In: Proceedings of the second annual scientific conference of the SADCC-Land & Water Management Research Programme, Mbabane, Swaziland, October 7-9, 241-257. Redclift, M. 1987. Sustainable development. Exploring the contradictions. Routledge, London, UK Reh, l., Van Oosterhout, S.A.M., Campbell, EM. and R.F. Du Toit. 1989. Resource use and agricultural production. 15-33 In: B.M. Campbell, R.F. Du toit and CAM. Attwell (Eds.), The Save study: Relationships between the environment and basic needs satisfaction in the Save Catchment, Zimbabwe. Supplement to Zambezia, University of Zimbabwe, Harare, Zimbabwe. Reynolds, P. 1991. Dance civet cat. Child labour in the Zambezi Valley. Boabab Books, Harare, Zimbabwe. Richards, P. 1980. Community environmental knowledge in African rural development. In: Brokensha, D.W., D.M. Warren and 0. Werner (Eds.), Indigenous knowledge systems and development. University Press of America, Lanham, MD, USA. Shumba, EM. 1990. Zimbabwe’s experience in agricultural research priority setting for communal area households. In: Food security policies in the SADCC region. Proceedings of the sixth annual conference of food security research in Southern Africa. Harare, Zimbabwe, November, 1990. Shumba, EM. 1988. Maize technology research in Mangwende, a high potential communal area environment in Zmbabwe. Part 1: Developing a research agenda. Farming Systems Newsletter, 34, 12-34. Shumba, EM. 1985. Application of the farming systems research approach in Mangwende communal area in the Murewe District. Zmbabwe Agricultural Journal, 82, 31 -35. Shumba, E.M. 1986a. A comparison of maize and groundnut husbandry practices under communal area production. Zmbabwe Agricultural Journal, 83, 137-147. 156 Shumba, E.M. 1986b. Farmer maize production practices in a high potential communal area environment: Implications for research. Zimbabwe Agricultural Journal, 83, 175-179. Shumba, E.M. 1984a. Reduced tillage in the communal areas. Zimbabwe Agricultural Journal, 81, 235-239. Shumba, E.M. 1984b. Yields of maize in the semi-arid regions of Zimbabwe. Zimbabwe Agricultural Journal, 81, 91 -94. Stanning, J.L. 1987. Household grain storage and marketing decisions in surplus and deficit communal farming areas in Zimbabwe: Preliminary findings. Working Paper AEE 1/ 87. Department of Agricultural Economics and Extension, Harare, Zimbabwe. Stanning, J.L 1987. Policy implications of household grain marketing and storage decisions in Zimbabwe. In: Southern Africa: Food Security Policy Options, Harare, Zimbabwe, 1-5 November, 1987. 329-358. Stanning, J.L 1989. Smallholder maize production and sales in Zimbabwe: Some distributional aspects, Working Paper AEE 2/89. Department of Agricultural Economics and Extension, University of Zimbabwe, Harare, Zimbabwe. Swift, J. 1979. Notes on traditional knowledge, modern knowledge, and rural development. IDS Bulletin, 10, 41 -43. Tattersfield, JR. 1982. The role of research in increasing food crop potential in Zmbabwe. Zimbabwe Science News, 16, 610. Thompson, J.G. and W.D. Purves. 1978. A guide to the soils of Rhodesia. Rhodesian Agricultural Journal, Technical handbook No. 3. Thompson, JG. 1965. The soils of Rhodesia and their classification, Department of Research and Specialist Services, Harare, Zimbabwe. Tversky, A. and D. Kahneman. 1981. The framing of decisions and the psychology of choice. Science, 211, 453-458. Tversky, A. and D. Kahneman. 1974. Judgement and uncertainty: Heuristics and biases. Science, 185, 1124-1131. Warren, D.M. 1991. In: Using indigenous knowledge in agricultural development, World Bank Discussion Papers, No. 127. The World Bank, Washington, DC. 157 Whingwiri, E.E. and G. Harahwa. 1985. Maize varieties for the low yielding environments in Zimbabwe. Zmbabwe Agricultural Journal, 82, 29-30. Whitlow, JR 1980. Agricultural potential in Zimbabwe. A factorized survey. Zimbabwe Agricultural Journal, 77, 97-106. Zinyama, LM., D.J. Campbell and T. Matiza. 1987. Traditional household strategies to cope with food insecurity in the SADCC region. In: Southern Africa: Food Security Policy Options, Harare, Zimbabwe, 1-5 November, 1987. 183-205. 158 Chapter 5. Simulatlon model: development, structure and testlng INTRODUCTION 'All models are wrong, but some are useful.’ (Box, 1979). To make meaningful evaluations of the performance of an agroecosystem using performance criteria such as stability, sustainability, resilience, efficiency or productivity, we need a thorough understanding of the structure and functioning of the agroecosystem and the ability to predict agroecosystem behavior. In the case of Masoka, and perhaps most communal area farming systems in Zimbabwe, neither this level of understanding nor these predictive capabilities are, as yet, well developed. The development and implementation of models can be a useful method of enhancing our understanding of the structure and functioning of complex systems (Shannon, 1975; Dent and Anderson, 1971; Wright, 1971). They may also provide hypotheses with which to test our understanding, both of the structure and of the functioning of the agroecosystem. At this early stage of our investigations, our objective is as full an understanding as is possible. We expect therefore, a model of the Masoka agroecosystem to mirror the complexity of the real world system (Wright, 1971). The simulation of entire farming systems or agroecosystems is a complex and difficult task. Difficulties arise from a) the lack of suitable biological data; b) uncertainty in the input data and in the state of the system at any time; c) the problem of obtaining the correct balance of detail in the various components of the model; d) the propensity of farming systems to change; and e) uncertainty in the influence of the controls farmers may use to achieve their objectives (Halfon, 1979; Dent and Anderson, 1971; Wright, 1971 ). The difficulty of the task is reflected in the literature where no published accounts could be found of farming systems simulation models, that simultaneously simulate multiple households producing multiple crops. Whilst there are several examples of conceptual models of agroecosystems in Zimbabwe (Swift et al., 1989; FSRU, 1985), as elsewhere (Fresco, 1986), none of these models have been implemented as simulation or analytical models. Where peasant farming systems have been simulated the models are of only a single household or farm unit and most often use some form of mathematical programming (Berdegue et al., 1989; Maino ef al., 1993; Ngambeki et al., 1992) or a mixture of dynamic models and mathematical programming models (Crawford, 1982). Dynamic models of single farms or farm enterprises have been used in commercial agricultural analysis (Deybe and Flichman, 1991; Ungar, 1990) 159 160 but here again mathematical programming models dominate (Kingwell et al., 1992; Kingwell and Pannell, 1987). When dealing with regional studies that require information on micro-scale impacts Klein et al. (1989) used individual farm dynamic simulation models to generate activities that were then fed into a regional input- output model. Deybe and Flichman (1991) used a dynamic model (EPIC) to generate activities that were subsequently incorporated into a linear programming model for the region. Owsinski (1982) used a suite of linear programming models for agricultural development planning in Poland. Despite their widespread use, mathematical programming models are far from ideal for modeling agroecosystems because they have only limited capabilities for simulating temporal change in state variables and of incorporating the effects of feedback and feed- forward processes on production (Dent and Anderson, 1971). Missing from previous modeling studies of peasant farming systems is the ability to examine the effects of household interactions on household needs satisfaction and on the biophysical resources of the agroecosystem. In virtually all of these studies, the independence of households is implicitly assumed. In Zimbabwean communal area farming systems however, inter-household sharing of equipment and labor is important (Steinfeld, 1988; FSRU, 1985). Equally important is the inter-household provision of support in drought years (Zinyama et al., 1988). Commonly, analysts simplify the task of modeling farming systems by assuming households to be homogenous. They are then justified in using 161 aggregated data (means) to predict household performance. The heterogeneity of Zimbabwean communal area households has however, been well documented (Steinfeld, 1988; Shumba, 1988; FSRU, 1985). Debrah and Hall (1989) clearly show the dangers of using aggregated data for the development of farm plans; aggregated data grossly underestimated the income variability faced by farmers when selecting a farm plan. Individual farm yields were between two and a half and eight times more variable than aggregated yields. Thus, whilst aggregated analyses will be useful for some purposes they are unlikely to be as productive, in terms of increasing our understanding of the Masoka agroecosystem, as analyses that allow us to examine the behavior of individual households interacting with other households. Models may take many forms. There is a considerable body of literature concerned with their classification (Naylor et al., 1966; Shannon, 1975; Steiner, 1989; Vemuri, 1978). When dealing with complex and large scale systems simulation models are frequently identified as the most suitable approach to anaan the system of interest (Anderson, 1974; Manetsch et al., 1971; Naylor et al., 1966; Vemuri, 1978). There is no generally accepted definition of simulation (Pritsker, 1979). For the purposes of this study simulation is defined as "the process of designing a model of a real system and conducting experiments with this model for the purpose of either understanding the behavior of the system or of evaluating various strategies (within the limits imposed by a criterion or set of criteria) for the 162 operation of the system" (Shannon, 1975). Models represent a synthesis of our understanding ofthe system of interest. The ultimate test of this understanding is our ability to accurately predict the behavior of the system, given any set of initial conditions and system inputs. This is typically called model validation. The terms verification and validation are however, often used synonymously. In this analysis, verification is used to mean the process of determining that the simulation model is working as intended whilst validation is the process of determining whether the model is an adequate representation of the real world system of interest (Sargent, 1982). The problem of model validation is complex and has not been resolved (Balci and Sargent, 1984; Naylor at al., 1966). Much of the difficulty stems from the philosophical problems of proving that a hypothesis is true (Giere, 1991; Naylor at al., 1966; Popper, 1968) but also from the objectives for which the model is designed. Officer and Dillon (1968) argue that results based on classical significance tests have no relevance to real world managerial decision making because they are “absolutely devoid of any economic appraisal of real-world consequences of alternative decisions". Mankin et al. (1975) suggest that, by definition, all models are invalid (model behavior does not correspond to real system behavior under all conditions of interest) and we should focus on questions of usefulness, especially when the models are designed to further understanding. These authors define a useful model as one that accurately represents some of the system behavior under consideration and claim it is useless if it does not. In the 163 business literature, the issue of validity is extended to include credibility, or the extent to which a simulation model and its results are accepted by the user and are used to assist in making decisions (Law and Kelton, 1991). Law and Kelton (1991) describe a comprehensive approach to validation based on the three stage approach of Naylor and Finger (1967). In this Chapter I describe a dynamic stochastic simulation model of the Masoka agroecosystem. The model was developed to improve our understanding of the structure and behavior of the agroecosystem and also as part of a developing methodology for the analysis of complex agroecosytems. More specifically, the model was designed to enable me to examine the interactions among agroecosystem components and in so doing identify those which significantly contributed to a household being able to satisfy needs. It was hoped that, once fully validated, the model might assist in evaluating policy and household decision options from both a household and a community perspective. MODEL DESCRIPTION Overall approach There are an infinite number of possible models of the Masoka agroecosystem. The model that was developed was the result of attempting to satisfy a changing set of modeling objectives within the constraints imposed by the abilities and world view of the modeler as well as by the available technology. In this section, those aspects of the modeler's world view that were important 164 influences on development of the model are briefly discussed. The model was designed to be used in Zimbabwe and therefore, was designed for use on a personal computer. The temporal and spatial scale at which the system was modeled was expected to have a significant impact on the results that were produced and the interpretations that could be made (Firbank, 1993; Meentemeyer and Box, 1987; \Megart, 1988; Mans 1989; Wiens et al., 1986). The primary scale of interest was the scale at which Masoka residents manage their resources. Essentially, this was the scale of the field, usually between a half acre and one acre. Our hope was to model the Masoka agroecosystem at three scales. Whilst the current model went some way towards meeting this objective, the limited resolution within the household, as well as within seasons, is less than what was originally desired. Complexity and uncertainty were considered important properties of the Masoka agroecosystem. A useful and valid representation of the Masoka agroecosystem was expected to incorporate these characteristics. Complexity was defined as the property of having interacting components with non-trivial detail at several scales. Uncertainty was defined as the inability to know or predict the outcome of some event or action, with accuracy, before the event had occurred (Anderson, 1991). Spatial heterogeneity and spatial interactions among agroecosystem components were believed likely to be important determinants of agroecosystem behavior (Baudry, 1993; Sauget and Balent, 1993). The model was therefore, 165 designed so that these agroecosystem attributes could be included where appropriate. Spatial interactions were not however, included in this version of the model. 1 As discussed in Chapter 4, knowledge of the goals, objectives and experiences of Masoka residents were essential building blocks for our understanding of the agroecosystem. The model was, therefore, developed to incorporate both local farmer knowledge as well as formal scientific knowledge. The models presented in this chapter were the result of numerous false starts, the exploration of several paths of investigation and multiple iterations of model development. These were essential precursors to the models presented in this chapter. Not all of the capabilities that were originally planned were incorporated into the model but where these were considered important, the model was designed so that these aspects could be added in the future. Model overview The Zambezi Valley Agroecosystem Model, (ZVAM) comprised three main sub-models (Figure 5.1): An initializing model (INITIAL), a rainfall generating model (RAINGEN) and an agroecosystem simulation model (MASOKA1). Both INITIAL and MASOKA1 used ASCII data layers (maps) developed using a geographical information system (GIS). GIS data layers were in raster format with cells (pixels) of 64m. INITIAL was used to generate a number of households, each with a 166 A ”Fan—#30 —' .._2<>N. _0005. E0.0>00000.m< >0=0> _N0nE0N 05 .0 0.:0can00 .90.). Om 0.3.”. 0.00 0.9.003... 0.00 0.9.0030... 22.0. 0.00 0.0 0.0.0. 800 0.6 JMDOE .5002 F<¥Ow<2 l...W0\ 02_.w..._wm.._ I‘ m \ 00000.0 0.00» .0 50:5... p.520 "00505.0 0.2.0030... n..<_._._Z. ...__»<0o._.._< _F / \g. mcoVw\ — wOéOhw 50.562205. SO.>C§EDw sown—Aw; 50.5008 . SAVEDSS SOdJImeOI \0§.0So \GENHDO \08 , _ 5.50 l ///// i “i 111 \i . E50 AEE; 2_.mn.0m _.ZO_.wO _.mOmm... , , 02.4.00 ‘ F<2llllo0<2 \ tlv. 02..m.:mm... l‘ is 178 C START D Initialization MONTE CARLO LOOP YEARLY LOOP CELL LOOP Allocate crops l Generate yield Update cell and household states Estimate erosion Statistics Household loop Estimate labor needs ' m '1 Correct yields > Budget calculations Balance needs Statistics I 1" Statistics V In L V Figure 5.4. Sequence of major components of MASOKA1. END 179 Production greater than needs? Yes—D Update stores Store greater Update deficit and Yes—b than deficit? store No . Update deficit and sufficuent to safisly Yes—D stores T Household classed as deficit Retum to main program Figure 5.5. Flow chart illustrating sequence of events in STORAGE sub-system. 180 prices used in the model (Figure 5.5). For most simulations these were the local, Masoka cash prices. Production in excess of needs was placed in storage, from which there was no loss. If household needs could not be met from production an attempt was made to meet them from storage (including using cash stores to purchase needs). If this was not possible the household was considered deficit for that particular need. The ALLOCATIVE subsystem The ALLOCATIVE subosystem allocated each household to a production and to a tillage strategy. Fixed proportions of households were allocated to each strategy at the beginning of each simulation. Households maintained the same strategy for the duration of the simulation. The production strategy governed what crops a household grew. the proportion of cells that the household planted to each crop and household fallowing practices. Tillage strategies governed whether a household used hand or tractor cultivation (tractor cultivation was only used on cotton and had no impact on yields and whether a household planted with, or at an angle to, the contour. These factors were important only in their effect on erosion. Four production strategies could be used in any simulation. These strategies differed in the parameters that defined: a) the proportion of fields households planted to each crop; b) the number of years a household cultivated a field before leaving it fallow; and c) the proportion of fallowable fields households 181 C mm D Set parameters Randomly generate test for whether or not b have hum" For each strategy and ........................ 99'!!YR°.--.-------------- 3 '1 eady for fallowin I and not currently You Leave fallow? Yes—s» : fallow? No M! v 3 Call yield I subroutine. . Plant sorghum? Yes—u Update household I sorghum store. I No : can yield . . subroutine. I Plant maize? Yes—h” I t 9 household . maize store. I No I Plant maize. - . Call yield I Fugdsozgfgrct’ent No—usubroutine. ' ' Update household I maize store. : Yes i ' Plant cotton I Call yield subroutine Update cell n ' Update household state I cash store. Retum to main program Figure 5.6. Flowchart of activities in the ALLOCATIVE sub-system. 182 actually left fallow each year (Figure 5.6). These parameters were important control variables that could be set for any simulation experiment. Households planted a fixed proportion of their fields to each crop in each year. These proportions were fixed for each production strategy. Whilst not a realistic representation of household decision making the use of fixed proportions simplified model development. The model was designed so that more complex cropping decision patterns could be incorporated at a later date. For sensitivity analyses only two strategies were used. income maximizers (INCMAX) and needs satisfaction variance minimizers (MINVAR). INCMAX households produced a higher proportion of cotton (50% of cells on all soil types), planted little sorghum (15% of jecha cells) and used shorter fallow periods (minimum cropping period four years. maximum of 10 years) than other households. MlNVAR households planted mostly maize (50% of bepe and river jecha and 35% of jecha cells) with a little cotton (50% of bepe and river jecha cells) and sorghum (35% of jecha cells) and used longer fallow periods (minimum cropping period of four years and a maximum of seven years). The decision whether or not to leave a particular field fallow was only pertinent on jecha soils; neither bepe nor riverland soils were left fallow. The decision to leave a field fallow was a probability based on the household production strategy and the number of years the field (cell) had been cultivated. In Figure 5.7 the cumulative probability curves of a household leaving a jecha field fallow are shown for each production strategy. An income maximizing (MAXINC) 183 household. for example, would not leave a field fallow before the field had been cultivated for at least four years. Thereafter, the probability of leaving a field fallow increased with increasing cultivation until, after ten years of cultivation the probability of leaving the field fallow was one. Other production strategies used different cumulative probability curves reflecting what i believed were reasoned approximations of some likely approaches to fallowing (Figure 5.7). The LABOR sub-system In the LABOR sub-system available household labor for each month of the growing season was estimated by multiplying the number of adults by 25 working days in the month. Based on Reynold’s data (Reynolds, 1991) 44% of children were considered to be in the age group that contributes to the labor force (10 to 20 years of age). Children were considered to provide labor equivalent to 42% of an adult. This value was a mean of the proportions given by Reynolds (1991) for clearing, planting, re-planting and weeding. Once a household's fields had been planted to a crop the LABOR sub— system identified the labor requirements (person days) in each month of thegrowing season (October to March) based on the VR labor budgets presented in Chapter 4. These requirements were compared to household labor availability. if households were labor deficit and they had cash or grain balances they could search for available labor from other households (Figure 5.8). Households that had labor greater than their needs and were deficit in cash or maize could supply labor 184 Cumulatlve probabllltg (300000000 01 l 0 I r r l 0 2 4 6 8 10 12 Years oi culttvatlon + MRXINC + MINURR + SRTIS —+— GREEN Figure 5.7. The cumulative probability of leaving a jecha field fallow for households using income maximizing (MAXINC), variance minimizing (MINVAR), needs satisfying (SATIS) and environmentally protective (GREEN) production strategies. 185 C sum D Estimate household labor needs and labor availability Calculate employment potential. Search all households for available labor until Sufficient deficit = 0. cash to hire all Yes—t Update employee and employer cash balances and labor stocks. Search all households for available labor until deficit = O. 03:33:: 3:; to Yes-h First use cash and then deficit? grain for payments. Update cash and grain balances and labor No i Search allhouseho for available labor until cash and grain stores are empty. , < Return to main > First use cash and then program grain for payments. Update cash and grain balances and labor Figure 5.8. Flowchart of major activities in the EMPLOY component of the LABOR sub-system. 186 to households that were labor deficit. Labor was hired from other households at the local wage rate (232.54 day") in cash or grain equivalent calculated using the Masoka prices for maize. Sorghum was not used to pay for hiring labor. When labor was employed the labor balances of both the employing and the employed households were updated. if households were still labor deficit, either through there being no available labor or through not having the resources to hire labor then the yields of each crop were adjusted. Where labor deficits occurred the resulting yield reductions were first allocated to sorghum, then to cotton and finally to maize (Figure 5.9). Thus, if the total labor deficit was less than the total labor required for sorghum then only the sorghum yields were corrected. If the total labor deficit was greater than the labor required for sorghum then as much of the deficit as possible was used on sorghum before yield reductions were made on cotton. The reduction in yield due to labor deficits was calculated as the product of the proportion of labor needs that were deficit times the yield reduction for that time period expressed as a proportion. deficit in month 'm'? Yes Establish month(s) of labor defict(s) and call appropriate subroutine Reduce total sorghum production. Sorghum grown? Yes—h Update stores and labor deficit. No Reduce total cotton production. Cotton grown? wk. Update stores and labor deficit. 2 bor deficit = 0? No Reduce total maize . production. No labor deficit. Update frequency of labor defict households in month. Return to main program Yes-D YOG-D Figure 5.9. Flowchart of major activities in yield correction component of the LABOR sub-system. 188 Final yield (Y,) was expressed as: Y, = YI'W'*LD*YR) Where Yi = the initial, uncorrected yield. LD = the labor deficit expressed as a proportion of the required labor. YR = the yield reduction factor expressed as a proportion of the uncorrected yields. For sorghum the yield reduction factors were based on the results of Burnside and WleS (1967). Sorghum was assumed to be planted in mid- November. A labor deficit in December led to a 12% yield reduction. A deficit in December and also in January resulted in a 36% yield reduction. A labor deficit in January (but not in December) led to a 3% yield reduction. Yield reduction factors for cotton were derived from Buchanan and Burns (1970) and Schwerzel and Thomas (1971). Cotton was assumed to be planted at the beginning of December. A labor deficit in December reduced yields by 13%. A labor deficit in December and in January resulted in a 54% yield reduction. Labor deficits in December, January and February reduced yields by 100% as did labor deficits in December, January, February and March. A labor deficit in January (but not in December) reduced yields by 56%. Weld reduction factors for maize were derived from Hall et al. (1992). Maize 189 was assumed to be planted in November after the first rains. A labor deficit in November had no effect on yields. A labor deficit in December reduced yields by 10%. A labor deficit in December and January reduced yields by 35%. If there was also a deficit in February then yields were reduced 45%. A deficit in January (but not in December) reduced yields by 25% whilst a deficit in February (but not in December or January) reduced yields by 10%. The CROP sub-system Yields for three crops, (maize. cotton and sorghum) were generated in MASOKA1. Households planted their fields (cells) to one of these crops (or left the cell fallow) each year (Figure 5.10). Yields. for each of these three crops, were a function of soil type, time since fallow (on jecha soils only) and rainfall season. Yield functions were derived from the yield possibility diagrams developed, by the VHS, for each crop, soil and season combination. To derive a rainfall season from the annual rainfall amount generated by RAINGEN membership functions for good, average and poor seasons were derived, using fuzzy set membership functions. Good. average and poor seasons were considered to be fuzzy sets. The membership functions described the relationship between annual rainfall amount and the degree of membership in each of these sets. Annual rainfall amounts had graded membership, from zero (not a member) to one (full member) in one or more of these sets. The data on Masoka 190 (mm) Determine degree of membership in season. Generate p value based on season membership Calculate yield as a function of p. Convert from per hectare to cell value. Update household production store and counter of cell cultivation history. Retum to main program Figure 5.10. Flowchart of major activities in CROP sub- system. 191 resident classification of each season (Chapter 4. Table 4.2) were used to derive the membership functions for good and poor season. The proportion of respondents allocating each rainfall year to a particular season class (good. average or poor. Table 4.2) were plotted against Angwa Bridge rainfall (Figure 5.11). Clear peaks were discemable for a good season (1000mm) and a poor season (500mm) but the trends for an average season were not at all clear. These peaks were used (to define the boundary of rainfall amounts that were definitely members of the set of either good or poor seasons. Thus. for any rainfall amountless than or equal to 500mm the membership of that rainfall amount in the set of poor seasons was set equal to one (ux poor greater than or equal to 1000mm the membership of that amount in the set of = 1). For any rainfall amount good seasons equalled one (uxgood = 1). For any rainfall amount greater than 500mm but less than 600mm the degree of membership of that rainfall amount in the set of poor seasons declined according to the membership function fitted to the curve shown in Figure 5.12. Initial tests of model yield predictions found good season yields to be consistently more than predicted yields when comparing the simulated means with those derived from the VR distributions. To reduce this deviation from the expected means the value at which uxgood = 1 was increased to 1124mm. With this correction simulated mean and median yields were closer to the expected values. For any rainfall amount less than 1124mm but greater than 767mm the degree of membership in the set of good seasons declined according to the function fitted to the good season curve shown in Figure 5.12. These curves 192 QDIS respond 0 C) C) ¢ 01 ox KKK Irratf /‘ oi . FD . (.0 xx. #/ M,” on Proport c3 400 600 800 1000 1200 140 500 700 900 1100 1300 Annual rainfall 1978/79 - 1991/92 + Good —B— Hveraqe + Poor Figure 5.11. The proportion of respondents allocating each growing season from 1978/79 to 19911/92 to good, average or poor seasons plotted against the rainfall for that season measured at Angwa Bridge. Rainfall data from the Department of Meteorological Services, Harare. 193 30:6 \ / \K 1... \/ \/ 25K K . . 400 600 800 1000 1200 500 700 900 1100 Ra1nial] (mm) + Good —5— Average + Door Figure 5.12. Membership functions for good. average and poor seasons used in NVUNDKA1. 194 describe the membership functions for good and poor seasons and were defined as follows: up”, = 5.592 - rain / 108.89 500 <= rain < 600 = 1 rain < 500 = 0 otherwise ugood = 1 rain >= 1124 = raln / 357.14 - 2.15 767 <= rain < 1124 = 0 otherwise Where rain = The annual rainfall in millimeters. Because there was no clear trend with which to describe the average season membership function the mean annual rainfall (725mm) was identified as having a membership of one in the set of average seasons. The membership function was then described so the lower and upper boundaries of annual rainfall events, that were members of set of average seasons, occurred plus or minus one standard deviation (247mm) from the mean (i.e. 478 and 972mm). The membership function for the set of average seasons was therefore: ummg. = rain / 247 - 1.935 478 < rain <= 725 3.935 - rain / 247 725 < rain <= 972 0 otherwise 195 The maximum membership value was used to class seasons when a rainfall amount had membership values greater than one in more than one season. Functions were fitted to each yield cumulative probability distribution generated by the VHS (Chapter 4). The general form of these functions were as follows: YIELDi-ae '“Pwe ‘d‘P' (5) Where YIELD = The yield in kg ha" i,j,k = The ith crop grown on the jth soil in the kth was season a.b.c,d = parameters fitted to the function e = the base of natural logarithms p = a pseudo random number between zero and one. To derive the specific yield for a cell the following procedure was used. The maximum seasonal membership value was identified and used to class the season as good, average or poor. A random number (r) in the range (0.1) was generated and compared to the membership value of that season. For a poor season the probability (p) value used in the yield function equalled the minimum of the random variable and the degree of membership of a poor season (MIN(r,upoo,)). For a good season, p = MAX(1- gown-r). These derivations may be more clearly 196 explained with reference to Figures 5.12 and 5.13. As the p value increases the generated yield decreases (Figure 5.13). In a poor season, as rainfall increased the membership of that season decreased (Figure 5.12) and, logically. we expected higher yields. Thus, by using the minimum of the membership value and r we generated yields whose lower bound was the yield achievable for that membership value and whose upper bound was the maximum yield achievable for that season. In Figure 5.13. for example, if the membership of the poor season was 0.4 we expected yields of between 2000 and 2900 kg ha". In a good season the reverse was true. As rainfall increased. ugood increased and yields were expected to increase to the maximum possible for that season. In this case, however, p needed, to be smaller as ugood got bigger. I, therefore, took the maximum of one minus the membership value or one minus r. In this case, and looking at Figure 5.13. if ugood = 0.4 then expected yields would fall in the range 1100 to 1700 kg ha". The logic behind the average season yield derivation is a little more complex. With rainfall less than 725mm, um”ago increased with increasing rainfall but then decreased again as rainfall increased beyond 725mm. Yields were, however. expected to increase with increasing rainfall. This problem was dealt with in the following way. Average season membership values were greater than the poor and good season membership values between annual rainfall amounts of 568 and 873mm. A linear regression model (I2 = 0.9985. 12 DF) was developed to describe the relationship between rainfall and k, where R was the membership value of a dummy average season. Membership of this season was governed by 197 3000 R 2800 \ 2600 \ '8 2400 _C \ \ 2200 2000 “F 1800 -\ 1600 \\ 1400 1200 \\l 1000 (kg Yleld 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 p Figure 5.13. Example yield generation function from MASOKA1. Dependent variable is yield (kg ha ) and independent variable is the random variate p. 198 the equation (k = -1.70476 + 0.03069 * rainfall) between rainfall amounts of 550 and 875mm and was zero otherwise. After determining the R value for the average season, the procedure for deriving a p value was much like that of deriving the p value for a good season but with p = MAX(1-k,1-r). For crops grown on jecha soils, soil fertility had an important effect on yields. If a soil had been previously cultivated, yields were derived using the following form: YIELD = 5YR_YLD + (VIRG_YLD - 5YR_YLD) I YRSCLT Where 5YR_YLD = The yield from a soil continuously cultivated for five years derived from equation 1, above; VIRG_YLD = The yield from a virgin soil derived from equation 1, above; YRSCLT = The number of years the soil had been previously cultivated. The EROSION sub-system The Soil Loss Estimation Model for Southern Africa (SLEMSA) developed by Elwell (1980) was used in ZVAM. SLEMSA calculates erosion (Z), in t ha'1 as a function of three sub-models yielding K. C and X: t Where 2 199 K*C*X Predicted mean annual soil loss. t ha'1 yr‘1 from the land under evaluation; Mean annual soil loss (t ha'1 yr") from a standard tilled field plot 30m * 10m at 4.5% slope, for a soil of known erodibility (F) under a weed free, bare fallow; The ratio of soil loss from a cropped plot to that from a bare fallow; the ratio of soil loss from a plot of length L meters and slope percent (S), to that lost from the standard plot. Elwell (1980) identified five control variables which were used in SLEMSA: rainfall energy (E), soil erodibility (F), percent energy intercepted by the crop G), slope percent (8) and slope length (L). Rainfall energy (E) was calculated for non-guti14 areas using: E = 18.846 * rainfall Where Rainfall = the annual rainfall in millimeters. ‘4 Gutl is a light rainfall that occurs in parts of Zimbabwe but not In the study area. 200 The erodibility of soils in the study area was determined by using a rainfall simulation method described in Elwell (1986). The erodibility factor is an estimate of the soil loss from a bare soil at 4.5% slope. Soil samples of approximately 300kg were collected from virgin and from cultivated sites representative of each soil type in Masoka. Each soil sample was collected from the top 10 to 15 cm of soil in a contiguous area or plot. Based on these simulations Elwell (personal communication) estimated the Fb values used in the model (Table 5.1). Table 5.1 Erodibility values for cultivated and uncultivated jecha and bepe soils in Masoka. Source Elwell, personal communication. Soil Fb Cultivated Uncultlvated Jecha 2 Bepe Crop cover curves were based on measured data provided by Elwell (personal communication) for maize. cotton and sorghum. For maize the minimum cover values were doubled to reflect the under sowing with cucubits that was common practice in Masoka. The following cover equations were used: MC 48.02 + 0.00608 * malze_yleld CC = 46.97 + 0.0043 * cotton_yleld 201 SC 69.1 + 0.005366 * sorghum_yleld Where MC, CC and SC are maize, cotton and sorghum cover values (%) and the yield values are expressed in kg ha". From these crop cover curves the soil loss ratio (C) was estimated using the following equations (Elwell. 1980): c = e'°-°‘*' i < 50% = (2.3 - 0.01*l) / 30 50% <= i Where i = the crop cover value (MC. CC or SC) indicating the percentage energy intercepted. Slope values were determined for each cell using the lDRlSl GIS system. Ten meter contours, over the entire study area, were digitized into IDRISI. This vector file was converted into a raster file (with pixels 64 * 64 meters) and the lDRlSl lntercon routine used to create a digital elevation model (DEM). A mean (low pass) filter was used to smooth the image. The SURFACE routine in lDRlSl was then used to calculate the slope of each 64 meter cell. Slope length was considered to be the length of one side of a cell (64m). The tillage practices of concern in Masoka were hand cultivation and tractor ploughing. Elwell (1980) provided data used to update the Fb values based on tillage type and planting and ploughing directions. For the EROSION sub-system 202 the soil loss from the previous year factor, the planting and ploughing direction factors, the land rough-ploughed only factor and an estimate of hand cultivation factor were used. According to Elwell (personal communication) hand cultivation (planting of seeds into holes made with a hand hoe) would increase the soil erodibility compared to a soil with a fine powdery tilth. The hand cultivation factor was therefore set to -0.75. From these values the F factor is estimated as: F = Fb + tf + ppf + prvf Where ti = the tillage factor set = 1 if tractor ploughed and to -0.75 if hand cultivated; ppf = the ploughing and planting factor, set to 0 if planting and ploughing carried out on or level to the contour, set to -0.25 if these operations at an acute angle to the contour and set to -0.5 if these operations at right angles to the contour; prvf = the previous years erosion factor, set to 0 if the previous years erosion was < 10 t ha", set to -0.5 if the previous years erosion was between 10 and 20 t ha'1 and set to -1.0 if the previous years erosion was greater than 20 t ha”. 203 The K value in the EROSION sub-system was thereafter estimated using (Elwell, 1980): K = o((0.4631 + 0.7663*F) * In (E) + 2.884 - 3.1209 . 5 Where F = The soil erodibility factor defined above. E = The rainfall energy defined above. The final factor, X, was estimated using the following equation (Elwell. 1980): x = L” * (0.76 + 0.53s + 0.07682) / 25.65 s > 4% = s*i.°-5/ (10.7425 + 0.030) 1% <= 3 Where L = The slope length in meters: S = ‘ The slope angle expressed as a percent. Erosion was estimated for each cell in the Masoka landscape and then summed to give a total landscape erosion (t ha"), a mean and standard deviation (t ha“) as well as the mean, and standard deviation, per hectare of cultivated land for each household. The sequence of activities in the EROSION model are shown in Figure 5.14. Estimated soil loss did not affect soil properties or yields. Yes Estimate Fb “Now—h Yes Yes i ii Estimate cover as a function of yield Generate cover as random variable Read cover from GIS file Estimate tillage practice effect and previous erosion effect l Estimate F Estimate K Estimate % rainfall energy intercepted (C) Estimate X Estimate erosion (Z) Z a K * c ‘ X Convert to cellular erosion Retum to main program D Figure 5.14. Flowchart of major activities in EROSION sub-system. 205 The ECONOMIC sub-system I assumed that households were able to sell all of the cotton they produced, to the Cotton Marketing Board (CMB), as well as all of the maize and sorghum they wished to sell. Maize and sorghum were only sold within Masoka and were traded at the 1990/91 season prices used in Masoka: Z$0.78 kg'1 for maize and Z$0.55 kg'1 for sorghum. Labor was valued at the local (Masoka) daily wage which varied between Z$2.00 and Z$3.44 per day. The average value (Z$2.54 day") was used as a constant in the model. Hired labor could be paid in cash or the grain equivalent of the daily wage. All households were assumed to face the same costs except INCMAX households that hired a tractor for ploughing had to pay the going rate of Z$148.00 ha'1 for ploughing. Households growing cotton were assumed to use the cotton packs supplied by Agritex which included all fertilizer and chemicals for 0.4096 ha (one acre). In allocating crops to a cell (ALLOCATE sub- system) a test was made to determine if the household had sufficient cash or maize reserves to pay for the basic costs of producing one cell (0.4096ha) of cotton. This value was set at Z$155.0015. Only the variable costs were used in calculating this amount and it excluded weeding and ploughing costs. A zero / one switch could be set in the model that would disallow / allow households to obtain loans for this amount and up to a maximum number of cells. The number of cells for which loans were obtained were recorded and later used to update the returns 15 This value was derived from the cotton partial budget presented in Chapter 4 (Table 4.13) and consists of seed (2330.88) + seed transport (232.47) + fertilizer and chemical costs (25345.80) multiplied by the hectare to cell conversion factor of 0.4096. 206 C enter 3 Set parameters Estimate costs for crop x using own labor Estimate costs for crop x using own labor and hired labor Estimate total returns Calculate returns to land, labor and investment Update household cash store Retum to main program Figure 5.11. Flowchart of major activities in ECONOMIC sub-system. 207 to cotton. No interest was charged on these loans. The general sequence of activities in the ECONOMIC sub-system are shown in Figure 5.15. A partial budget format similar to that of Chapter 4 was used to calculate the returns to land, to labor and to initial investment for each crop. Returns to labor were calculated as the returns to household labor and management. The OUTPUT sub-system The OUTPUT sub-system facilitated sub-sampling of individual households, yields, Fb values or erosion periodically during any simulation experiment. Thirteen output files were produced as required (Figure 5.3). Data for each household in each season were written to the HOUSEHLD.OUT file (T able 5.2). These data describe general household characteristics, household needs and household production. Data for all households or for a random selection of households could be written to this file. If the model was being run in monte carlo mode then the same set of randomly sampled households could be used in each monte carlo run. Mean erosion over the simulation period, for each cell in the landscape, was written to a GIS image file (T OTEROSJMG). The model could also write the mean erosion of any season to a GIS image file (EROSIONJMG). Yields, uncorrected for labor deficits, were written to file (YIELDS.OUT) for each cell in each season. Data written to this file included the year identifier, the season classifier, the soil type of the cell and the number of years the cell had been cultivated. A GIS data layer file (CROPSJMG) was generated for each season to show what crops had been grown in each cell in that season. Each cell in the landscape was assigned a crop identifier to indicate what crop had been grown in that season. If required, a GIS data layer (NEWFERTJMG) of the cropping history of the landscape (how many years each cell had been cultivated) could be written at the end of a simulation experiment. The labor requirements, labor available and labor surpluses or deficits for each month of the growing season could be written to a file (LABOUR.OUT) for each household in each season or for randomly sampled households in randomly sampled seasons. Also written to this file were each household’s production strategy class and the number of cells of each crop planted by the household. Returns to land, labor and to initial investment as well as information on the number of cells of each crop the household planted, the number of labor days hired out and hired in and the number of cells for which cotton loans were received were written to the file BUDGET.OUT for each household in each season. If required the data from a random sample of households could be written to the file. Two summary data files were generated (T able 5.3). In the first, SUMMARY.OUT, data averaged over all households were produced each season. These included season attributes as well as production, erosion and economic 209 Table 5.2 Summary of output variables written to household output file (HOUSEHLD.OUT). VARIABLE DESCRIPTION ADLTS Number of adults (is. 16yrs old or greater) in the household. CHLD Number of children (is. 16yrs old or greater) in the household. CELLS Number of nomriverland cells household owns. RIV Number of riverland cells household owns. PROD Household production strategy (MAXINC, MINVAR, SATIS, GREEN). TILL Household tillage strategy. CASHND Household annual cash needs (25). MZND Household annual maize needs (kg). SORND Household annual sorghum needs (kg). CASHBAL Household cash balance (23). MAIZEBAL Household maize store balance (kg). SORBAL Household sorghum store balance (kg). CASH Total household cash generated for the year (23). MAIZE Totalhouseholdmaize productionfortheyear (kg). SORGHUM Total household sorghum production for the year (kg). CASHDEF Surplus (deficit) of household cash production minus cash needs (23). MAIZEDEF Surplus (deficit) of household maize production minus maize needs (kg). SORDEF Surplus (deficit) of household sorghum production minus sorghum needs (kg). DAYEMPL Labor days hired in that season. DAYWORK Labor days hired out that season. LOANCEL Number of cells for which cotton loans received. TEROS Total erosion over household fields (t he“). CEROS Mean erosion (t ha") over household fields. 21 0 Table 5.3 Summary of output variables written to SUMMARY.OUT and to MONTECAR.OUT. VARIABLE m S NT WEAR Annual or seasonal idfiifier X‘— MONTE Monte-carlo run identifier X SEAS Season class (good, average. X poor) TOTEROS Average erosion over the X X landscape (t ha'1) ANCASHDEF Households that were cash deficit X X (96) ANMAIZEDEF Households that were maize X X deficit (%) ANSORDEF Households that were sorghum X X deficit (%) MZEHA, MZESD, COTHA, COTSD, Average and standard deviation of X X SORHA, SORSD uncorrected yields for maize, cotton and sorghum (kg ha") MZCELLS. COTCELLS, SORCELLS Total number of cells planted to X maize, to cotton and to sorghum OCTDEF, NOVDEF, DECDEF, Percentages of households with X JANDEF, FEBDEF, MARDEF, labour deficits in each month APRDEF (October through April) (%) MNMZREI', SDMZRET, MNCTRET, Mean and standard deviation of X SDCTRET, MNSORET, SDSORET total returns to maize, cotton and sorghum as household") MNMZACR, SDMZACR, MNCTACR, Mean and standard deviation of X SDCTACR, MNSORACR, per hectare returns to maize, SDSORACR cotton and sorghum (2S ha") MNMZLAB, SDMZLAB, MNCTLAB, Mean and standard deviation of X SDCTLAB, MNSORLAB, SDSORLAB returns to labor for maize, cotton and sorghum (25 ha“ 1) MNMZINV, SDMZINV, MNCTINV, Mean and standard deviation of X SDCTINV, MNSORINV, SDSORINV returns to initial investment for maize, cotton and sorghum (25 ha") MNEMPLDYS, MNWRKDYS Mean number of labor days hired X in and hired out per household MNCRCTMZ, SDCRCTMZ, Mean and standard deviation of X MNCRCTCT, SDCRCTCT, corrected maize, cotton and MNCRCTSR, socacrsn sorghum yields (kg ha") GOOD, AVERAGE, POOR Frequencies of good, average and X poor rainfall seasons INITFRT, ENDFRT, CHGFRT Initial, final and change in years X X cropped summed over all cells TOTFB Total Fb value summed over all X X cells 21 1 attributes. In the second summary file, MONTECAR.OUT, essentially the same data were presented except they were the averages over each year in a monte carlo run. Data that were in the SUMMARY.OUT file but not in the MONTECAR.OUT file were labor deficit data and household returns data. Data included in MONTECAR.OUT but not In SUMMARY.OUT were the frequencies of good, average and poor rainfall seasons. METHODS Verification, validation, usefulness All model sub-systems and code segments were checked several times for correct representation, logic, the accuracy of outputs, data correctness and the correct sequencing of events. Wherever possible, analytical checks were made of calculations made by the model. The maximum and minimum as well as mean and standard deviation of output data for each module were used to identify incorrectly functioning model components. For example, the ranges of possible yields derived from VR cumulative probability distributions were be used to check that simulated yields were within the correct range. Given the objective of an enhanced understanding of the Masoka agroecosystem, evaluation of model performance was more concerned with model usefulness than with the accuracy of predictions. The model was considered to be useful if it did any of the following: . ‘ ‘7“ ‘ "‘ - ._h.l 212 a) Demonstrated the utility (or lack thereof) of the approach to analyzing complex agroecosystems. Did the model provide insights that would not have been likely or possible with any other analytical approach; b) Helped identify gaps in our understanding of the Masoka agroecosystem; c) Enabled us to improve the management of the Masoka agroecosystem: or d) Enabled us to make comparisons of different system configurations in a design situation. I assumed, a priori, that the model was an inaccurate representation of the real world system. As far as was possible major components of the model were tested to establish the degree to which they accurately represented the real world processes of interest. This was difficult however, as there are no measured data on production, erosion or on household needs satisfaction. A major part of model validation that was not completed was the iterative evaluation of model results with Masoka VRs and leaders as well as with development professionals actively involved in the design, analysis and management of Zambezi Valley agroecosystems. The model presented here was not expected to withstand rigorous deductive testing. Rather, it was developed as a working hypothesis with which to focus discussion and further research. Given the complexity of the Masoka agroecosystem the model was neither complete nor sufficiently accurate to facilitate making reliable predictions of Masoka’s future. The model was expected to provide useful information on the direction and perhaps relative magnitudes of 01—-.- 213 changes in the agroecosystem. The model was also expected to be able to identify what known populations could be supported with the levels of technology and needs used in the model. Evaluation of the Masoka agroecosystem model broadly followed the procedures described by Law and Kelton (1991): Input data were checked for correctness and representativeness. The distributions of output variables generated by the model were tested for goodness of fit to the original distributions using Chi square, log-likelihood and Kolmogorov-Smimov tests. Model outputs were compared to observed system data where these existed or to theoretical distributions where appropriate. The total area cultivated each year in a simulations was found to fluctuate notably in the first few years of a simulation. For all simulations with MASOKA1 the data from an initial warm up period of 20 years were discarded. The length of this period was calculated following Welch (1983). This procedure requires estimating and plotting moving averages, with different window sizes of the variable of interest. The window size would be steadily increased until a relatively stable plot of the variable of interest against time was achieved. For these calculations the total number of cells cultivated in any year were used. To verify that the crop yield component of MASOKA1 generated yields that did not differ from expected yields (Table 4.9) the model was run for 50 years with 100 households on a landscape generated with INITIAL. In generating the initial conditions the land allocation multipliers were set to one. Yield data, uncorrected 214 for labor shortages, were collected for each call over the 30 year period starting from year 21. Mean rainfall over the 70 year period was 721mm with a standard deviation of 243mm. Sensitivity analysis Sensitivity analyses were carried out to identify the sensitivity of model response variables to changes in model parameters and inputs (factors). There were more than 100 factors and decision variables that could have affected model responses. Examining the effects of all of these was not possible. A 25 fractional factorial design16 was used with 13 factors (T able 5.4). Where appropriate I selected model parameter values so that the +1 level was the normal (standard model) mean plus one standard deviation and the -1 level was the mean minus one standard deviation. To facilitate the examination of a larger number of factors, some factors were selected to represent a group of factors. The assumption being made was that the interaction effects among these grouped factors would not cancel each other. Where grouped factors were found to significantly influence model response variables the effects of factors within the group could be evaluated at a later data. With INITIAL, two sets of initial conditions were generated. In the first 16 I am grateful to Professor D. Gilliland, of the Department of Statistics and Probability, Michigan State University, for his able assistance In generating, and analyzing the results of. this design. 215 Table 5.4 Factors and factor levels used in sensitivity analysis of MASOKA1. Factor +1 -1 Mean annual rainfall (mm) 972 478 Good season classification membership um s 1 um,“l a: 1 rainfall function rainfall >2 1124 >= 1000 Average season classification Mean _-l; 1*SD Mean 1 0.5*SD membership function Poorseasonclassification' membership u I,-1 u r=1 function reiniall’°<°- 500 raimair°<°= 442 Percentage of children that work 59 29 Loanordividend Loan-26155 Loan=2$0 Dividend a: 230 Dividend a 23155 Households could employ labor Yes No Maize price .00 23 kg" .26 23 kg" Household cash needs base+.41*base base-.41*base Proportion of households In production 80% MINVAR 20% MINVAR strategies 20% INCMAX 80% INCMAX Wage rate 233.07 day'1 Z3201 day'1 Land multiplier for riverland and non- 1.7 .63 riverland cells Minimum and maximum years jecha Min 0 Min 5‘ fields cropped before leaving fallow Max 10 Max 7 1. The model included an error for these analyses. For the income maximizing strategists the minimum number of years cultivated before a field was left fallow was sixandnotfive. (INIT63) the land multiplier was set to 0.63 for both riverland and non-riverland cells and the number of households was set to 100. The vegetation cover and cropping history maps produced in this run were used for all subsequent sensitivity simulation experiments. In the second run (INIT 17) the land multiplier was set to 1.7 for both riverland and non-riverland cells. The household data and field map files were saved for use in subsequent sensitivity simulation experiments. 216 Two rainfall files were generated. For the first, the rainfall mean was set to 972mm. Using the equation of Hutchinson and Unganai (1992) this mean yielded a k value of 14.41. For the second rainfall file the mean was set to 478 with a k value of 7.44. Two membership functions were used for classifying good season rainfall. In the first, the membership function described earlier in this chapter was used. In the second, the membership function was derived so that am = 1 when rainfall = 1000mm. Two membership functions were also used for each average and poor season classification. For average seasons the first function was that described earlier in this chapter. The second was fitted so that uavemge = 0 when rainfall was less than 600mm (725 - 0.5*247) and when rainfall was greater than 850mm (725 + 0.5*247). The poor season membership function described earlier in this chapter was used for the +1 level of this factor and for the -1 level the function was fitted so that upoor = 1 when rainfall = 600mm. Reynolds (1991, Table 1.1) provides data on the number of children in each of six classes (broken down by age and gender) from which the mean and standard deviation of the percentage of children that worked were estimated. Children in the 10 to 20 year-old class were considered likely to work. The mean dividend received by households over the four year period 1989 to 1992 was Z$150 (Chapter 2). I decided to examine the relative effects of these dividends compared to a loan of the same amount, but which was provided in the form of inputs to cotton production. The +1 level of this factor was a loan of 23155 (sufficient for one cell) with no dividend and the -1 level was a dividend of 26155 217 and no loan. The two levels of employment used were simply an on / off switch. At the +1 level households were able to employ labor and at the -1 level they could not. The maize price factor levels were set at the mean (Z$0.53 kg") plus one standard deviation (260.27 kg") and the mean minus one standard deviation of the government producer price (260.25 kg"), the consumer price of maize meal (260.55 kg") and the Masoka maize price (Z$0.78 kg") were used to calculate a mean and standard deviation maize price. The +1 level maize price was this mean plus one standard deviation (Z$0.80 kg") and the -1 level was the mean minus one standard deviation (230.26 kg"). To derive a standard deviation for household cash needs the data from the questionnaire survey described in Chapter 2 were used to develop a regression model of cotton needs (dependent variable) on family size (independent variable). The standard error of the coefficient on family size (0.413) was used to derive the range in household cash needs. The factor levels were set at the VR derived value plus or minus 41%. The VRs provided data on the number of households in each well-being class (Chapter 4). The proportion of households in the class of households who always satisfied needs (19%) were assumed to be income maximizers. The remaining households were considered to be risk averse or variance minimizers. These values were rounded to 20 and 80% and were used as the +1 factor level for the proportion of households in each production strategy. For the -1 level these 218 proportions were reversed. The levels of wage used in the sensitivity analysis were derived from the VR data on the prices that were paid for weeding and the mean times to completing a weeding task. Daily wage rates ranged from Z$1.77 to Z$3.04 day". The mean (Z$2.54) and standard deviation (260.53) of these were calculated. Factor levels were set at the mean plus and the mean minus one standard deviation. The values used for the minimum and maximum number of years before a field was left fallow were derived from my discussions with Vrs and other community members. The ranges used in the sensitivity analysis were selected to reflect a mean of 5.5 years with a large variance (+1 level) and a mean of 6.5 years with low variance (-1 level). Ten monte-carlo runs, each of 50 years with a 20 year warm-up period, were run for each combination of experimental factors. Data were collected from all montecarlo runs. A random sample of 20% of the 100 households were saved for subsequent analyses. Mean effects due to each factor were computed using Yates algorithm (Yates, 1937, described in Box at al., 1978). Tests of significance for each response variable were made by comparing the following ratio to the t-distribution with three degrees of freedom: A B 7138-) iii-‘8— 219 where; s. . (4302 + (00512 + @002 3 A, B... were the mean effects of factor A, B,...N on the response variable of interest, (N=13); ABE, CDE, 800 were the effects of runs 30,31 and 32 combined as error terms. The results of sensitivity analyses were presented as the mean effect of each factor changing from the low (-1) to the high (+1) level. The sign on these means indicated the direction of the effect. Also shown in the results tables were the mean effects expressed as percentages of the overall mean for the response variable of interest. To calculate the value of the response variable at the low level one need only subtract half of the mean effect for the factor of interest from the overall mean effect for the variable of interest. The value of the response variable at the upper limit would be calculated by adding half of the mean effect to the overall mean effect. 220 VERIFICATION AND VALIDATION Rainfall and seasons RAINGEN was used to generate 10,000 annual rainfall values. The mean (725mm) and standard deviation (222mm) of these values were no different to those observed for Angwa Bridge (t=-.004, DF = 9998, p>.99). The season classifier in MASOKA1 was used to classify each season for which there was data from Angwa Bridge (1977/78 to 19991/92). The difficulties of comparing these classes to those derived from local knowledge have been discussed in Chapter 4. Where we have the broadest data for classification (1988/89 to 1990/91) from the questionnaire survey, the model and data classifications were the same. In the period 1980/81 to 1987/88 the model and VR classifications agree completely in five out of the eight seasons, differ by one season class in one season and differ by two season classes in two seasons. Given the good agreement between the survey and model classifications, and the reasonably good agreement between VR and model classifications, the model season classification system was accepted as being useful and sufficiently accurate for the purposes of this study. Initial conditions Soil characteristics were assumed to be homogenous for specific types of soils. The accuracy of the soil, the slope and the vegetation cover data layers were not determined. The spatial patterns of household placement on the Masoka 221 landscape were similar to patterns observed on air photographs of the Masoka agroecosystem. To verify that INITIAL was working correctly it was used to generate 100 households in 10 separate and independent runs. The household data files produced were saved and the distributions of adults, children and land analyzed. The distributions of number of adults per household, produced by INITIAL, did not differ from observed distributions (Mann-Whitney U = 36195.5, p>.937). The distributions of numbers of children for each number of adults, generated by INITIAL, did not differ from observed distributions (T able 5.5). Table 5.5 Goodness of fit test statistics for the distribution of numbers of children per number of adults. Number of Adults Chi Square‘ '2 2.16 3.44 4.84 8.94 10.40 5.77 6.97 4.89 20.21 CDQNOTUI-hwmd 1. Critical value of the Chi square distribution for v=14, iii-.05 is 23.685. The hypothesis that the generated data were from the specified distribution was rejected if the Chi square values in the table were greater than this critical value. 2. Cells with fewer than five values were grouped. 222 The distributions of land for each household class (is. based on the number of adults) were derived from only 70 data points. In many cases the data in the tails were smoothed to give a continuous distribution of land area rather than the discontinuous distributions described by the data. To compare the non- riverland areas generated by INITIAL with those of the survey data, the 70 data points were compared to the 1000 points generated by INITIAL Ouantile plots were produced of the generated data and of the survey data (Figures 5.16a,b). These plots show the proportion of the data that were in each quantile. The similarity in these plots suggests that the land generation routines in INITIAL produced land holdings from the same distribution as occurs in Masoka. The proportion of households with riverlands (60%) was not very different from the observed proportion (54%). The mean areas of riverland per household generated by INITIAL were not different from those provided by the Vrs (Mann- Whitney U = 33825.5, p>.33). The distributions of these two data sets were however, somewhat different. The survey data were from a more positively skewed distribution (G117 = 4.578) than the generated data (G1 = 1.353) and exhibited a very much more platykurtic distribution (G2 = 27.712) than the generated distribution (G2 = 0.956). With the exclusion of the one outlier in the survey data these distributions (Figures 5.17a,b) were however, sufficiently similar to warrant The G1 and 62 statistics refer to the third and fourth moments about the mean - the skewness and kultosis of the distribution (Zar, 1984). oa- - 06- I 4 < ( I l- i- 305" ~ 306)- l . 8 8, | 30‘.- -1 20‘- I -I 02- - oz-l ~ OD 111 11 0011111111 01 204687 012046878 mm “W Figure 5.16. The cumulative proportion of a) actual (n=70) and b) simulated (n=1000) Masoka households cultivating a given area of non- riverland. 224 a) b) 1.0 T I T I T ‘IO U I I V as '- " 08 I- I -l < < I I- f- 3 on - . a on - ~ 8 8 ga‘ " "1 20.4 I- cl M «It 02 l- " 0.0 l 1 1 1 00 1 1 -2 O 2 4 8 8 -1 D 1 2 8 I-iarlvsrhdedlrated “WNW Figure 5.17. The cumulative proportion of a) actual (n=70) and b) simulated (n=1000) Masoka households cultivating a given area of riverland. 25 our acceptance of the INITIAL riverland generating routine. In summary, the data generation functions built into the INITIAL model produce distributions of adults, children and land that are similar to those observed in Masoka. Maize yields The simulated yield statistics of maize grown on each soil for good and average seasons (T able 5.6) were the same as the expected means derived from VR probability distributions (T able 4.9). Poor season, simulated yields were between 11 and 25% higher than expected means (Table 5.6). These differences were expected. When, in the model, the season was classed as poor with membership one, any yield value within the set of poor season yields was possible. As the membership of a poor season declined (and rainfall increased) the lower yields from the set of possible yields were excluded. This would result in an increasing mean yield with increasing rainfall. The simulated yields for poor seasons were, therefore, expected to have higher means than those of Table 4.9. Statistics of yields for each season and for each soil type were confounded by the proportions of seasons and each soil type used in the model compared to those used in deriving statistics for Table 4.9. The proportion of good, average and poor seasons in the verification rainfall data were .16, .56 and .27 compared to the .3, .3 and .4 values used in Table 4.9. These probabilities of each season type were used to calculate yield statistics for each soil type, based on the data of 226 Table 5.6 Simulated maize yields (kg ha") for each soil and in each season. Data are means with standard deviations in parentheses, medians and the sample size. Soil type Bepe Jecha River jecha Mean1 Season (kg ha") (kg ha" (kg ha ) (kg ha ) 1800 1600 3350 2150 Good (884) (486) (401) (918) 1700 1500 3350 1900 n=131 n=433 n=248 n=812 1800 1050 3450 1950 Average (428) (423) (268) (1 139) 1750 950 3400 1550 n=473 n=1499 n=936 n=2908 850 500 1500 850 Poor (218) (191) (218) (494) 850 500 1500 700 n=212 n-674 n=421 n=1307 1550 1000 2900 1700 M98" (566) (522) (892) (1092) 1550 850 3300 n=816 n=2606 n-1605 n=5027 Table 4.9. The means derived for bepe, jecha and river jecha soils (1600, 1000 and 2900 kg ha") were no different to the simulated means (Table 5.6). These results suggest that the approach to season classification and yield calculation used for maize was appropriate and produced acceptable maize yield values. Seasonal mean yields were expected to be more variable than soil means owing to the dependence of jecha yields on cropping history. The mean yield of virgin jecha and of jecha cultivated continuously for five years from Table 4.9 and the probabilities of each soil type derived from the verifying simulations (bepe=.16, jecha=.52 and river jecha=.32) were used to calculate seasonal mean yields. The good and average season yields (2150 and 2050 kg ha") were essentially the same as simulated yields whilst the poor season mean yield (750 kg ha"), as 227 expected, was lower than the simulated value. Cotton yields The simulated yield statistics for cotton (Table 5.7) were similar to those derived from VR yield probability distributions. Mean yields for bepe, jecha and river jecha (850, 700 and 1100 kg ha") calculated using the VR derived means (Table 4.9) and the season probabilities used in the verifying simulations, were Table 5.7 Simulated cotton yields (kg ha") for each soil and in each season. Data are means with standard deviations in parentheses, medians and the sample size. Soil Bepe Jecha River Mean 1 jecha1 1 (kg ha' ) (kg ha") (kg ha' i (kg ha ) 1000 1000 1850 1150 Good (379) (476) (421) (471) 900 1 100 1300 1200 n=44 n=290 n=167 n=501 1050 800 1400 1050 Average (370) (430) (295) (469) 950 750 1300 1 100 n-=157 n=1015 n=558 n=1730 350 300 700 450 Poor (159) (180) (342) (308) 300 300 650 350 n=68 n=421 n=243 n=732 850 750 1200 900 Mean (452) (473) (448) (513) 800 650 1200 850 n=269 n=1726 n=968 n=2963 28 close to the simulated mean yields for each soil type (Table 5.7). As with maize yields the simulated, poor season means were higher than VR derived means for each soil type. The simulated yields on each soil type in a good season suggest a bias toward yields higher than might be expected from the VR derived yield probability distributions (T able 5.7). The mean yields for each season were compared to those derived from the VR yield probability distributions using the proportions of each soil type used in the model simulations. Mean cotton yields for a good, an average and a poor season were 1000, 1050 and 300 kg ha". Apart from the poor season mean which we expect to be a little higher than the VR derived mean, these values are satisfactorily close to the means derived from VR yield probability distributions. Sorghum yields The simulated yield statistics for sorghum, grown on jecha soils (T able 5.8) were slightly higher than VR derived values in a good season (Table 4.9). Average and poor season yields were the same as means derived from VR probability distributions. The overall simulated mean for sorghum (500 kg ha") was not different than the mean yield derived from the VR yield probability distributions (450 kg ha"). 229 In general, the simulated yields compared favorably with expectations. These results increased our confidence in the season classification system and in our ability to generate yield patterns that conform to the VR probability distributions. Table 5.8 Simulated sorghum yields (kg ha") on jecha in each season. Data are means, with standard deviations in parentheses, medians and the sample size. SOII Jecha Season (kg ha") 800 Good (337) n=275 550 Average (268) 500 its 1 01 5 100 Poor (122) 50 n=438 500 Mean (351) 400 n=1728 As discussed in Chapter 4 there were no published measurements of actual yields that could be used as empirical comparisons of these simulated yield statistics. The yield values of Table 4.7 were all within the range of values that the 230 MASOKA1 model generated. The mean, poor season maize yields on jecha that households claimed to have harvested (500 kg ha", Table 3.2) was the same as that predicted by the model. Both the poor season cotton and sorghum yields reported in Table 3.2 appear lower than those derived from the model but were still within the range of yields predicted by the model. The yields of Table 4.7 were difficult to compare with the model derived values because these statistics were from situations where the proportions of soil types were unknown. Despite these unknowns the model generated yield statistics for poor season maize and cotton and average season cotton and sorghum that were reasonably close to the observed yields presented in Table 4.7. The average season means derived from the model, whilst appearing far higher than the average season data of Table 4.7, were variable and we expected mean yields as low as 800 kg ha" with a high probability. There were too few samples of good season yields presented in Table 4.7 to enable meaningful comparisons with the model. Erosion SLEMSA was validated in the high rainfall areas of Zimbabwe (Elwell, 1978) but not in the low rainfall areas, such as the Zambezi Valley. The model was, however, in general use throughout the SADC region. Erosion results were perceived as an index of erosion risk rather than actual erosion amounts that might be experienced under the conditions described in any model simulation. Simulated erosion for cultivated areas of Masoka were between 281 1.4 and 9.3 t ha" year" on cropped lands. These values were reasonably close to the erosion values estimated for each soil type by Elwell (1978), using a rainfall simulator (0.8 and 9.3 t ha" for bepe with 3.8 and 5.7 t ha" on virgin and cultivated jecha). The EROSION sub-system was the least closely examined of the sub-systems in MASOKA1. In the original model design I planned to use a spatially dependent erosion model. That aspect of MASOKA1 was not yet developed and an unaltered SLEMSA was used as an Interim measure. RESULTS Sensitivity analysis Model response variables were grouped as needs, yield, economic and resource responses. Needs responses included the mean proportion of households who failed to satisfy their cash (ACDF), maize (AMDF) or sorghum (ASDF) needs each year (Table 5.9). Yield responses were the mean yields (kg ha") of maize, cotton and sorghum, from all fields, corrected for labour deficits, achieved over the 30 year simulation. These means were then averaged over the 10 monte carlo simulations. Economic response variables included, for maize, cotton and sorghum, mean total returns to a household (MZRET, CTRET, SRREI'); mean returns per hectare (MZHA, CTHA, SRHA); mean returns to own labour and management (MZKAB, CTLAB, SRHA); and mean returns to initial investment (MZINV, CTINV, SRINV). As with the yield data these were the means of all households averaged over the 10 232 monte carlo runs. The resource response variables were the changes in the total number of years that each jecha cell had been cultivated summed over the whole landscape (CHGFRT), the mean erosion (t ha") over the landscape (MNERS), and the change in the total Fb value of the landscape, or the Fb value of each cell summed across the entire landscape at the beginning of the simulation minus the sum of all cell Fb values at the end of the simulation. Lower Fb values indicated a higher erodibility. Household needs satisfaction For the most part, the responses of ACDF, AMDF and ASDF were in the direction and of magnitudes that were expected. The effects of rainfall and the price of maize are obvious and require no explanation. The effects of factors such as the land multiplier, the proportion of households in each production strategy or the effects of the membership functions used were easily explained or the effects were too small to warrant discussion. Increasing household cash needs had a large impact on the proportion of households that satisfied cash and food needs. The effect was contrary to what was expected and required a more elaborate explanation. At the +1 level of land availability, households attempted to cultivate, on average, 70% more land than the real world households in Masoka cultivated. Given that in the model households cultivated fixed proportions of the land 233 available to them, irrespective of their labor resources, it was not surprising that more land would increase the levels of deficits as households would have had greater labor shortages and consequently achieved lower yields. Cotton’s greater sensitivity to weed cover was reflected in the greater response of ACDF to increases in land availability (T able 5.9). The mean percentage of households that were deficit for each need each year decreased when the poor season classification changed from the low to the high level (Table 5.9). This was expected because at the +1 level the season was classified as poor (upoor > uaverage) at higher rainfall levels than at the -1 level. The yield functions for all three crops showed higher yields as upoor approached zero than when uavmae was small. Thus, at the +1 level of poor season classification I expected higher mean yields and, therefore, a lower percentage of households that were cash, maize or sorghum deficit. The classification of good seasons had a similar trend for households that were cash deficit but the opposite trend for the proportion of households that were maize deficit (T able 5.9). For the negative response of the percentage of households that were cash deficit the explanation was much the same as that for the poor season classification; the yields at the high end of an average season were higher than those at the low end of a good season. At the +1 factor level for good season classification, ugood is dominated by uaverage over a wider range of rainfall amounts than was the case at the -1 factor level. I expected, therefore, the mean yields to be higher at the +1 factor level and for there to be fewer cash deficit households. The explanation for .96.$2.9seasoasesssossosacafiaooaosaaeecooesoofiaecssofiaao .0 see. .0 0.0.00 .0. as. 80 .02. 00.3. a 9% .02. 020.00 6.. 0:0 E 00.1 va a... 5.01 va 2:... 50.01 va .0 ecu 59. 0.0.00.0 so; 285. so. . 008030 00.0: 08.05.00 8:80.090 .0 seawataboaeaaoefiososs .. 0.: 0.3 0.00 000:. .065 e .0 02 02 0208.800 888 000.02 02 a: 0. 0.0. 02 000.505 30:0“. 2. 00 0.00 : cm 0.0 02 0:00.20 0 08.. wz N 0; m. .6. 2020050000 000000 0000 02 0. ha. 0 0.0 20.00.06 $32.2 .0 5082a v. 0.0. n. E- 0. mi. £550.86 80000 .ooc 0.- 0...- 02 e 0.0. 80.. 0002 t. 09 0.00. 3+ 0.0. 0.- 0.0.- 008.. 0000 t R. «.8. 00- 0.00. 0.- 0.0.- 5.0.00 0 0.0 02 0. «.3 0.0.0.05 000.. .0000 6000 0.0 a. :8... 0.0 0. aces. «0.0 0.... 6000 :8... 00.288; 0... “.00... 00.9.83. .0... ”.052 000088.. 0... “.00.. been. . so. .902 a. 80:05 c. .00.. coco .5000 50$ 5:093 0:0 .002... case 600$ :08 so; a... 000083.. a o... .o 08:. o... .0 0.20800 0.0 0.0:. 235 the reverse of this expectation for maize deficit households was not clear. Firstly, the difference was slight (1%) and despite the significance (P<.05) could be due to random error. Secondly, the difference in yields between a high uaverage membership compared to a low ugood membership was not as great for maize as it was for cotton. As expected, when the proportion of households that were classed as variance minimizers was low (level -1) the percentage of households with a cash deficit was higher and the percentage of households that were maize deficit was lower (T able 5.9). Income maximizers planted a greater proportion of their lands to cotton and less to maize than did variance minimizers. At the low factor level (80% income maximizers), therefore, I expected fewer households to have a cash deficit but more to have a maize deficit. The percentage of households that had a maize deficit and the percentage that were sorghum deficit were sensitive to whether the Z$155 subsidy was paid as a dividend or as a loan (Table 5.9). In both cases (AMDF and ASDF), the provision of a loan resulted in a greater percentage of households with a deficit than did the payment of a dividend, and for sorghum the provision of a loan resulted in an almost 30% increase in deficit households. Loans were provided to only those households that grew cotton and were provided only as inputs to cotton. Dividends, on the other hand, were paid to all households and were not tied to any productive activity and could therefore, be used to satisfy household needs. With lower total inputs of cash to the community (not all households 236 receiving loans) and loans being tied to cotton production, we expected a lower proportion of maize and sorghum deficit households with dividend payments than with tied loans. Household fallowing practices had a significant impact on the percentage of households that had a maize deficit each year (Table 5.9). At the -1 factor level the minimum number of years that a field would be cultivated before being left fallow was five. All yields of maize grown on jecha at this factor level were based on the five-year cultivated jecha function which had a much lower mean yield than the uncultivated jecha function (Table 4.9). I expected, therefore, that at the +1 factor level, where yields could be based on the virgin and the cultivated functions, the mean maize yields would be higher and fewer households have a maize deficit. This expectation is consistent with the data of Table 5.9. The effect of increasing household cash needs was to reduce the percentage of needs deficit households (Table 5.9). Similarly, several other model response variables responded to increasing household cash needs in ways that were contrary to expectations. A tentative explanation is presented (Figure 5.18). The argument had two components. The first was as follows: When household cash needs were increased more cash was required to satisfy basic household cash needs and less was available to meet the relatively high initial costs of 237 00000.00 2000 0:0 000:. 0.020000: :0 0000.. 2000 0.9.0000: 05000.0... .0 0.0020 0... .0 0.0.00.0 000. .00000 .90 0.00.... 20.8000... 0~.0E .90... l 0.5.00 0.00 .0. 050:. 00.0....w muse: :08 29.0030: 00020:. 0:00 .0200 :0 0200.0 0.2.9.096 0000.00 l 050.... .0 2000 2000 202.000.... Hi I 00:0E00 .000. 00:00 0. 002.0... 0~.0E 0. 00.00.... - ..... - .v i 0.00000... . . .- -- . {.0590} . ‘ 0 ‘ - - - - . 00.0... 00000.. . \ I .. 00:00 .. 000.2 4.... ...00_0..0000... .. .. 0.00.300... - - + .. 5.600 .. ...--+a... § ' ' - ‘ - ‘ I I \ l 0 0 0 238 producing cotton- The per household area planted to cotton was therefore, expected to decrease. This decrease would have three important consequences- Firstly, household expenditure on agriculture would have decreased, leading to higher household cash balances which could be used to satisfy cash needs. Secondly, smaller areas of cotton would have resulted in lower labor demands, a lower likelihood of labor deficits and lower labor deficit induced yield reductions. The third. and probably most important consequence of reduced cotton planting would have been the increased areas planted to maize, including areas of the better soils. The greater area planted to maize, coupled with the higher average maize yields would lead to higher total maize production and hence to lower household maize deficits. Households with maize surpluses could sell their surpluses to offset their cash deficits. To reduce the same cash deficit households required between three and 10 times as much maize as they would cotton. Thus, once household maize and cash deficits were satisfied there was not likely to be a cash balance sufficient to support cotton production. From this hypothesis a number of predictions were deduced. Based on this hypothesis we would have predicted that average maize yields would be higher when household cash needs were at the higher level. We would also predict that there would be a lower percentage of maize deficit households when household cash needs were at the higher level. Maize yields were higher when household cash needs were at the higher level (Table 5.10). From Table 5.9 we had already seen that higher household cash needs resulted in a lower percentage of 239 households that were maize deficit each year. Acceptance of this hypothesis was contingent on acceptance of a few key assumptions. the most important of which was that whilst households unable to produce cotton would increase the probability of their satisfying household cash needs. they would not be earning sufficient cash to enter into cotton production. These assumptions required further analysis for me to have felt entirely comfortable with the hypothesis- A second, and perhaps more plausible hypothesis (but not exclusive of the first) was as follows (depicted by the dotted lines in Figure 5.18): Wlth increased household cash needs, many of the more marginal cotton producers (those with poor soils or who had a labour deficit) would be excluded from cotton production due to their not having the capital to invest in inputs as described in the previous hypothesis. These households would have to switch to maize production. The remaining cotton producers would be those households that were more likely to achieve high yields as they would not be labor constrained. They would also be more likely to satisfy household cash needs. The marginal cotton producers (i.e. those who were likely to be labor deficit) who had switched to maize were likely to have been those households that were not satisfying household cash needs. in part because of the high costs of cotton production. These households, relieved of the burden of cotton production might have been more likely to satisfy household needs through maize production owing to the lower costs of production. I assume here, that these households were generally smaller (hence the labor constraints) and therefore quite likely to satisfy household cash needs 240 from maize and sorghum production. As with the first hypothesis this one required further investigation to be acceptable. Although they may both be plausible, these hypotheses are related to explaining the behavior of the model and did not necessarily provide acceptable explanations of real world processes. It was clear that some aspects of the model specification were contributing to behavior that we would not have expected in the real world. In particular, households in the model planted a set proportion of their fields to specific crops, irrespective of the labor they had available. This could lead to households’ reducing their chances of satisfying needs by planting areas that were larger than they could cope with. This was an unrealistic expectation which was to be corrected in future versions of the model. By proposing these hypotheses, I did not discount the possibility that the model contained an error that resulted in the observed effects of increasing household cash needs. Crop yields As expected. mean yields of maize, cotton and sorghum were all markedly sensitive to changes in the mean annual rainfall (Table 5.10). The yields of maize and cotton more than doubled with the change from a low mean annual rainfall to a higher mean whilst the yields of sorghum more than tripled. The +1 level of fallowing practices led to a reduction in maize yields but increased cotton and sorghum yields. The effects on cotton and sorghum were what was expected; at ..0>0.$00.0..-0£Eo...00.0§.0.0.00.0500.0000020000000.00055000050526.3000 .0 0.00. .0 0.0.00 .0. 00. 000 .02. 00.00 .0 0.00 0.0.. 00.00... .00 000 .0 00." V0 43. 5.0" V0 .03... 50.0" V0 .0 Son 0.0.. 0.0.00.0 0.0.... 0.005 .80. . 0.000020 00.0: 0000.00.00 000000.090 .N 241 80008....o.0000...0.0>00£.o$ ... 000 can 003 :00... ..0.0>O z. 0 0. z. 0 00 2. 0 00 020050000 000000 .000 :5 . 0 02 :3 0 am 00000 0000 t .v E 02 . . 0.: 00.000.00.00 000000 000.02 2:. 0. mo 02 0:. 0. 2. 002.080 00.39.00 2; 0: 0.0 :5 2 .00 2:. 00 08. .0000 0.0 0.. .00.... 000.2 0.0 0.. .0000 000.2 N0.0 .0. .0000 0002 .000 9.. 0.500 .000 9.. 0.50 .000 9.. 00.5.. .000... .0003. .000. 0. 0000000 0. 6.5.0. 0500.00 000 6.50. 00000 .6052. 0000. .0 00.0... .8000 0000. .0 32.0000 3.0 0.0.» 242 the +1 level we would expect a larger number of fields cultivated for fewer years resulting in higher yields. The reduction in maize yields at the +1 factor level was difficult to explain. Despite the significance of the result, the response (70 kg ha") was within the range of error I expected from model yield functions. Mean yields for all three crops were sensitive to the average season and the poor season membership functions used. The direction of the effects were all the same with an increase in yields at the +1 factor level for both seasons. With the poor season membership function at the +1 level we would expect higher yields as at all but the lowest upoor levels the +1 level upoor dominated -1 level upoor (area c of Figure 5.19). For the average season membership function the explanation for higher yields at the +1 factor level was a little more difficult to visualize but was nonetheless straightforward; when moving from the +1 level to the -1 level the effect was to have a wider range of annual rainfall amounts dominated by low membership values of the set of poor season yields and a wider range of annual rainfall amounts dominated by low membership values of the set of good season yields (areas a and b in Figure 5.19). At the +1 level areas a and b of Figure 5.19 would have been dominated by low membership values of the average season (the A+1 curve). Yields at low membership values of an average season, on each soil type. were higher than yields at low membership values of either poor or good seasons. Thus, when areas a and b of Figure 5.19 O 0 0 mm QHLMLGDEQE Commwm Figure 5.19 243 1.2 0 1 - ,r ’7- H ' / I {7; d / I / L O a 8 A ’ /’ 0’ I. r o I / E I ’/ (D 0. 6 ‘ ’l / E .‘ .0 ./ .’ / C l /’ O 0 I 4 - {-1. /’ Ul I '1. 010.2“ ' . \ b 0 1". 0 400 560 660 760 860 960 who 11b0 1200 Qalnlall (mm) p+1 A+1 - ------- (3+1 —p-1--- 0-1- ------- 5—1 Figure 5.19. Good (G), average (A) and poor (P) season membership functions used for +1 and -1 levels in sensitivity analysis. (I'I 244 were dominated by the average season membership function (at the +1 level) higher yields were achieved. The responses of maize and sorghum yields to increases in household cash needs have been described in the previous section and with reference to Figure 5.18. Economic response variables Apart from the consistent sensitivity to rainfall and household cash needs there were no patterns of sensitivity to factor levels across all economic response variables. Not unexpectedly the returns to maize were sensitive to the change in maize price (T able 5.11). Increasing the maize price from 250.26 kg'1 to 230.80 kg'1 (i.e. 208%) resulted in increases of 22% in returns to land. labor and initial investment as well as an increase of 17% in total returns to households from maize. Increasing household cash needs by 41% increased returns to all inputs for maize. Given the explanatory model presented in Figure 5.18 these increases were expected. The model of Figure 5.18 indicates higher maize yields with increased household cash needs and therefore, higher returns to land, to labor, to initial investment and total returns to households. .TwNMN. .7000 0N. .~-0..|0.N. .xb.o..00:o0 0N. >2.N.2 0<..Ns. (INS. KENS. .200... . 0.059.: 000 «>2.NS¢ 0:00:80)... .325 O. 36.0 AmsN—Zv 0°95. 60‘:sz USS. 0. 00:03.05 .CrWKN—Zv 0:01.00: .GuOu 00 NCNOE 02. CO “@8530 .08. LOuONs 50 0000000 00s.». 0. .35 .sOIL. 245 .06. .+ 00.0. .-00.0.0..00.0.0..0.0.00.00.00.000000.00..0000.000E 50009005300000 0.00. .0 0.0.00 .0. 0.0. 000 .02. 00.00 .0 0.00 0.0.. 0.0.00.0 .00 000 .0. 00... v0 ..:. 5.0" v0 .2... 50.0» v... .0 0.00 0.0.. 0.0.00.0 0.0... 0.0000 .00. . 0.0.0030 00.00 0000.00.00 000000.090 00.00. .00. .0. .0000 0000. .0006 .0 0.. 00.0 00.. 000 000 0000. ..0.0.6 .00 O. 5...? .00 0 Md. 0:. 0. _.V- c N up- 80:00.5 30.3“— 0000.00 $2 $2 a. N. o _.l a. V mm \ :81— 00.000.00.00 0.0 O hrd a N de c N mw 00 m 3 $088 08a. : 0 000 02 . 0 0 z . 00 0.00.0... 000.. - 00.000000... 02 02 02 t v 00 000000 000.03. 0.050.000 m<>z=2 00 m hNd m2 0. N @w .25 GN :N 3 32:80.3. 0: 00 0.0 :0. k 00 z. 0. 000 :5 0.. 000 00.0.00 2; 00 00.. z. 0 0.0 2; 0 00 0: 00 00. 00000 0000 .2. «0. :0 z. .0. 00.0 2.. .0. 00. :5 t. 000. 00.... 00.0.2 .0000 .0000 .0000 .0000 0.0 0.. 000.2 0.0 0.. 000.2 0.0 0.. 000.2 00.0 .0.. 000.... ...0~ 0N. .100 0N. .700 0N. .~.0.0..0000.. 0N. >z_~s. 050.... $8... Emu... 0.00.. .0000... .0. $2.52. 0.0000000. .000. 0. 000 .32. .000. .252. 000. 0. 0.00.0. .5002. 0.00.0. .00. .0 00000. 00. 00 009.000 .000. .200. .0 0.00.... 0.: 2.0 0.00» 246 .000. .+ 00.0..-00.0.0..00.0.....0.0.00.00.00.000000.0000000.000.00.000000050600000 .0 0.00. .0 000.00 .0. .00. 000 .02. 00.3 .0 0.00 0.0.. 0.0.00.0 0.. 000 0.. 00.» V0 0:. .0..." V0 0:... .80...- V0 .0 0.00 0.0.. 0.0.00.0 0.02. 0.0000 .00. . 0.00020 00.0.. 0000.00.00 000000.090 .0 00.002050200500030... .. 00.0 0.0. 000. 000 000.0 00.000 02 00 - 00... 02 wz EoEBEEw . 0 00.0 02 0 0 0.. 02 0000 00.0.2 02 02 02 . 0 00 000.... $2 $2 wz 0 .. no. 3.50.0000... 50000 .0000 . 0 00.0 02 02 . 0 0.. 08000.0 30.0.. . 0- 00.0- 02 02 0. 0.- 00.- $000.00. 000.. z. 0. 0.0 : 0 00.0 : 0 00. z 0. 0... 0000000000 000000 .000 0.060.000 . 0- 00.? 0 0 00.0 02 z .0- 00.- m<>z..2 .0 00.0000... . 0a 8.- 0 0 .00 02 2; 0.. 0.0... 00000 0000 2. 0.. ...0. 02 . 0 00 2: 00- 03. 0000.20 .0 000.. .2. .0 00.0 0: 00 00.0 .2. 00 000. z. 00. 000 00.0.00. .0000 .0000 .0000 .0000 0.0 0. 000.2 0.0 0.. 000.2 0.0 .0 000.2 0m0 .0. 000.2 run 0N. .7000 0N. .00.. 0N. ..-0_0..00..0.. 0N. >zFo 055 <15 BE... 0.00.. . 08 .0. 92.5. 000.000... 00.0. 0.000 .058. .000. .218. 000. 0. 00.0.0. .9050. 0.00.0. .0.0. .0 00000. 0... 0000000 .000. .900. .0 0.0000 00. 0.... 0.00.- J- 1.1.5:: I «teen 2: 2 .m 32: 247 .65. fro—:9TofizfihggfibofifiofimsgeowcoamecaoE£ogcoon~BOcmmfio .m .98. 8 £58 .2 5. com .Asz 8.3 a 22 52. Eemsu 8: can 8 Siva as... Ed" va 6.5L Soda Va a San Ea... 822% So; 385 .82 . 9.536 as? 35.52% 8:85:9m .u :03 «we: 50* EOE =95>O *0 * .w Nh 8.0 CNN war camp...— .3505 5%.". cat 0 3N a: 0 N v.0 «Eu 0 m :5 0 V 838 gm at. Gr 36 «an, E. 8.6 :5 MP mm :5 _.w m-. «00302 393“. «cc w 8.9 cc. F 8.9 .2: w 0 :5 ON QN mg: 2&0 9!: hp 9 0.3 :5 h S. 0.? 9:. hr w @ rm :5 0:. hmw _aucam «00:0 50% 900:0 gm :8: 9m x 8% :8: 9m x. :8: «gm 3.. 522 A73 «NV :38 3 A72 «N. $52832 mu >Z_¢w 05mm (IN—fl hum—mm 58m“- .neacaa» 9 92.me Enos. E.E. 2 as 655. .82 .5m@ 2m. 2 8.32 .Eme 25.2 .29 .o 885 2. 8 £55 .26. .28. .o 2005 9P 2.» oz.» 248 At the +1 factor level total returns (per household) more than doubled. providing supporting evidence for the hypothesis of Figure 5.18 in which I suggested that the areas under maize production would increase when household cash needs increased. From greater areas I expected greater total returns. The unexpected higher total returns to cotton with increased household cash needs did not fit neatly into the explanatory model of Figure 5.18. One explanation was that, with the reduced areas planted to cotton and the consequent reductions in labor demands (and deficits) the yields of cotton (as well as those of sorghum and maize) would have increased. Whether those increases would have been sufficient to increase total returns by almost 50% required further investigation. Rainfall had the expected effect of increasing total returns for maize, for cotton and for sorghum, as well as to all inputs for each of these crops. The effect of increasing the proportion of variance minimizers from 20% (factor level -1) to 80% produced the results I had expected; variance minimizers planted a greater proportion of their land to maize so I expected higher total returns and slightly higher returns to land. labor and initial investment for maize. Only the returns to labor were not significantly improved at the higher level of variance minimizers (Table 5.11). With respect to cotton. the effect of changing the proportion of variance minimizers was not as clear. Variance minimizers planted less of their area to cotton so I expected. at the -1 factor level, the higher total returns to cotton as were found (Table 5.12). With larger areas however, I expected greater labor deficits and hence, lower yields per hectare with lower 249 returns per hectare. per labor day and per 23 invested. The data of Table 5.12 confirmed this expectation for returns to land and to labor but not for returns to initial investment. The effect for the latter, was however, very low (1% of the overall mean) and could be due to error. ' Economic response variables for all three crops were sensitive to the poor season membership functions used, although. except for the effect of the poor season membership function on cotton returns, these effects were small compared to the factors discussed previously. Only total returns to maize were sensitive to the average season membership function used and the effect was small (Table 5.11). For the most part the effects of changes in the seasonal membership functions were due to changes in yields as discussed above. The increased yields due to the +1 level of the poor season membership function probably accounted for higher total returns and returns to land, labor and initial investment for cotton (Table 5.12). Maize and cotton returns Were notably sensitive to the land multiplier used 0' ables 5.11. 5.12). Maize returns (except returns to labor) increased with increasing land availability whilst those for cotton decreased. These trends were not unexpected. With larger land areas households would have been less likely to have had sufficient capital to plant all of their normal quota of cotton fields than with smaller land areas. This would have been due to households having to plant a proportion of their fields to cotton. More maize and less cotton would, therefore, have been planted, thereby increasing the total returns to maize and reducing 250 those to cotton. Fallow practices had notable effects on sorghum returns and lesser effects on maize and cotton returns (T ables 5.11. 5.12. 5.13). For maize the returns to land. to labor and to initial investment decreased when fallow practices changed from the -1 to the +1 level by between 3% (MZINV) and 6% (MZHA, MZLAB) of the overall mean. Given the reduction in maize yields with the +1 level of fallow practices (Table 5.10) these changes in maize returns were expected. Cotton returns increased when fallow practices changed from the -1 to the +1 level. Total returns increased by 8 % and returns to initial investment increased by 5 % (Table 5.12). Yields at the +1 level of fallowing practice were expected to be higher than those at the -1 level; at the -1 level the cultivated jecha yield function would always be used (with lower yields) whilst at the +1 level the uncultivated jecha yields function would be used at least some of the time, leading to higher average yields. Sorghum returns increased by about 12% when the fallow practice factor changed from the -1 to the +1 level. This was equivalent to the increase in mean sorghum yield (13%) due to the fallowing practice factor. Total cotton returns increased by 7% as the wage rate increased from 232.01 to 233.07 (i.e. by 53%). I suspected that as the wage rate increased more of the marginal cotton producing households (i.e. those likely to experience labor shortages) would have gone out of cotton production. Expecting the mean total returns to cotton production to increase would have been plausible. Mean total returns r- househo Tc sensitive decreasi Loans ir constrair cotton, t< bepe or cotton. T PTOduce Although however, ETOSiOn. A CHGFRT examinin hOW the l the total These Dr r980“ rCe 251 returns represented the total returns to cotton averaged among cotton producing households. Total cotton returns, as well as returns to land and initial investment, were sensitive to the form that subsidies took with the mean of total cotton returns decreasing 50% when subsidies were paid in the form of dividends (Table 5.12). Loans increased the area planted to cotton but also facilitated more labor constrained households, that would otherwise not have been able to produce cotton, to plant cotton. Labor constrained households or those without access to bepe or riverlands faced relatively high risks of low or even negative returns to cotton. The effect of increasing the number of these households that were able to produce cotton would be to reduce the mean of the total returns to cotton. Although the effect was small, the higher returns to land from cotton were however, inconsistent with this model. Erosion, erodibility and "fertility" A number of the responses of the resource performance measures (MNEFl, CHGFRT and CHGFB) were completely contradictory to my expectations. In examining the model I realized that many of these responses were an artifact of how the changes had been calculated as well as the large temporal variability in the total number of years the landscape had been cultivated (and hence Fb). These problems made interpretation of the results of factor level changes on resource responses difficult. The landscape Where A sin developed WOW. The total ”Umbl ”0mm , simulations the ”limb er Clear W are practices W; 252 The change in the total number of years of cultivation summed over the landscape (CHGFRT) was calculated as: CHGFRT = ENDFR‘I’so - STRTFRT21 Where CHGFFlT = The change in the total number of years of cultivation in each cell summed across the 30 year simulation; ENDFRTso = The total number of years of cultivation in each cell summed over the landscape at t=50; STRTFFle = The total number of years of cultivation in each cell summed over the landscape at t=21. A simple simulation model that used the fellow functions of MASOKA1 was developed and initialized, with the same functions used in INITIAL, for cultivation history. The model was used, with only 100 cells, to examine the changes in the total number of years of cultivation of each cell summed across the landscape (T OTFRT) and over the same 30-year simulation period. Several independent simulations were carried out and produced the same results. TOTFRT as well as the number of cells left fallow each year were plotted in Figure 5.20. What became clear were the very large fluctuations in TOTFRT when the -1 level of fallow practices was used (5 year minimum and 7 year maximum). Despite the obviously 253 lower mean TOTFRT at the +1 factor level, by using the above equation to calculate CHGFRT the +1 level of fallowing practices appeared to increase CHGFFlT compared to the -1 level (Figure 5.20). These appearances were deceptive; the mean of several runs with the fallow test model showed that with 0-10 fallow practice mean TOTFRT over 30 years was 50% less than with the 5-7 fallow regime and 60% less with the 6-7 fallow regime that was used. in error. Unfortunately, the sensitivity analyses were conducted using the CHGFRT and CHGFB values as calculated above (CHGFB was calculated in a similar manner) and the results were therefore, of little use. In future versions of the model it was planned to use more suitable analytical methods for these data, autoregressive moving averages or spectral analyses were tools that were thought likely to prove insightful. Although mean erosion was calculated over the entire landscape and then averaged for each year, its dependence on Fb suggests that interpretations of the sensitivity of erosion (as well as Fb) to factor level changes should be made with far more attention paid to the time series than to the average values. SUI a O 5 0 r3 PL 4: 4. 3 3 2 2 JLmeaL COHHM>HHH3U P.» y C l 1 HmVHOH FiQUre 51 254 150 ‘ u 100 1 I 20 25 30 35 40 45 50 Slmulatlon gear -*'—'O-10 Fallow -43-'5-7 Fallow Figure 5.20. Total number of years each cell had been cultivated, summed over the landscape for +1 fallow practice factor level (min = O, max = 10 years) and -1 factor level (min = 5, max = 7 years). Result of single shnubfion. pr l str de' slrr SOT nee SOU retu sum to th the r Yield relJre and l fUncfi amoU CE” in 255 SUMMARY AND CONCLUSIONS The development and structure of simulation models of the Masoka agroecosystem were described and tests of model validity and sensitivity were presented. The model was developed to enhance our understanding of the structure and functioning of the Masoka agroecosystem and as part of a developing methodology for the analysis of complex agroecosystems. Three models were described: The agroecosystem model (MASOKA1) simulated household decisions regarding which of three crops (maize, cotton and sorghum) to plant on each field in each year, compared household cash and food needs to production, compared labor needs to available household labor and sought to employ labor to make up deficits. The model also calculated household returns to land, labor and initial investment for each of the three crops and summary statistics for all performance measures on an annual time step. Inputs to the agroecosystem model were generated by the initializing model (INITIAL) and the rainfall generator (RAINGEN). Yields were generated, within MASOKA1, using yield probability distribution functions fitted to yield probabilities provided by village representatives for each of three soils and three classes of season (good, average and poor). Elements of fuzzy set theory were used to develop continuous yield functions. The degree of membership in each season of each annual rainfall amount was used to determine the yields achieved by each household on each cell in that season. 256 Household production was corrected if households experienced labor deficits in key months of the growing season for each crop. Households that were labor deficit could employ labor from households that had labor surpluses but were food or cash deficit if the employing household had the cash or food to pay for the hired labor. Once household production was corrected for labor deficits the model used household yields, land areas planted and labor hired to calculate the total returns to the household for each crop as well as the returns to land, to labor and to initial investment. The model also estimated the erosion on each cell of the landscape, using the SLEMSA model of Elwell (1980). Outputs from the model were written to data files each year. Outputs of individual household performance could be written to files or random samples of household data could be written to data files. Household data were aggregated each year and written to separate output files. These data in turn were aggregated when the model was run in monte carlo mode and the output written to a monte carlo output file. GIS data layers of soil, slope, household field holdings, cultivation history and vegetation cover were key input files to the model. The model produced an updated cultivation history.GlS data layer at the end of the simulation as well as data layers of erosion and the change in soil erodibility over the landscape. The latter data layers could be produced at up to three times during the course of a simulation. Outputs from the rainfall generator and initializing model were tested against the original data distributions taken from Masoka and were found to generate at rai ea Co for 257 rainfall and initial conditions that were acceptably close to the original distributions. The yields generated by the agroecosystem model were compared to the yield distributions developed by the VRs and were found to work adequately. Interactive validation of model results with Masoka residents and VHS as well as Zimbabwean development professionals was still required at the time of writing. Preliminary sensitivity analyses of the model examined the effects of 13 factors on response variables. Crop yields were particularly sensitive to rainfall. The percentage of households that were cash, maize and sorghum deficit each year were sensitive to rainfall, the land area available to households, the cash needs of households, the membership functions used for each season, whether subsidies were provided as loans or dividends and fallowing practices. For the most part the relationships between factor levels and response variables conformed to expectations. The decrease in households experiencing needs deficits with increased household cash needs was surprising. A tentative model, not yet tested, was described to explain the effect. The effects of factor levels on economic response variables were largely attributable to the effects of rainfall on yields, and in the case of maize, to changes in the maize price. Maize returns were particularly sensitive to the maize price, rainfall, household cash needs and the proportions of households that were in each of the two household production strategies used for these simulations. Cotton returns were particularly sensitive to rainfall, the membership functions used for each season, the proportion of households in each production strategy and fallow house for ea dklnc ofthe chang invesfi Masok 1) 2) 3) 258 fallowing practices. Sorghum returns were particularly sensitive to rainfall, household cash needs, fallowing practices and to the membership functions used for each season. The model estimates of changes in cultivation intensity and soil erodibility did not take the cyclic nature of these changes into consideration. Interpretation of the results of sensitivity analyses on the responses of these variables to changes in the 13 factors were therefore, ambiguous and required further investigation. In conclusion, I proposed that the models developed for analyzing the Masoka agroecosystem, although incompletely evaluated, were useful: 1) 2) 3) 4) The models enabled me to examine the responses of a very complex system to changes both in exogenous inputs typical of policy decisions as well as to endogenous changes typical of household production decisions. The models enabled me to examine the feasibility of integrating indigenous knowledge as well as formal scientific knowledge to provide a predictive simulation model. The models enabled me to examine, simultaneously, the effects of various management or policy options on household and community performance measures. The models enabled us to use biophysical, social and economic performance criteria, in our evaluations of policy or management options. This capability would enable us to observe the relative trade-offs between inl so; the inc rea 259 these criteria and could lead to more informed decision making. As an aid to understanding the structure and behavior, the models proved informative and represent a largely untapped source of insight and hypotheses. The modeling methodology shows promise as an aid to decision making in data scarce environments. We would do well however, to remember that "..in running the models we investigate the consequences of the guesses and assumptions incorporated in them - model output may tell us nothing new about empirical reality." (Landsberg, 1987). BIBLIOGRAPHY Anderson, JR 1991. A framework for examining the impacts of climate variability. In: Muchow, R. C. and J. A. Bellamy (Eds.), Climatic risk in crop production: Models and management for the semiarid tropics and subtropics. C.A.B lntemational, Wallingford, U.K. 3-17. Anderson, JR. 1974. Simulation: Methodology and application in agricultural economics. Review of Marketing and Agricultural Economics, 42, 3-55. Balci, O. and R.G. Sargent. 1984. A bibliography on the credibility assessment and validation of simulation and mathematical models. Simuletter, 15. 15-27. Baudry, J. 1993. Landscape dynamics and farming systems: Problems of relating patterns and predicting ecological changes. In: Bunce, R.G.H., L Ryszkowski and M. G. Paoletti (Eds.), Landscape ecology and agroecosystems. Lewis Publishers, Ann Arbor, USA. 21 ~39. Berdegue, J.A., M. Installe, C. Duque, R. Garcia and X. Ouezada. 1989. Application of a simulation software to the analysis of a peasant farming system. Agricultural Systems, 30, 317-334. Box, G.E.P. 1979. Robustness in the strategy of scientific model building. In: Launer, R.L and GM Wilkinson (Eds.), Robustness in statistics. Academic Press, New York, USA. 201 -236. Box, G.E.P., W.G. Hunter and J.S. Hunter. 1978. Statistics for experimenters. An introduction to design, data analysis, and model building. John Wiley & Sons, New York, USA Buchanan, GA and ER. Burns. 1970. Influence of weed competition on cotton. Weed Science, 18, 149-154. Burnside, 0.0. and GA. Wicks. 1967. The effect of weed removal treatments on sorghum growth. Weeds, 15, 204-207. Debrah, S. and H.H. Hall. 1989. Data aggregation and farm risk analysis. Agricultural Systems, 31, 239-245. Dent, J.B. and JR Anderson. 1971. Systems, management and agriculture. In: Dent, J.B. and JR. Anderson (Eds.), Systems analysis in agricultural management. John Vlfiley & Sons, Australasia, Sydney, Australia, 3-14. 260 261 Deybe, D. and G. Flichman. 1991. A regional agricultural model using a plant growth simulation program as activities generator - an application to a region in Argentina. Agricultural Systems, 37, 369-385. Elwell, H.A. 1980. Design of safe rotational systems. Department of Conservation and Extension (AGRITEX), Harare, Zimbabwe. Elwell, H.A. 1986. Determination of erodibility of a subtropical clay soil: a laboratory rainfall simulation experiment. Journal of Soil Science, 37, 345-350. Elwell, H.A. 1978. Modelling soil losses in Southern Africa. Journal of Agricultural Engineering Research, 23, 117-127. Firbank, LG. 1993. The implications of scale on the ecology and management of weeds. In: Bunce, R.G.H., L. Ryszkowski and MG. Paoletti (Eds.). Landscape ecology and agroecosystems. Lewis Publishers, Ann Arbor, U.S.A. 91-103. Fresco, LC. 1986. Cassava in shifting cultivation. A systems approach to agricultural technology development in Africa. Royal Tropical Institute, Wageningen, The Netherlands. FSRU. 1985. Annual Report of the Farming Systems Research Unit, Department of Research and Specialist Services. Government Printers, Harare, Zimbabwe. Giere, RN. 1991. Understanding scientific reasoning. Holt, Rinehart and Winston, Inc, Orlando, Florida, USA. Halfon, E. 1979. The effects of data variability in the development and validation of ecosystem models. In: Zeigler, B.P., M.S. Elzas, G.L. Klir and T.I. Oren (Eds). Methodology in systems modelling and simulation. North-Holland Publishing Company, Amsterdam, Holland. 335—343. Hall, M.R., C.J. Swanton and G.L. Anderson. 1992. The critical period of weed control in grain corn (Zea mays). Weed Science, 40, 441 -447. Hutchinson, P. and LS. Unganai. 1992. The effect on rainfall frequencies of changes in mean annual and monthly rainfalls. In: Kronen, M. (Ed.). Proceedings of the Second Annual Scientific Conference of the SADCC-Land and Water Management Research Programme. Mbabane, Swaziland, October 7-9, 1991. 42- 55. Kingwell, R.S., D.A. Morrison and AD. Bathgate. 1992. The effect of climate risk on dryland farm management. Agricultural Systems, 39, 153-175. 262 IGngwell, RS. and DJ. Pannell, (Eds.). 1987. MIDAS, a bioeconomic model of a dryland farm system. Pudoc, Wageningen, Holland. Klein, K.K., Kulshreshtha and SA. Klein. 1989. Agricultural drought impact evaluation model: Description of components. Agricultural Systems, 30, 117-138. Landsberg, J.J. 1987. Book review. Forest Ecology and Management, 18, 161-163. Law, AM. and W.D. Kelton. 1991. Simulation modeling and analysis. McGraw—Hill, Inc, New York, USA. Maino, M.J. Berdegue and T. Rivas. 1993. Multiobjective programming - an application for analysis and evaluation of peasant economy of the Vlllth Region of Chile. Agricultural Systems, 41, 387-397. Manetsch, T. J., M.L Hayenga, A.N. Halter, T.W. Carroll, M.H. Abkin, D.R. Byerlee, K Chong, G. Page, E. Kellog and G.L Johnson. 1971. A generalized simulation approach to agricultural sector analysis. Michigan State University, East Lansing, Michigan, USA Mankin, J.B., R.V. O’Neill, H.H. Shugart and SW. Rust. 1975. The importance of validation in ecosystem analysis. In: Innis, G.S. (Ed.), New directions in the analysis of ecological systems. Part 1. The Society for Computer Simulation, La Jolla, California, USA. 63-71. Meentemeyer, V. and CE. Box. 1987. Scale effects in landscape studies. In: Turner, M.G. (Ed.), Landscape heterogeneity and disturbance. Springer-Veflag, New York, USA. 15-34. Naylor, T.H., J.L Balintfy, D.S. Burdick and K. Chu. 1966. Computer simulation techniques. John Wiley & Sons, New York, USA. Naylor, TH. and J.M. Finger. 1967. Verification of computer simulation models. Management Science, 14, 8-92 - B-101. Ngambeki, D.S., R.R. Deuson and RV. Peckel. 1992. Integrating livestock into farming systems in northern Cameroon. Agricultural Systems, 38, 319-338. Officer, RR. and J.L Dillon. 1968. Probability and statistics in agricultural research and extension: A pro-Bayesian view of modern developments. The Journal of the Australian Institute of Agricultural Science, September, 121-129. Owsinski, J.W. 1982. Agriculture, population, environment and balanced regional development: A modelling project in Poland. Agricultural Systems, 8, 87-104. I“ CF C\ C\ Q\ 263 Popper, Sir KR. 1968. The logic of scientific discovery. Harper and Row, New York, USA Pritsker, A.A.B. 1979. Compilation of definitions of simulation. Simulation, August, 61-63. Reynolds, P. 1991. Dance civet cat. Child labour in the Zambezi Valley. Boabab Books, Harare, Zimbabwe. Sargent, R.G. 1982. Verification and validation of simulation models. In: Cellier, F.E. (Ed.), Progress in modelling and simulation. Academic Press, London, UK Sauget, N. and G. Balent. 1993. The diversity of agricultural practices and landscape dynamics: The case of a hill region in the southwest of France. In: Bunce, R.G.H., L Ryszkowski and MG. Paoletti (Eds.), Landscape ecology and agroecosystems. Lewis Publishers, Ann Arbor, U.S.A. 113-129. Schwerzel, P.J. and P.E.L Thomas. 1971. Weed competition in cotton. PANS, 17, 30-34. Shannon, RE. 1975. Systems simulation, the art and science. Prentice-Hall, New Jersey, USA Shumba, E. 1988. Maize technology research in Mangwende, a high potential communal area environment in Zimbabwe. Part 1: Developing a research agenda. Farming Systems Newsletter, 34, 12-34. Steiner, J.L 1989. Modeling agroclimatic systems: Guidelines and Perspectives. Soil, crop and water management systems for rainfed agriculture in the Sudano-Sahelian zone. ICRISAT Sahelian Center, 11-16 January, 1987. Steinfeld, H. 1988. Livestock development in mixed farming systems: a study of smallholder livestock production systems in Zimbabwe. Wissenschaftsverlag Vauk IGel, FR Germany. Swift, M.J., P.G.H. Frost, B.M. Campbell, J.C. Hatton and KB. Wilson. 1989. Nitrogen cycling in farming systems derived from savanna: Perspectives and challenges. In: M. Clarholm and L Bergstrom (Eds.), Ecology of arable land. Kluwer Academic Publishers, Holland. 63-76. Ungar, ED. 1990. Management of agropastoral systems in a semiarid region. Pudoc, Wageningen. 264 Vemuri, V. 1978. Modeling of complex systems. An introduction. Academic Press, New York, USA. Welch, PD. 1983. The statistical analysis of simulation results. In: Lavenberg, S.S. (Ed.), Computer performance modeling handbook. Academic Press, New York, USA. 267-329. Wiegart, R.G. 1988. Holism and reductionism in ecology: Hypotheses, scale and systems models. OIKOS, 53, 267-269. Wiens, J.A., J.F. Addicott, T.J. Case and J. Diamond. 1986. Overview: The importance of spatial and temporal scale in ecological investigations. In: Diamond, J. and T.J. Case (Eds.), Community Ecology. Harper & Row, New York, USA. 145153. Wiens, J.A. 1989. Spatial Scaling in Ecology. Functional Ecology, 3, 385-397. Wright, A. 1971. Farming systems, models and simulation. In: Dent, J.B. and JR. Anderson (Eds.), Systems analysis in agricultural management. John Wiley & Sons Australasia, Sydney, Australia. 17-33. Yates, F. 1937. The design and analysis of factorial experiments. Imperial Bureau of Soil Science. Harpenden, Herts, England. Zar, J.H. 1984. Biostatistical analysis. Prentice Hall, Englewood Cliffs, New Jersey, USA. Zinyama, L.M., D.J. Campbell and T. Matiza. 1988. Traditional household strategies to cope with food insecurity in the SADCC region. Southern Africa: Food Security Policy Options. Harare, Zimbabwe, 1-5 November, 1987. Chapter 6. Overview and key Issues for further research INTRODUCTION In this chapter the results of Chapters 3 through 5 are integrated and discussed with respect to the research objectives defined in Chapter 1. In the last section of the chapter, key issues requiring further research are identified. OVERVIEW In Chapter 1 it was suggested that two problems needed to be overcome in conducting agroecosystems analyses. The first of these was the problem of applying performance criteria such as sustainabilty, stability, resilience, productivity and equitability. The second was the lack of a sound philosophical and methodological basis for conducting the analysis. The checklist of Grimm et al., (1992) was suggested as being a minimum set of definitions required when using these performance criteria; these factors were a) the level of description; b) the variable of interest; c) the referential behavior of the variable of interest; d) the nature of the disturbance; e) the spatial scale; and f) the temporal scale. 265 266 The field data collection and modeling methodologies presented in the preceding chapters not only required that I make explicit the primary level of analysis but facilitated use of several levels of analysis simultaneously. With these capabilities the modeling methodology in particular, could be used where emergent properties and multi-objective evaluations were likely to be problems. By using several levels of analysis, the household compared to the community for example, the policy maker could examine the impacts of policies from the household perspective. Given that several scales of analysis were possible in any simulation experiment, the impacts of household practices or policies on emergent properties could be examined. The large number of performance measures in model output provided an integrated assortment of social, economic and bio-physical variables that were available to evaluate agroecosystem performance. This cross section of variables facilitated evaluation of the trade-offs that might have to be made in the analysis of real world agroecosystems. The simulation models developed in this study were capable of generating expected or referential behavior based on a wide variety of assumptions regarding initial conditions, inputs and model parameter values. These referential behaviors could be used in evaluating the behavior of the agroecosystem. The large number of factors that were changed in the sensitivity analysis were indicative of changes that could be made to examine the effects of particular disturbances on agroecosystem behavior. The approach thus enabled the me to In n) (I) 267 specify the nature, direction and magnitude of a disturbance and to observe the impacts of these changes on the behavior of the agroecosystem using any of the large number performance measures estimated by the model. Both spatial and temporal scales were fixed parameters in the model with respect to some variables but not to others. The simulation time step and cell size were fixed parameters. The number of households was, however, a variable factor whose value might affect the values of performance criteria. Household numbers were explich set in each simulation. Viewed as a whole, the field and modeling methodologies provide a useful set of tools with which to address questions about agroecosystem behavior when either static or dynamic performance criteria were to be used in evaluating agroecosystem behavior. Some aspects the model expanded the criteria suggested by Grimm et al., (1992); the model provided information to examine the effects of changes in endogenous or exogenous control variables on agroecosystem performance criteria at several levels simultaneously; the model provided information that could be used to examine the effects of changes in endogenous or exogenous control variables on several response variables and several dynamic performance criteria. This capability would enable decision makers to evaluate the relative trade-offs that would be made under different policy or experimental conditions. The development of a simulation model as part of the methodology would enable me to conduct controlled simulation experiments which could be used to C0 268 explore the consequences of policy or household level changes in control variables. The unavailability of data was a major constraint in planning the analysis described in previous chapters. Methodologies were developed and used that enabled me to develop useful insights in to the structure and behavior of the agroecosystem using the best available data. Much of this data was in the form of the knowledge of agroecosystem residents. The methodology presented was sufficieme flexible to incorporate data from local knowledge sources as well as more formally measured data. These data varied greatly in quantity and quality but were integrated into a simulation model. Very little data was available on the Masoka agroecosystem before the analysis was conducted. The methods developed in this study enabled meaningful examinations of the agroecosystem to be conducted in spite of the limited availability of quantitative data. In particular the use of spidergrams facilitated quantitative analyses of the components and relative importance of inputs to and outputs from these components for an entire production system. The resulting graph of the production system could prove useful for focusing further investigations. The spidergrams also helped select those components of the production system that were likely to be most important and needed to be included in the simulation model. Although an approach to analyzing the behavior of the Masoka agroecosystem was presented the analysis was far from complete. Few statements could therefore, be made regarding the actual performance of the agroecosystem w'rl cle or 269 with regard to sustainability, stability, resilience and productivity. What became clear as the analysis proceeded was the dependence of these performance criteria on those factors listed in the checklist of Grimm ef al. (1992). The widespread use of these performance criteria belies the diffictu of actually applying them in the real world. Although tentative, the analysis presented in Chapter 5 indicated that the ability of households to satisfy cash and food needs were critically dependent on rainfall, the land available to them both through physical availability as well as through the labor available to manage a given area and also on the level of cash needs that households expected to satisfy. The land available to each household and the quality of life expected by households were key community and household decisions. The analytical approach adopted, with the emphasis on community participation, could assist the community and households to make these decisions with at least some information on the trade-offs that needed to be made. There was no doubt that aspects of the agroecosystem model lacked realism; The fixed proportion of a household’s lands that were allocated to each crop in each season was certainly not a reflection of reality. A more flexible household decision making algorithm, in which household plantings were made as a function of both cash and labor resources available was planned for future versions of the model. The use of a constant number of households with unchanging land holdings and number of adults and children were also not realistic but considerably simplified the model. The use of constant prices and wage rates was obviously an over simplification of what is likely to be a complex as of 270 and changing set of values. Within season rainfall distributions are likely to have significant impacts on yields and erosion, yet the poor within season resolution of the model made analysis of these effects impossible. Despite these many shortcomings however, the model provided a useful first attempt at simulating a very complex production system at several scales; the complex interactions that were apparent in the results of the sensitivity analyses justified using the multi- scaled systems approach; despite the scarcity of measured data I was able to develop a meaningful model of the agroecosystem. l was also able to use the model to examine the effects of changes in the policy environment (crop price changes for example) as well as the household decision making environment (switches in fallowing practices for example) on several performance measures. The model could therefore, have utility both as a tool for planning extension packages as well as for examining the effects of various agricultural policies. In conclusion, whilst the analysis and evaluation of the methodology were not complete, the methodology developed for analyzing the Masoka agroecosystem was useful and enabled the development of a fuller understanding of agroecosystem behavior. Some of the methods developed could prove useful in analyses of similar problems in similar classes of agroecosystem. 271 KEY ISSUES FOR FUTURE RESEARCH The approach to agroecosystem analysis described in this dissertation is clearly far from adequate. Several issues emerged from the analysis and subsequent writing. 1. A methodological approach to validating expert knowledge has not been reported in the literature. Validation of indigenous knowledge (Chapter 4) was difficult and the methods of doing so were not well developed. If, as is likely, indigenous knowledge is to be an important source of information then the issue of validating this knowledge needed to be addressed. Methods of analyzing the results of data collected from VRs were not readily apparent and require further investigation. In this regard Bayesian probability theory appeared to be a suitable starting point. The simulation models developed for analyzing the Masoka agroecosystem were quantitatively based. Although some use was made of fuzzy set theory, greater use of these concepts could have resolved some of the problems associated with using indigenous technical knowledge, such as the conversion of total production to production per hectare. Some analyses were not completed due to time constraints. Important among those was the expansion of model sensitivity analyses to finer scales of resolution. Were different classes of household, for example, differentially sensitive to changes in factor levels? How sensitive were the same performance criteria to simulation experiments conducted over longer 272 periods of simulated time? A key issue that was not dealt with in the prototype agroecosystem model was the effect of spatial dependence on erosion simulation. The erosion modeling component of the version of MASOKA1 presented in these chapters was far from adequate and perhaps the weakest aspect of that model. As was discussed earlier, the model and model results should be validated against VR opinion. The opinions of WWF MAPS Project staff would also be needed to establish the usefulness of the model and the approach. It was likely that the spidergrams that were developed during the field data collection were biased by conditions that were dominant at the time (the drought for example). Investigations were needed that would indicate the relative changes in the components and weights of components of spidergrams if these analyses were conducted at different times and places. The model as well as the overall approach to analyzing agroecosystems have potential utility both for evaluating government policy (extension packages or resettlement in the Zambezi valley for example), community level policies of land allocation and expansion as well as for identifying optimal household production practices. To utilize that potential the approach and results needed to be communicated to audiences that differed greatly in their technical understanding of the approach as well as in their objectives and values. One of the major challenges for the future 273 would be to present the findings of the study to these different groups and then to integrate their responses into future analyses. Despite its complexity the model was a greatly simplified version of the Masoka production system; only three of the 13 production activities listed by female VRs were simulated in the model. It was quite possible that a greater diversity of production activities could significantly affect the results of analyses of the agroecosystem. Methods of evaluating the sensitivity of model response variables to the way the model was specified needed to be devised and tested. A potential solution tothis problem might have been to examine the effects of allocating income to specific households at specific times in a simulation to represent income generated by these other activities. The income so generated could be either in the form of food (to represent vegetables or poultry for example) or as cash generated by craft or beer brewing. An advantage of the modeling approach used was the ability to examine effects such as these on different classes of household. 274 BIBLIOGRAPHY Grimm, V.E. Schmidt and C. Wissel. 1992. On the application of stability concepts in ecology. Ecological Modelling, 63, 143-161. "III’LIIIILIIIIIIIIIIIIIIIII