MODELING INTEGRATED SOIL FERTILITY MANAGEMENT OPTIONS IN MALAWI
UNDER VARIABLE SOIL AND CLIMATE CONDITIONS
by
Mary Helena Ollenburger

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Crop and Soil Sciences
2012

ABSTRACT
MODELING INTEGRATED SOIL FERTILITY MANAGEMENT OPTIONS IN MALAWI
UNDER VARIABLE SOIL AND CLIMATE CONDITIONS
by
Mary Helena Ollenburger
Sustainable agriculture requires integrated soil fertility management, which relies to a
considerable extent on the presence of legumes. On-farm participatory research in Africa has
identified pigeonpea (Cajanus cajan) as a promising legume for improving fertility management
in maize (Zea mays) systems. Including a long-duration legume like pigeonpea creates a tradeoff between improved soil fertility and increased water stress, and impacts of the legume on
maize production will depend on soil type and rainfall regime. This effect is difficult to quantify
using short-term field trials. Using the simulation model APSIM, results from on-farm research
in Malawi were extended to evaluate maize-pigeonpea rotation and intercrop systems relative to
continuous maize with modest (24kgN/ha) fertilizer inputs in all systems. Performance over
multiple years from system establishment was evaluated in different soil types, utilizing longterm weather records from two sites. Despite trade-offs between soil fertility and moisture, maize
diversified with pigeonpea produced over double the yields of continuous maize on sandy, low
fertility soils, with very low risk of failure. System performance under climate change scenarios
was evaluated through modification of weather station data, including changes in rainfall of +/10% and +/-25% and temperature increases of up to 4 °C to compare a maize-pigeonpea rotation
to continuous maize at two nitrogen fertilization levels. Overall, maize yields in the rotation
system were maintained under climate change scenarios, though declines were seen under some
reduced rainfall scenarios. These results suggest that maize-pigeonpea systems can improve
productivity in poor soils and are resilient to current and future climate variation.

ACKNOWLEDGEMENTS

Thank you first of all to Dr. Sieglinde Snapp, for taking bringing me to MSU despite my
unusual background, for giving me tremendous freedom in exploring problems, some of which
never made it into this thesis, and for supporting me throughout this entire process. Thanks as
well to my committee members, Dr. Laura Schmitt-Olabisi and Dr. Alexandra Kravchenko for
their support and input.
Special thanks to Dr. Wezi Mhango, without whose previous work (and data!) I wouldn’t
have been able to do any of this, and who was a great help and friend during the time I was in
Malawi.
This project was funded by the McKnight Foundation’s Collaborative Crops Research
Program, as part of Legume Best Bets II. Thank you to the foundation for their financial support.
Thanks also to the collaborators in in Malawi who helped me so much: particularly to Mr.
Ladson Chirwa and Ms. Esnart Kanyenda, without whose help in Ekwendeni I would have been
lost. Thank you to Mrs. Lizzie Shumba at Ekwendeni Hospital, who is quite simply incredible. I
am grateful for and humbled by the generosity of farmers who allowed us into their fields and
invited us into their homes: they are an inspiration and a motivation to me, always.
Thanks to Roger and David Stern, the greatest father-and-son math team I’ve ever met,
for hospitality and good work and great conversations. Thanks also to Thomas Mwakudisa,
James Musyoka, and Joyce Odour at Maseno University for excellent collaborations and new
friendships. Thomas and David helped me define and classify the rainfall scenarios in Chapter 2,
and Thomas wrote GenStat code that I developed into what are now called the B and C scenarios.

iii

Nathan Moore first suggested I use a scenario approach to climate change analysis. Todd Benson
generously shared weather data from Malawi.
Thanks to my lab mates and friends here, to Christine Sprunger for help and
commiseration, to Brendan O’Neill for helping me with lab work, to John Green for help
grinding soils and Mark Freeman for analyzing them.
Beyond the academics: Thank you to Jeff for so much love, support and encouragement,
for an occasional second home, and for many grand adventures along the way. And for
reminding me to always update my priors.
And finally: to my parents, your love, support and encouragement has been unwavering.
You supported this work—as you have supported everything I do—with laughter and hugs and
good food when I most needed it. Wherever I go, I know that I will always have a place to come
home to.

iv

TABLE OF CONTENTS

LIST OF TABLES

vii

LIST OF FIGURES

viii

LIST OF ABBREVIATIONS

x

LITERATURE REVIEW
Introduction
Legume diversification for improved soil fertility
Crop models and the Agricultural Production Systems Simulator (APSIM)
Modeling maize-legume cropping systems
Climate variability and risk in rainfed agriculture
Climate change
Objectives

1
1
2
3
7
9
12
15

CHAPTER 1
Extending participatory research results through simulation modeling:
Benefits, risk and trade-offs to using long-duration legumes in Malawi
1.1 Introduction
1.2 Materials and Methods
1.2.1 Field sites
1.2.2 Model Setup
1.2.3 Analysis of model results
1.3. Results
1.3.1 Evaluation of model performance
1.3.2 Systems including pigeonpea show higher yields than continuous
maize systems.
1.3.3 Risk and yield distribution change with cropping system and over time
1.3.4 Nitrogen effects of legumes
1.3.5 Water stress effects of legumes and impacts on yield
1.4 Discussion
1.4.1 Integrating modeling with participatory research
1.4.2 Maize-legume systems increase maize yield with low risk of failure
1.4.3 Soil water and nitrogen trade-offs
1.4.4 Long term impacts of maize-legume systems
1.5 Conclusions

28
29
30
32
33
33
34
35
35
36

CHAPTER 2
Modeling maize-legume systems under variable climate scenarios in Northern Malawi
2.1 Introduction
2.2 Materials and Methods
2.2.1 Generation of climate scenarios from current weather data

54
54
58
58

v

17
18
22
22
24
27
27
27

2.2.2 Weather data used
2.2.3 Scenarios used in simulations
2.2.4 Simulation setup
2.3. Results
2.3.1 Rainfall in climate change scenarios
2.3.2 Yield response to increases in CO2 concentration
2.3.3 Yield response to increases in temperature
2.3.4 Differences in yield response in rainfall scenarios
2.3.5 Water and nitrogen stress in maize
2.3.6 Rotation effects under climate change
2.4 Discussion
2.4.1 Soil fertility remains the limiting constraint in low input systems
2.4.2 Effects on high-input continuous maize
2.4.3 Effects on maize in rotation
2.4.4 Effects on Pigeonpea
2.4.5 Implications for use of long-duration legumes
2.5 Conclusions
REFERENCES

60
60
61
62
62
63
63
65
66
68
68
68
69
69
71
72
73
90

vi

LIST OF TABLES

Table 1.1: Properties used in APSIM soils input files

39

Table 2.1: Rainfall scenarios used in climate change simulations

76

vii

LIST OF FIGURES

Figure 1.1: Soil characteristics at field sites

37

Figure 1.2: Schematic depiction of time series model setup

38

Figure 1.3: Comparison of field-measured data and model results

40

Figure 1.4: Mean maize yields and years from system establishment

41

Figure 1.5: Cumulative probability distributions of maize yields

42

Figure 1.6: Mean effect of legume systems on maize yields

44

Figure 1.7: Risk of yield reduction in maize-legume systems

45

Figure 1.8: Inorganic N in soil at planting

46

Figure 1.9: Maize yield and cumulative nitrogen stress

47

Figure 1.10: Effects of legume diversification on maize yields and cumulative
nitrogen stress

48

Figure 1.11: Maize yields and in-crop rainfall

49

Figure 1.12: Maize yields and cumulative water stress

50

Figure 1.13: Mean soil water stress

51

Figure 1.14: Maize yield and extractable soil water at critical plant growth stages

52

Figure 1.15: Mean extractable soil water at critical plant growth stages

53

Figure 2.1: Baseline (2000) rainfall estimates for Lilongwe, Malawi from
statistically downscaled global climate models

53

Figure 2.2: Rainfall projections for Lilongwe, Malawi for 2050 and 2090 in two
scenarios from the Special Report on Emissions Scenarios of the
Intergovernmental Panel on Climate Change

75

Figure 2.3: Mean total annual rainfall in empirically generated rainfall scenarios

77

Figure 2.4: Mean rainy season end and start dates in empirically generated rainfall
scenarios
.

78

Figure 2.5: Mean modeled maize yields in empirically generated climate scenarios

79

Figure 2.6: Mean modeled pigeonpea yields in empirically generated climate scenarios

80

viii

Figure 2.7: Mean maize yields in empirically generated climate scenarios as fraction
of maize yield with baseline rainfall.

81

Figure 2.8: Percentage of years in which cumulative soil water stress in maize exceeds
ten in R75 group rainfall reduction climate scenarios

84

Figure 2.9: Mean cumulative N stress under increasing temperatures

86

Figure 2.10: Risk of yield reduction due to legume diversification in empirically
generated climate scenarios

88

ix

LIST OF ABBREVIATIONS

APSIM: Agricultural Production Systems Simulator
CO2: Carbon Dioxide
ha: Hectare
ISFM: Integrated soil fertility management
kg: Kilogram
MZ24: Maize with 24 kilograms of nitrogen per hectare fertilizer input
MZ92: Maize with 24 kilograms of nitrogen per hectare fertilizer input
N: Nitrogen
OC: Organic carbon
Rot24: Maize-pigeonpea rotation with 24 kilograms of nitrogen per hectare fertilizer
input applied to maize

x

LITERATURE REVIEW

Introduction
Improving agricultural productivity and food security in Africa has been a challenging
and seemingly intractable problem. While much of the rest of the world experienced rapid
increases in crop yields over the 1960s and 70s with the introduction of new varieties and
increased fertilizer use of the Green Revolution, African yields have stagnated (Sanchez, 2002).
Smallholder farmers face numerous constraints to production, including lack of access to
markets and credit, poor infrastructure and the resulting high cost and lack of access to chemical
inputs (Morris et al., 2007). Due to these constraints, many smallholder farmers are mining
nutrients from already degraded soils, and as a consequence yields are low and extremely
variable. This production environment will only grow more challenging as population growth
increases the pressure on already limited land, while climate change brings additional risk and, in
some areas, more marginal production environments for many common crops (Funk et al., 2008).
Sustainable intensification strategies are needed to improve smallholder productivity
while protecting natural resources. Strategies for sustainable intensification of cropping systems
are often placed in the contest of integrated soil fertility management (ISFM). ISFM techniques
are management practices, tailored to local conditions, that make efficient use of mineral and
organic sources of fertility and improved crop varieties to increase crop productivity (Vanlauwe
et al., 2010). Increased use of multipurpose legumes has been proposed as one strategy for
integrated soil fertility management in sub-Saharan agriculture (Snapp et al., 2010). Integrating
semi-perennial legumes has been shown to improve productivity and reduce maize yield
variability in specific cases. Because of the extremely variable soil and climate conditions found

1

in many parts of sub-Saharan Africa, it is vital to assess integrated soil fertility management
strategies in order to fit management practices to environmental conditions. Crop simulation
models are capable of using long weather records, soil characterization and information about
management practices to quantify interactions among these variables (Dixit et al., 2011).
Legume diversification for improved soil fertility
Integration of legumes into cropping systems is key to implementation of ISFM
techniques, because legumes fix atmospheric nitrogen and their residues provide some of this
fixed N to following cereal crops. A number of agroforestry strategies have been promoted, with
varying degrees of success (Garrity et al., 2010). Green manure legumes such as Mucuna puriens
have also had limited farmer adoption, with use mostly based on benefits other than soil
fertility—for example, Mucuna is used for weed suppression. Grain legumes in rotation also
have the potential to improve soil fertility if they can generate sufficient soil amelioration
through roots and vegetative cover that can be retained in the field. Pigeonpea is one legume that
shows potential as a dual use crop, producing both a high-protein grain and prolific leafy
biomass, some of which senesces during the growing season and can provide a source of
intermediate to high quality organic residue (Myaka et al., 2006). In Malawi, pigeonpea has
shown promise in a variety of soil types as a way to improve nutrient cycling, fertilizer use
efficiency and maize yields in long term countrywide and participatory trials (Snapp et al., 2010).
In multiple sites, use of pigeonpea in rotation increased N fertilizer use efficiency by 50-120%,
indicating significant positive impacts of the rotation system on soil fertility. Farmers
consistently rank cropping systems including pigeonpea highly in participatory trials, due to its
benefits for nutrition and food security as well as soil improvement (Bezner Kerr et al. 2007).

2

One concern in promoting increased use of long-duration, deep-rooted legumes is their
potential for increased water use and thus reduce yield of main food crops in the following year
(Wortmann et al., 2000). In semi-arid Zimbabwe, long-duration or high biomass producing
legumes depleted water in the deep soil profile, though in this sandy soil, the differences
vanished after the first major rainfall event (Ncube et al., 2007). In addition, drought stress can
decrease the capability of legumes to fix nitrogen, reducing their soil fertility benefits (Dakora
and Keya, 1997).
Performance of integrated soil fertility management strategies such as crop diversification
with legumes will be affected by soil type and rainfall distribution. Interactions among
management and environmental variables are difficult to demonstrate in field experiments,
because of the short duration of such experiments. Yield response of cereal-legume systems in
response to long-term climate variability and climate change has not yet been tested, and many
“outscaling” questions remain. Crop simulation modeling can be used to answer some of these
questions by extending the results of field experiments based on long-term climate records.
Crop models and the Agricultural Production Systems Simulator (APSIM)
APSIM was designed as a modular modeling framework for simulating a variety of
cropping systems. Biophysical crop and soil modules are linked by a central simulation engine
and controlled by management modules that allow the user to establish rules to control the
biophysical modules (Keating et al., 2003). The system runs on a daily time step and requires
daily climate data and a minimum set of soil data and crop variety information as inputs.
Management practices can be set to follow specific operations in the case where a specific trial is
being modeled, or to follow logical rules, as in the case of simulating management scenarios.
Crop modules

3

Crop modules in APSIM follow a set of routines based on physiological relationships
describing growth of plant components based on temperature, light, water, and nutrient
availability. Individual crop modules are stored as parameter files, which include crop and
variety-specific relationships.
APSIM uses the Plant module to simulate all legume crops, as well as several others.
This model uses a set of generic routines to simulate important plant growth parameters, as
described in Robertson et al. (2002). Phenology is determined by thermal time and photoperiod,
with plant growth described as transitions between a number of phases, separated by key stages
such as emergence, flowering, and maturity. Leaf development is modeled by combining
functions for leaf appearance and expansion, so that the potential leaf area is a combination of
number and maximum leaf size. The simulated leaf growth will be less than the potential leaf
growth only if carbon is limiting, as may occur at very high plant population densities. Biomass
production is predicted initially from the plant’s photosynthetic capacity as determined by its leaf
area and radiation use efficiency. This biomass is then partitioned into roots, stem, leaves, and
later to grain, based on experimentally determined ratios. Root biomass is converted to root
length and depth based on a specific root length parameter, as well as a soil parameter defining
any physical or chemical limitations that affect root penetration. Because some of the plants
simulated with the legume model are perennial, this module allows for harvest at a specified
height above the ground, with or without subsequent regrowth of the plant (Robertson et al.,
2002).
Nitrogen uptake is simulated as a combination of mass flow (nitrate entering as roots take
up water from soil solution), active uptake (diffusive uptake of nitrate and ammonia by roots),
and nitrogen fixation(Robertson et al., 2002). The crop’s demand for nitrogen is determined by

4

the minimum, optimal, and maximum N concentrations in its various tissues. Required N is
taken up first by mass flow, then by either active uptake or fixation, based on parameters that are
set for each crop. These can also vary based on growth phase. Thus juvenile plants whose roots
have not yet developed nodules may be less able to fix nitrogen. Deficits in nitrogen lead to a
reduction in growth when compared to the optimal case, limited only by photosynthetic
capability(Robertson et al., 2002).
Water deficits also slow modeled plant growth. Water demand by the crop is calculated
as a function of crop growth divided by transpiration efficiency. Soil water supply is calculated
using available soil water, calculated in soils modules described below, and the potential of roots
to extract this water. Water extraction occurs in layers where roots are present, at rates defined
by an extraction coefficient. If the soil water supply is below demand, a water stress factor is
used to reduce the modeled rate of plant growth from the photosynthetically limited rate.. Water
stress also reduces the ability of N-fixing crops to fix nitrogen, and the degree of the reduction
will vary by species (Robertson et al., 2002).
The pigeonpea module in APSIM is one instance of APSIM’s Plant module. It was tested
on 38 datasets taken from existing literature (Robertson et al., 2001) for extra-short, short, and
medium duration varieties. A long-duration cultivar was added later. The model showed
generally good agreement with field data, with a few limitations. Pigeonpea leaves tend to
senesce and drop late in the growing season, so that final biomass is difficult to measure and to
simulate properly. The model did capture water stress accurately, but it did not adequately
account for waterlogging. The data sets used for model validation were all obtained in central
India, but phenological responses have been shown to be accurate for Kenyan conditions as well

5

(Carberry et al., 2001). Ncube et al. (2009) modeled pigeonpea yields in Zimbabwe and found
modeled yields and biomass were within about one standard error from mean observed values.
The maize module in APSIM is similar to that in many other cropping systems models
and was based originally on CERES-Maize (Carberry, 1989). It simulates largely the same crop
growth characteristics, with the addition of maize-specific traits such as tasseling and silking. It
follows its own routines and not those of the generic Plant module because it was developed
earlier than other plant growth models. A key component of this module, of particular
importance for the current study, is its ability to simulate plant response to the low fertilizer rates
common in African smallholder farming systems. Shamudzarira and Robertson (2002) studied
this response in Zimbabwe and found that APSIM simulated grain yields at fertilizer levels
ranging from 0-60 kgN/ha to within one standard error of mean observed yields at Makoholi
Research Station in Zimbabwe.
Soil modules
Soil processes are modeled in several different modules. Soilwat, the water balance
model, uses a cascading layer model first developed for the CERES suite of crop models (Ritchie,
1972). Soils are input as a series of layers with set depths and volumetric water holding
capacities at key pressures. These are the lower limit (LL15, 1500 kPa), drained upper limit
(DUL, also known as field capacity, 33 kPa), and saturation. Water flow processes include runoff,
evaporation and transpiration, as well as both saturated and unsaturated vertical flow. The
module interfaces with a separate module, SurfaceOM, which calculates decomposition of
surface residues generated by the various crops, to account for the effects of these residues on
runoff and evaporation(Probert et al., 1998). This system has been shown to be very robust in

6

accurately modeling soil water content of different cropping systems over a range of soils and
environments (e.g. Ncube et al. 2009, Probert et al. 1998, Keating et al. 2003).
APSIM’s SOILN module is responsible for simulating both carbon and nitrogen cycling
in soil, since these are usually coupled (Probert et al., 1998). It treats soil organic matter as three
pools, plus an additional pool consisting of fresh organic matter (FOM). Residues incorporated
into the soil and decomposing roots first enter the FOM pool, where they are transformed into
either the rapid turnover microbial biomass (BIOM) pool or the slower turnover, less available
humic (HUM) pool. There is an additional component of inert OM, which represents highly
recalcitrant OM and has a turnover rate on the order of centuries. In APSIM, inert OM is
considered to be a fraction of the HUM pool which is not allowed to decompose (Probert et al.,
1998). Carbon to nitrogen ratios (C:N) are assumed to be constant in the BIOM and HUM pools.
The flow between pools is governed by carbon dynamics, with nitrogen released as a function of
changes in C:N ratios between pools. Mineral N produced by decomposition is available for
plant uptake or immobilization in soil. If insufficient N is present in FOM, the rate of
decomposition is slowed (Probert et al., 1998).
Modeling maize-legume cropping systems
Accurate modeling of crop growth, nutrient flows, decomposition and other processes is
key to simulating the effect of including legumes in the cropping system. Because of legumes’
ability to fix atmospheric N2, their residues tend to be enriched in N and some of this N may be
available to the subsequent maize crop. APSIM is capable of simulating these effects. This has
been shown for low-input smallholder farming systems in Africa, in the case of grain legumes
benefitting sorghum (Ncube et al., 2009) in Zimbabwe, and in the case of maize response to
previous velvet bean (Mucuna pruriens) green manures in Malawi (Robertson et al., 2005).

7

Ncube et al. (2009) report simulation modeling of a trial conducted at Matopos Research
Station in the semi-arid region of Zimbabwe, which tested several different types of grain
legumes for their productivity and their residual effect on sorghum. The experiment used two
varieties each of cowpea (Vigna unguiculata), groundnut (Arachis hypogaea), pigeonpea, and
Bambara groundnut (Vigna subterranea) in rotation with sorghum (Sorghum bicolor) (Ncube et
al., 2007). Phosphorus fertilizer was applied based on soil test results, but no inorganic nitrogen
was applied. Nitrogen fixed by legumes was measured by

15

N natural abundance and N

difference methods, and total biomass and yield were recorded. Soil water was measured by
neutron probe and gravimetric methods (Ncube et al., 2009). This study showed generally good
agreement between model and experiment for legume growth, N fixation, and water use. Water
use at the end of the growing season was underestimated, most likely due to the lack of weeds
and their water uptake. The simulated response of sorghum in rotation was accurate in predicting
total biomass, but underestimated grain yields, especially after pigeonpea. For sorghum grown
after pigeonpea, model yields were up to 30% below observed yields where pigeonpea residues
were incorporated. The authors attribute this discrepancy to the model not simulating pigeonpea
leaf fall during the season. Allowing for leaf fall improved the simulated sorghum yields, but
resulted in an under-prediction of pigeonpea biomass at the end of the growing season (Ncube et
al., 2009).
A study in Malawi by Robertson et al. (2005) tested the response of maize to nitrogen
from a Mucuna pruriens green manure and developed a Mucuna module in APSIM. The Mucuna
module was developed from the same general plant template used for pigeonpea and other
legumes, and APSIM was used to simulate growth of the green manure itself as well as the effect
of the green manure on subsequent maize yields. A calibration field trial was established in five

8

sites and over two growing seasons from 1998-2000. At all five sites, maize was planted in the
1998-99 growing season with nitrogen fertilizer levels of 0 and 69 kgN/h, and Mucuna was
planted in adjacent plots. In the second year, the previous maize plots were planted with the same
fertilizer levels. The Mucuna plots were planted to maize with 0 and 34.5 kgN/ha. An additional
fertilizer response trial was established at Chitedze, one of the five sites, in 1997-98. In the
fertilizer response portion of the trial, fertilizer rates ranged from 0 to 92 kgN/ha applied as urea.
(Robertson et al., 2005).
Field results showed variation in response to both inorganic fertilizer and the green
manure legume. The highest responses were seen in soils with low initial inorganic nitrogen and
low organic carbon content. When fertilizer was applied to maize following Mucuna, the
response to the green manure was only significant at the lowest fertility soils, which had organic
carbon contents of less than 0.7%. APSIM simulated maize yields well over the range of
fertilizer rates and fertilizer-green manure combinations. Thus the legume model is able to
generate the appropriate quantities and qualities of biomass, the soil modules are processing it,
and the maize module is responsive to this type of N input in appropriate ways. The authors
reported very high variability in maize response to green manure legumes in field trials based on
soil, crop management, and climate factors and note the importance of using simulation
modeling to better identify the possible yield responses in a variety of environments.
Climate variability and risk in rainfed agriculture
Agriculture in sub-Saharan Africa is predominantly rainfed. Especially in drier areas,
variable rainfall leads to a great deal of variation in yields from season to season, making
agriculture inherently risky, particularly for farmers whose livelihood and food security depend
on their crop production. Understanding the degree of climate risk and developing improved

9

strategies for coping with this risk are important in agricultural research and extension in these
areas.
The simplest way to quantify climate-related risk is by a statistical analysis of rainfall
records. Where long-term records exist, valuable insights can be gained from analysis of various
components of these data. Stern and Cooper (2011) report several methods for analyzing rainfall
data in a case study from Zambia. These include identification of successful planting dates,
length of growing season, and risks of dry spells at various key points in crop growth. These
types of analyses can be much more informative than simple annual or monthly totals, since dry
spells at key crop development stages can cause low yields or crop failures in a year when total
rainfall appears adequate.
Crop models can expand upon these types of analyses by accounting for interactions
among climate, soil, and crop. They also allow the evaluation of cropping systems or
management strategies as methods for coping with variability. One key issue that can be
addressed by models is the targeting of fertilizer recommendations, which are often based on
maximizing potential crop yield and do not accurately account for the economic realities of
smallholder farmers (Marenya and Barrett, 2009). These farmers have limited resources to spend
on fertilizer and also tend to be highly risk-averse. Because of this they tend not to apply
fertilizers at recommended rates. In some areas, fertilizer recommendations have been developed
that specifically consider these issues, and crop models can be used to test these
recommendations for risk and overall profitability.
In Zimbabwe, M.I. Piha (1993) developed a system of variable fertilizer application for
maize based on rainfall amounts early in the season. This system, which was tested in trials at
three sites and over three growing seasons, included low application rates of N at planting

10

followed by variable top-dressing rates of between 0 and 150 kgN/ha applied at 4 and 6 weeks
after planting and at maize tasseling. The rates varied based on the location of the trial site (those
located in drier regions received lower fertilizer rates) and amount of rainfall up to the
application date. In this study, yields and economic return to fertilizer were significantly higher
under the variable-N system than under fixed N application at the recommended rates.
Shamudzarira and Robertson (2002) investigated the long-term productivity and profitability of
maize production using a similar approach, with low and variable N application in simulations of
fertilizer response using 46 years of daily climate data. They tested both fixed and variable N
applications at rates ranging from 0 to 60 kgN total. At these low rates of fertilizer use,
conditional application of fertilizer resulted in lower levels of fertilizer application, but overall
similar yield and economic returns, with conditional application resulting in slightly lower
economic return to fertilizer. The model captured the risk distribution in fertilizer response, and
that this response was only correlated to rainfall within a range of 250 to 550 mm of rainfall over
the season (Shamudzarira and Robertson, 2002). Below this level, crops failed regardless of
fertilizer, and above it yields tended to plateau. This trend was also observed in field trials. These
results indicate that establishing a viable fertilizer recommendation that accounts for resource
constraints and risk is more important than varying application rates by year.
Crop models can also be used to extend the results of short-term field trials by accounting
for long-term variability in high productivity areas. Dixit et al. (2011) used APSIM to simulate
maize fertilizer response trials to establish a profitable rate of fertilizer over a 50-year simulation
dataset in Kitale, Kenya. They used extensive data from variety trials conducted in 1977 at the
National Agricultural Research Station in Kitale to calibrate APSIM, and this calibrated model
produced yield results which agreed to within 5% of observed yields. This study also validated

11

and used the stochastic weather generator MarkSim (Jones and Thornton, 2000) to create longterm sequences of representative rainfall and temperature data where these are unavailable.
MarkSim uses climate surfaces interpolated from existing weather stations to create sequences of
temperature and rainfall that are statistically similar to the long-term patterns at a given location.
This tool is potentially very useful for other studies in sub-Saharan Africa, where weather data,
particularly the types of long-term records needed to evaluate risk, are often missing. The Kitale
study first demonstrated the validity of the MarkSim generated weather by comparison to a
shorter existing climate station record, then used these data to evaluate probability distributions
of maize yield at various fertilizer levels and weed control treatments. They found that given the
high yield potentials in this area, even very high levels of fertilizer can be economically justified.
-1

For this analysis they use current maize and nitrogen fertilizer prices of $2.86 kg for N and
-1

$0.252 kg for maize. In this case nitrogen applications of 150 kg/ha applied as 30 kgN at
planting and 120 kgN at 5 weeks post emergence produced a rate of return of at least 3 in 9 out
of 10 years.
Climate change
As concern over the potential impacts of climate change increases, strategies for coping
with year-to-year climate variability are seen as key components of long-term adaptation to
climate change (Cooper et al., 2008). Climate risk management strategies are important both
because they suggest longer-term adaptation possibilities and because climate variability is likely
to increase in the future (Christensen et al., 2007).
Studying potential adaptation strategies is complicated by the fact that climate predictions
for many parts of sub-Saharan Africa are highly uncertain. In many parts of eastern and southern
Africa, General Circulation Models (GCMs) do not agree on the direction of rainfall change,

12

much less the magnitude (Christensen et al., 2007). This uncertainty is compounded by the need
to downscale GCM projections from their typical 100-300 km resolution to account for local
variation (Winkler et al., 2011). This can be done in a variety of different ways. Dynamic
downscaling uses regional climate models similar to GCMs, applied over a smaller area, to
generate predictions at finer resolution. These models are provided with boundary conditions
from GCMs. This process is computationally intensive, and variation among RCMs is as large as
that among GCMs (Winkler et al., 2011). Statistical, or empirical, downscaling uses regression
analysis to relate current climate variables found in historical weather records to modeled climate
variables from GCMs. Although this method is less rigorous because it does not consider the
dynamic relationships within the climate system, well-performed statistical downscaling has
been shown to produce similar results to RCM outputs (Wood et al., 2004). All these methods
rely on extensive historical data for parameterization, which is lacking in much of the developing
world (Jones et al., 2005).
Given the uncertainty surrounding climate change projections, one possibility is to adopt
a sensitivity analysis approach. This can be done by varying rainfall and temperature in ways
consistent with the range of possibilities for future climate and evaluating potential management
strategies based on the range of conditions under which the management practices perform well.
The sensitivity analysis approach was adopted in a report by scientists from the International
Crops Research Institute for the Semi-Arid Tropics (ICRISAT) (Cooper et al., 2009). This study
evaluated the impacts of 1° and 3° C increases in temperature as well as a 10% reduction in
rainfall on maize production in Makindu, Kenya and Bulawayo, Zimbabwe, and on groundnut
production in Kasungu, Malawi. Weather inputs for this study were generated by modifying
historical data by the specified levels. While this process is relatively straightforward for

13

temperature, there are a number of different ways to generate a 10% reduction in rainfall amount.
For this study the authors reduced the amount of rain on each rainy day by 10%, leaving the
overall distribution of rainfall events intact. This method is likely to underestimate the effects on
agricultural production of an actual 10% reduction in rainfall, since some of the reduction will
come on days with extremely high rainfall totals and therefore will only reduce runoff.
Nevertheless, the effects of this simulated climate change were significant: reductions in rainfall,
coupled with increased evapotranspiration due to increased temperatures, led to reductions in the
length of potential growing season, by 5-10% in Makindu for example. Additionally, yield
reductions may be expected because increased temperatures lead to more rapid plant growth and
thus lower yields. Despite these negative potential impacts, Cooper et al. (2009) title their paper
a “Hypothesis of Hope.” They point out first that while climate change is projected to reduce
potential yields, actual farm yields in sub-Saharan Africa are well below this potential level, and
second, that by using existing germplasm, such as longer-duration groundnut varieties in the case
of Kasungu, and improved management practices to conserve soil and water, yields can be
increased substantially above current levels.
Another method for using models to investigate the impact of climate change is to couple
downscaled global climate models or regional-scale models to GIS systems and crop models to
make regional predictions of changes in productivity. This approach was taken by Thornton et al.
(2009) to examine changes in maize and bean production in six countries in East Africa. They
used the DSSAT cropping systems model (Jones et al., 2003), with climate projections from two
GCMs, and with representative soil profiles from publicly-available databases, to produce
representative estimates of crop production under a 2050 climate scenario.

14

This spatially explicit approach allows evaluation of climate change impacts on a
regional level. Thornton et al. (2009) show site-specific differences for maize production in this
region. While some areas will suffer from decreased rainfall and higher than optimum
temperatures for maize production, other, particularly highland areas, benefit from increasing
temperatures. In some parts of East Africa, rainfall is expected to increase, resulting in yield
improvements. Bean production follows a similar pattern, but beans show greater yield reduction
as a result of high temperatures. This large-scale view omits some of the nuances of local
production, however, and the computing power needed to run models over a large number of
pixels in these spatial simulations does not allow evaluation of complex management strategies.
Case studies in specific areas can complement this broader view. While some case studies have
been conducted (Thornton et al., 2010), integrated soil fertility management strategies such as
crop diversification with legumes have not been rigorously tested under future climate scenarios
in Eastern and Southern Africa. These need to be explored as researchers strive to find strategies
for improving smallholder crop production that will be resilient to future climate change.

Objectives
1. Assessing the impact of climate variability on maize-legume cropping systems in Malawi
Climate in Malawi is both heterogeneous in space and variable over time (Ngongondo et al.,
2011). Participatory field studies over three years at one site show the promise of diversified
cropping systems including the long-duration legume pigeonpea. Crop simulation modeling will
be used to extrapolate from these studies to a second site, and to evaluate risk associated with
climate variability in these systems,
2. Exploring the response of maize-legume cropping systems to climate change

15

Because of the uncertainty in climate change projections for Malawi we follow a sensitivity
analysis framework similar to that of Cooper et al. (2009) to evaluate the impact of climate
change on soil and crop processes in maize-pigeonpea systems. In this study, distribution of
rainfall as well as amount will be adjusted, allowing for changes in start or end date of the rainy
season to result in varying lengths of season, changes in number of rainy days within a rainy
season of similar length and changes in daily rainfall amounts.
3. Analysis of trade-offs in a changing climate
Trade-offs related to use of long-duration legumes under current and future climate variability
will be quantified. Trade-offs between soil moisture reductions and increased nitrogen
availability through N fixation are expected, particularly in drier conditions.

16

CHAPTER ONE
Extending participatory research results through simulation modeling: Benefits, risk and
trade-offs to using long-duration legumes in Malawi
Abstract
Integrated soil fertility management strategies for intensifying crop production in sub-Saharan
Africa have shown promise for improving smallholder yields. In Malawi, participatory research
methods have identified pigeonpea (Cajanus cajan) as a promising, multipurpose legume for
improving fertility management in maize (Zea mays) systems. Including a long-duration legume
like pigeonpea creates a trade-off between improved soil fertility and increased water stress.
Benefits and risk in these cropping systems will depend on soil type and rainfall regime. This
effect is difficult to quantify using short-term field trials, particularly in the context of
participatory research where variability is high. The current study provides an example of the
integration of modeling and participatory research methods by parameterizing the Agricultural
Production Systems Simulator (APSIM) with data from on-farm trials to evaluate maizepigeonpea rotation and intercrop systems. Model experiments at two sites—Zombwe and
Kasungu—show that at low fertilization levels (24 kgN/ha applied to maize) both intercrop and
rotation systems outperform continuous maize. While soil water and nitrogen trade-offs exist,
particularly at Kasungu, the drier of the two sites, the overall effect is positive. Increased time
from establishment increases beneficial effects of both systems, and mean effects are greater in
sandy, low fertility soils than in finer-textured, higher fertility soils. Risk of negative effect from
pigeonpea is less than 10% in all cases for the rotation system. Risk in intercrops is high in the
initial establishment phase (10-30 %) and drops below 10% after 3 years. These results suggest

17

that maize-pigeonpea systems are viable, low-risk options for improving crop production and
soil fertility, and can be targeted to low fertility soils to obtain the maximum benefit.
1.1 Introduction
In the past decades, agricultural research has led to new strategies showing great potential
to increase yields for smallholder farmers in sub-Saharan Africa, but yields on the continent have
stagnated (Sanchez, 2010). Lack of adoption may be due to promotion of inappropriate
technologies, or lack of capacity in extension systems to disseminate new information effectively.
Participatory research methods have been developed to mitigate some of these issues, and have
been used in areas including plant breeding and testing of new varieties, as well as soil fertility
and natural resource management. Methods such as mother-baby trials (Snapp et al., 2002) are
used to create opportunities for co-learning, where researchers and farmers can collaborate to
achieve research objectives. As a result of engaged research, integrated soil fertility management
strategies have been developed which incorporate fertilizer and organic technologies to
maximize nutrient use efficiency and crop productivity (Vanlauwe et al., 2010).
Simulation modeling has been used successfully in concert with field experimentation in
an agricultural development context to extrapolate findings and assess technology performance
(e.g. Gowing et al. 2003; Chikowo et al. 2008). Models have also been used to match integrated
soil fertility management practices to appropriate agro-ecological niches defined by climate and
soil type (Giller et al., 2010). Crop models can be used as a tool for engagement with farmers, or
as a learning tool for policymakers. A review of these uses is found in Whitbread et al. (2010).
Despite the clear value for inference over time and space, use of crop simulations has rarely been
linked to participatory on-farm research (Dimes et al., 2003). There are considerable challenges
to model applications in this context, due to variability of the natural resource and management

18

environment, as well as multiple non-modeled effects on yield, such as pest and disease pressure
(Robertson et al. 2005). This complexity also creates difficulty when trying to identify
underlying mechanisms from field research because of the multiple sources of variability in such
systems. Simulation modeling can be used to isolate and quantify variability due to soil and
climate variation.
The Agricultural Production Systems Simulator (APSIM) is a modular cropping systems
model that has been developed for simulation of a wide variety of cropping systems. Modules for
diverse crops including maize, pigeonpea, soybean, and groundnut have been developed, as well
as modules for simulating various soil processes including water and soil carbon dynamics, and
phosphorous and nitrogen budgets (Keating et al. 2003). The model has been used extensively in
Eastern and Southern Africa to simulate crop growth and development under both research
station and smallholder conditions, and has been found to perform well, including for simulating
response to nitrogen fertilizer in Zimbabwe (Shamudzarira and Robertson, 2002), and rotational
effects with green manure legumes in Malawi (Robertson et al., 2005). APSIM has also been
used to conduct risk analyses, using longitudinal climate records to test management strategies.
Dixit et al. (2010) used APSIM to evaluate maize response to fertilizer and economic returns
over many years for a high-productivity site in Kenya. Shamudzarira and Robertson (2002)
explored the combined effects of N and water in a semi-arid environment, and evaluate risk for
different fertilization scenarios.
APSIM, like other crop simulation models, requires high-resolution data for
parameterization, and obtaining this data in the context of on-farm trials and participatory
research is difficult. Ideal methods for characterizing soil profile properties require extensive infield measurements (Dalgliesh et al., 2009). Crop models also require daily rainfall and

19

temperature data, and while methods exist for infilling missing data (Pickering et al., 1994) or
stochastic generation of daily data (Jones and Thornton, 2000), complete weather station data is
much preferred. A challenge for integrating modeling and participatory research is to determine
the true minimum data set required for adequate confidence in model results.
Integrated soil fertility management strategies including legumes such as green manures
and agroforestry tree species, show promise for soil fertility improvement. In northern Malawi,
participatory trials have identified diversification of maize cropping with the long-duration
legume pigeonpea (Cajanus cajan) as an integrated soil fertility management system that is
farmer preferred and shows promise for improving soil fertility (Snapp et al. 2010). Farmers
have highly rated these systems, including pigeonpea in rotation with maize and a pigeonpeagroundnut (Arachis hypogaea) intercrop rotated with maize. The benefits noted include food
during off-season periods, high value protein-enriched products, and improving maize yields in
the following year (Bezner Kerr et al., 2007; Mhango et al., 2012). However, including a longduration and deep-rooted legume like pigeonpea or an agroforestry tree crop in a rotation
increases water use and thus has the potential to reduce yield of main food crops in the following
year (Wortmann et al. 2000). In addition, drought stress can decrease the capability of legumes to
fix nitrogen, reducing their soil fertility benefits (Dakora and Keya 1997).
Based on these relationships, trade-offs are to be expected between water use and soil
fertility gains. The extent and intensity of trade-offs have only begun to be investigated, and are
expected to be influenced by soil type and rainfall distribution. Soils in Malawi are spatially
heterogeneous, even over short distances (Snapp, 1998). Variability in soils, combined with the
multitude of confounding factors inevitably found in on-farm research, means that identifying
and quantifying these trade-offs in participatory field studies is nearly impossible. Linking

20

participatory research to modeling could markedly enhance understanding of processes
underlying interactions among soil type, climate, and crop yields. Better understanding of
interactions would permit extrapolation of findings. For example, characterizing the soil and
climate conditions that support long-duration legumes could play a critical role in identification
of areas appropriate for expansion of legume diversification. Evaluation of locations where longduration legumes may be deployed must be based on risk of crop failure as well as potential
benefits.
Most areas of Malawi show high temporal variability in rainfall (Ngongondo et al., 2011),
making risk of crop failure due to drought difficult to predict using short-term field trials. Models
can also be used to explore potential long-term impacts of changes in management. Modeling
studies usually reset soil parameters such as organic carbon, inorganic nitrogen, and soil water to
their initial values yearly or biannually to prevent carry-over errors that would reduce accuracy.
This limits the scope of such studies to effects measurable over one or two years. By using
simulation models run over longer continuous time series without reset, it is possible to test for
longer-term consequences of changes in management practices (Carberry et al., 2002).
Clearly, there is need to understand the risks, trade-offs, and long term consequences of
diversification of mazie-based systems with long-duration legume species such as pigeonpea.
The objectives of this study are to:
1. Improve linkages between participatory research and modeling through assessing the
feasibility of using data acquired through participatory on-farm research to parameterize APSIM
2. Quantify the vulnerability of rainfed maize and maize-pigeonpea systems to climate-related
risk over time given climatic and soil constraints representative of smallholder farming in
Malawi

21

3. Evaluate trade-offs between soil fertility benefits and increased soil water use in maizepigeonpea diversified systems across soil types and rainfall regimes.
1.2 Materials and Methods
1.2.1 Field sites
Data for model parameterization were obtained from field trials conducted in Ekwendeni,
Mzimba district, in northern Malawi, during the three growing seasons between 2007 and 2011
(Mhango, 2011). Yield and biomass data, as well as soils data including organic carbon content,
soil texture, and pH in surface and subsurface layers were collected from 19 farmer-managed
field sites in 5 villages. Soil types at the field sites were generally sand to sandy clay textured
with low levels of organic matter, classified as fine kaolinitic, thermic, typic kandiustalfs
(Mhango et al., 2012). Rotation and fertility treatments used in modeling consisted of:
continuous maize with two nitrogen fertilization levels (24 and 92 kg N/ha); a pigeonpea-maize
rotation with 24 kg N/ha applied only to maize; and a maize-pigeonpea intercrop, with 24
kgN/ha. Nitrogen was applied as urea at 4 weeks after planting (WAP) in the 24 kg N/ha
treatments. In the 92 kg N/ha treatment urea was applied in a split application with 23 kg N/ha
applied at planting and 69 kg N/ha applied at 4 WAP. The maize variety was MH-17, and the
pigeonpea variety was ICEAP 00040.
Field sites at each farm constitute single replications for the purpose of field experiments,
and many non-modeled factors may have influenced yield, including pest pressure, large-scale
soil erosion, and livestock (Mhango, 2011). Therefore, when modeling the results it is unrealistic
to attempt to precisely duplicate yield results at each farm by tailoring soil parameters
individually. Importantly, soils varied among farms, and we are interested in understanding the
effects of this variation, so it was also not appropriate to use a single soil type. Using data

22

collected during baseline soil sampling in 2007 (Mhango, 2011), field sites were divided into
three groups (Figure 1.1) by texture and organic carbon content, and soil types for modeling
were created to represent each group.
Soil characterization for model initiation required additional data, collected as follows.
Four field sites representative of the three soil groups were selected for additional sampling for
inorganic N, as well as deep soil sampling for organic carbon and bulk density (Figure 1.1).
Additional sampling was conducted at these sites in mid-July 2011. At each site two plots were
selected, one plot that had been planted to a maize-pigeonpea rotation system, and a second plot
with three sole maize sub-plots which had been fertilized at 0, 24, and 92 kgN/ha in the previous
year. Five samples were taken from the maize-pigeonpea plot, and three samples were taken
from each of the three sole maize sub-plots. Samples were collected using a 6.4 cm soil augur, to
a maximum depth of 120 cm,at intervals of 0-15, 15-30, 30-60, 60-90, and 90-120 cm, where
each depth interval was composited.. Samples were well mixed, air-dried and sieved to pass a
2mm sieve, and analyzed at Michigan State University in the Snapp Laboratory. Nitrate (NO3)N was extracted from upper layers (0-15 and 15-30 cm) using 1M KCl and measured
spectroscopically following the method of Doane and Horwath (2003). Samples for organic
carbon analysis, which were taken at all depths, were ground to powder and analyzed for total
organic carbon using a dry combustion C and N Analyzer (Costech ECS 4010, Costech
Analytical Technologies, Valencia, CA). Organic carbon values were averaged for each field site,
including both maize-pigeonpea rotation and sole maize plots, since our data confirmed previous
analysis showing no significant difference in total OC content between plots (Mhango 2011).
Samples were also collected for textural analysis but the resulting data was inconsistent and
could not be used. At the same time, 7.6 cm cores for bulk density measurements were taken at

23

each field site. Two pits were dug at each site and cores were taken from two faces of each pit.
Pits were located adjacent to trial plots and dug to a depth of 1.2 m. Four cores were taken at
depths of 0-15, 15-30, 30-60, 60-90, and 90-120 cm and averaged by depth.
1.2.2 Model Setup
The crop simulation model APSIM was used to model all treatments. APSIM requires inputs of
daily weather data, including maximum and minimum temperature, precipitation, and solar
radiation. For model validation, daily rainfall data (1945-2011) and long-term average monthly
temperature data were available from the Zombwe Extension Planning Area, approximately 5
km from the field sites. Weatherman, a part of the DSSAT software suite (Pickering et al., 1994),
was used to generate daily temperatures and solar radiation, as well as to fill some gaps in the
rainfall record. Missing values in rainfall records were concentrated during dry seasons, so
should have limited effect on model results. Zombwe (11.33°S, 33.82°E, altitude 143 m.a.s.l.)
has an average of 783 mm of rain per year, with a standard deviation of 203 mm. The model was
also run using a long-term record (1927-2010) from Kasungu, Malawi (13.03°S, 22.45°E,
altitude 1036 m.a.s.l.), also infilled using Weatherman, to investigate the applicability of this
system to a slightly drier site in a different part of Malawi. Kasungu has an average of 739 mm
of rain per year, with standard deviation 247.
The soils modules in APSIM require soil moisture content at permanent wilting point,
field capacity, and saturation, bulk density, organic carbon content, and pH. From field data,
three APSIM soil input files were created to represent the range of variability in soil types
present in field experiments. Organic carbon from deep samples collected in 2011 was averaged
by farm and depth, and two of these average profiles were selected for use as organic carbon
inputs for modeling. Water holding capacity parameters were derived from texture and bulk

24

density using pedotransfer functions following Saxton and Rawls (2006). This method was
selected based on evaluations of a number of common pedotransfer functions by Gijsman et al.
(2002). Their analysis found that an earlier version of Saxton and Rawls’ method (Saxton et al.,
1986) performed the best out of the eight methods they tested. The updated version used here is
similar. Bulk density and organic carbon measurements described above from 2011 sampling
were used. Texture at depths from 0-30 cm was calculated as the average within each of the three
soil groups (Figure 1.1) defined from previous (2007) data (Mhango, 2011). The highest-fertility
soil (Soil HF) is 0.78% organic carbon (OC) 63% sand, 27% clay in the topmost 15 cm. The
medium-fertility soil (Soil MF) has similar texture but 0.58% OC in the topsoil, and the lowfertility, sandy soil (Soil LF) has the same OC content as soil two, but has a higher (79%) sand
content. Data for soil texture below 30 cm was not available because of difficulties with
processing of deeper soil samples, so properties below this were assumed to be constant,
following Chikowo et al. (2008). Field examination of soils did not suggest any major changes in
soil texture at depth.
There were two types of model runs used: an evaluation model, and a time-series model.
The evaluation model was created to mimic five of the experimental treatments for the years
2009-2010. The time-series model used the full length of weather records from Zombwe (66
years) and Kasungu (83 years), and was run without reset for ten-year time series.
Evaluation model
The evaluation model includes simulations of continuous maize controls at 24 and 92 kg
N/ha as in the field trials. A pigeonpea-maize rotation system with 24 kgN/ha applied in maize
years only was simulated for both entry points, to provide annual yields for both maize and
pigeonpea. A maize-pigeonpea intercrop was also simulated at 24 kgN/ha yearly. Simulations

25

used maize cultivar MH 17, already parameterized in APSIM, at a plant density of 3.7 plants/m2.
The built-in long-duration pigeonpea cultivar showed satisfactory performance, and given the
lack of available calibration data it was used without modification. This was planted at lower
density (2.2 plants/m2) to account for high plant mortality and insect damage in field experiments.
Soil nitrogen, organic carbon, surface residues and water content were re-initialized on
July 15, every year in the case of the continuous maize and intercropping systems to eliminate
year to year carryover errors and every two years (after maize harvest and prior to pigeonpea
planting) for the rotation system, in order to accurately model the maize yield response to
pigeonpea in the previous year. Surface organic matter was set to 300 kg/ha maize residues,
consistent with farmer practice. Initial inorganic nitrogen (as nitrate) was set to the average of
measured values for fields that had been previously planted to unfertilized maize, then adjusted
slightly, both to better match yields in the continuous maize systems and to prevent large jumps
in soil N at reset, which would bias comparisons between rotations (reset every two years) and
other systems (reset yearly). This simulation was run using weather data for Zombwe, for the
years 2008-2010.
Time series model
The second set of simulations was identical to the first in overall setup but was run for
both Kasungu and Zombwe, for the entire duration of each weather data set: 83 years at Kasungu
and 66 years at Zombwe. These simulations were run over ten year time series and not reset each
year. This allowed novel insights into the medium to long-term impacts of using a diversified
system. In order to maintain the high level of replication possible with the long weather data
records available, the simulations were duplicated ten times, with start dates in sequential years
such that each year from system establishment was represented in each calendar year (see Figure

26

1.2 for a schematic representation). Residue management was performed separately for each
crop: 80% of maize residues were removed to avoid early-season immobilization of N which
otherwise substantially reduced yields, while pigeonpea residues were incorporated. The same
soil profiles were used for both sites. These profiles are consistent with soil properties measured
on smallholder fields in the Kasungu area (Phiri et al 2010), though soils there tend to be most
similar to the lower fertility, sandier type (Soil LF).
1.2.3 Analysis of model results
Analysis was conducted using the statistical software R (R Development Core Team, 2011).
Models were fitted using the built-in linear modeling function lm(), which calculates standard
model diagnostics for linear fixed-effects models. Because the objective of the data analysis here
is largely exploratory, we focus on trends in the model results that illustrate underlying
mechanisms. To simplify the analysis, the time since reset, representing time from establishment
of the system, was grouped into first-year, early (years 2-4), mid (years 5-7) and late (years 8-10)
periods. Analysis excluded the first year, because no rotation effect is expected before year two.
1.3. Results
1.3.1 Evaluation of model performance
Results in the evaluation model were generally within one standard deviation of
measured maize yields and biomass (Figure 1.3). In most treatments modeled yields are above
the experimental mean yield. This slight over-prediction of on-farm yields is to be expected,
since we are comparing model results, which do not account for pest and disease pressure, to onfarm yields, where these have significant impacts. Model yield is notably low in 0N continuous
maize treatments, but shows good agreement in all treatments considered in the time series
simulations.

27

Pigeonpea yield and biomass were overestimated by the model, however for
indeterminate long-duration pigeonpea such as that used in these trials accurate estimation of
total biomass and yield is difficult because of early leaf senescence. This, combined with high
levels of pest pressure, means that it is probable that field-measured yield and biomass are lower
than would be expected when considering only model-simulated effects. Mean measured
pigeonpea yields in 2008 and 2009 ranged from 119 to 266 kg/ha, with a maximum reported
yield of 953 kg/ha in the rotation system in 2009. Mean biomass ranged from 806 to 3307 kg/ha,
with a maximum of 7040 kg/ha, again in the rotation system in 2009 (Mhango, unpublished data).
Modeled yields ranged from 263 to 698 kg/ha, and biomass ranged from 3016 to 6981 kg/ha.
These modeled yields and biomass, are consistent with results reported elsewhere (Myaka et al.,
2006).
1.3.2 Systems including pigeonpea show higher yields than continuous maize systems.
Results of the time series model showed that rotation systems outperformed continuous
maize on all soil types and at both Kasungu and Zombwe (Figure 1.4). The intercrop resulted in
lower maize yields than those in continuous maize in the most fertile soil type (Soil HF) in the
early period in Zombwe. For the same soil type and period at Kasungu the intercrop produced
similar yields to continuous maize. In all other periods, and for the other soil types, maize yields
in the intercrop were better than those in continuous maize. All systems improved in productivity
over the course of the 10 years between resets, with most of the increase coming within the first
five years. Continuous maize yields increased slightly with time, because soil conditions in year
one represent soil status following unfertilized continuous maize. Repeated fertilizer inputs and
small additions of organic matter result in improved yields over time. Yield increases between
periods were greatest in intercropping systems, while rotation systems provided the highest yield

28

overall in maize years. The greatest differences between continuous maize and maize-legume
systems was observed in the least fertile soil (Soil LF).
At Zombwe, maize yield in intercrop systems was positively correlated with the previous
year’s pigeonpea biomass throughout the 10-year period from establishment. Rotation systems
show a more complex relationship, with interactions between soil type and establishment period.
Early in the 10-year cycle (2-4 years from establishment), effects were positive and significant
(at p < 0.05) for Soil HF, but insignificant for the other two soils. In later years, there was no
significant effect in Soil HF but soils MF and LF showed small negative trends. At Kasungu the
effect of pigeonpea biomass was minimal, and all regression r2 values were less than 0.1.
1.3.3 Risk and yield distribution change with cropping system and over time
Yield distributions are wider for maize-legume systems than for continuous maize, and
wider for intercrops than for rotations (Figure 1.5). In the early establishment period and in
higher fertility soils, the lowest yields of maize-legume systems are below those of continuous
maize, and their distributions are considerably wider. Once the systems are established, their
yield variability decreases and mean yield increases. Yield variability is higher at Kasungu than
at Zombwe in all cases. The effect of the legume component and the probability of decreased
yield can be calculated by comparing maize yields in rotation and intercrop to that of continuous
maize for each soil type and year from establishment. Yield effects are substantial, ranging up to
150% increases over continuous maize yields (Figure 1.6). Effects are most significant in Soil LF,
the sandy, low fertility soil. The probability of decreased yield is low, especially in mid and late
establishment periods (Figure 1.7). Intercrops are riskiest, particularly in the period of early
establishment, where maize yield is reduced in 10-30% of years at both sites. This risk is reduced
dramatically with time, so that in the 3-5 year period, only 4% of years show reduced yields in

29

intercrops. Risk in rotation is never more than 3.1% at Zombwe, and 6.7% at Kasungu, making
this a more stable system than the intercrop.
1.3.4 Nitrogen effects of legumes
In all systems, yield is limited by nitrogen availability, as expected for low-level fertilizer
applications in relatively low-fertility soils. Legume systems increase the amount of nitrogen
available to associated maize crops, particularly in rotation treatments. This can be seen from the
relative amounts of inorganic N in soil at planting (Figure 1.8), although some of the nitrogen
inputs from legume residues are not represented as they are in organic pools.
APSIM calculates a daily value of nitrogen stress for plant processes including
photosynthesis, biomass expansion, and grain fill. These values range from zero (no stress) to
one. Summing these values over the growing season provides a measure of total nitrogen stress
of the crop. For this analysis we use N stress for photosynthesis, since photosynthesis will occur
throughout the growth period. The effect of N stress on yield is most clear at Zombwe, the wetter
of the two sites, where nitrogen stress is clearly the dominant predictor of yield (Figure 1.9).
Reduction in nitrogen stress accounts for most of the yield increase in both intercrop and rotation
systems in Zombwe. Yield increase is calculated as a percentage of continuous maize yield by
the formula:

(1) Yield Effect =(Yield of maize-legume system — Yield of continuous maize)/(Yield of
continuous maize)

Reduction in N stress is calculated similarly, using model-calculated cumulative N stress for
photosynthesis over the growing season and the formula

30

(2) N Effect = (N stress in continuous maize —N stress in maize-legume system)/( N stress in
continuous maize)

Where either yield or N stress was zero in continuous maize, values were excluded. This
transformation isolates the nitrogen effects on maize due to the inclusion of pigeonpea, and at
Zombwe, it provides a clear picture of the effect (Figure 1.10). Remaining variability when
nitrogen stress effects are accounted for is small.
A linear model including period from establishment, soil type, and N effect as
independent variables was fitted separately to each of the rotation and intercrop yield effects data
at Zombwe. Period from establishment (grouped as above into early, mid, and late) and soil type
were considered fixed factors, with N effect as a covariate. A simple 3-way analysis of variance
showed that main effects and interactions were significant at the p<0.05 level, so we subset the
data by period and soil type. A simple linear regression of yield effect by N effect shows that
both the slope of the regression and the r2 value decrease with increasing time from establishment,
with these effects most pronounced in the lower-fertility soils. The overall magnitude of the N
effect also increases with increasing time from establishment.
The effect of N stress reduction in Kasungu is less pronounced. A similar model of yield
and N stress effects produces many non-significant slopes, and r2 values of less than 0.15 in all
cases. It is clear that while in some cases nitrogen is limiting yields in Kasungu, other factors are
also important determinants of maize yield.
With long-term use of legume systems, improvements in soil fertility will be expected.
Both intercrop and rotation systems increased organic carbon content in soils, particularly in the

31

upper 15cm. At Zombwe, in this layer, the organic carbon content in soil HF increased from
0.78% to 0.89% in the intercrop system and 0.86% in the rotation. Organic carbon in soils MF
and LF increased from 0.58% to 0.66% in rotation in both soils, and to 0.70% and 0.71% in
intercrop for soils MF and LF, respectively. Soil carbon in the continuous maize system
remained approximately constant in all soils. Results were similar at Kasungu.
1.3.5 Water stress effects and impacts on yield
Crop water stress was much more important at Kasungu than at Zombwe. In Zombwe
there were so few instances of high soil water stress that it is impossible to draw conclusions.
However, the drier conditions and longer record at Kasungu provided enough instances of
substantial water stress to allow for analysis. At this site there is a clear effect of rainfall. As
reported elsewhere (Shamudzarira and Robertson, 2002), this effect is most pronounced at
rainfall amounts below 550 mm (Figure 1.11). Cumulative soil water stress for photosynthesis
was calculated in the same manner as cumulative N stress described above. In cases where
cumulative soil water stress was greater than one, there is a clear relationship between soil water
stress and yield (Figure 1.12). Rainfall amounts in these stressed years range up to 743 mm
(above the mean value over all years at Kasungu of 739 mm), with a mean value of 478 mm.
Both intercrop and rotation treatments are more stressed than continuous maize (Figure 1.13).
Soil water stress levels increase with time from establishment, particularly in the “mid” period,
as nitrogen stress declines and yield increases, causing plants to use more water.
The effect of water stress is most pronounced at critical times in crop development. Using
model-identified crop stages, soil moisture content was examined early in the growing season
(from emergence through flag leaf stage) and near flowering (floral initiation and flowering
stages). For years with low rainfall (less than 550 mm during the growing season), mean

32

extractable soil water early in the growing season and at flowering show some correlation with
yield, though the relationship does not appear to be linear (Figure 1.14). There seems to be a
threshold effect, where extremely low soil water content results in a high probability of crop
failure.
Soil water status is influenced by both soil and treatment, with soil HF having generally
higher extractable soil water. Maize-legume systems have lower soil moisture content than
continuous maize systems, especially early in the season. Soil water at flowering is much closer
among systems, as soil water recharge occurs with early season rainfall (Figure 1.15).
In Zombwe, overall rainfall amount was not well correlated with maize yields: even for
years with rainfall below 550mm, correlations were not good. Relationships between yield and
soil water at critical crop stages were also unclear. Yields were, however, correlated with dry
spells of 10 days or longer during the growing season, for all rainfall amounts in both rotation
and intercrop systems. In contrast, continuous maize yields showed no correlation with dry spells.
This effect was not observed at Kasungu—in fact the number of 10-day dry spells was correlated
with a slight increase in yield.
Pigeonpea, as a long-duration crop, is more adversely affected than maize by low rainfall
and shortened rainy seasons, in both Zombwe and Kasungu. The effect of dry spells observed at
Zombwe for maize yields is not seen for pigeonpea.
1.4 Discussion
1.4.1 Integrating modeling with participatory research
APSIM was able to realistically simulate both rotation and intercrop systems, producing
maize and pigeonpea yields similar to those observed in participatory field trials in Northern
Malawi and reported elsewhere (e.g. Myaka et al. 2006, Høgh-Jensen et al. 2007, Mwale et al.

33

2011). While the field data used in this study do not provide validation of long-term effects,
other work has confirmed the ability of APSIM’s components to simulate long-term soil
processes (Probert et al., 1998). We have found that with limited additional soil sampling and
processing, the data collected from participatory trials was sufficient to adequately parameterize
APSIM. Some compromises were required due to limited parameterization data. A full
characterization of soil water retention parameters in the field was not possible. Texture data was
not available for deep soil layers, so following Chikowo et al. (2008) we used the values from
15-30 cm for deeper layers. While these compromises mean the precision of the resulting model
is not perfect, it does allow us to address our research objectives, and can serve as a tool for
targeting further field research.
1.4.2 Maize-legume systems increase maize yield with low risk of failure.
Overall, both intercrop and rotation systems outperformed the continuous maize control
treatment (Figure 1.3). As expected, we do see a trade-off between reduced soil moisture and
increased nitrogen availability to crops, and this leads to higher variability in maize-legume
systems. However, in nearly all years the balance of impacts from including pigeonpea is
positive. This confirms results of field studies in Malawi showing the benefits of pigeonpea and
other long-duration legume systems for increasing productivity of maize (Snapp et al., 2010).
The use of pigeonpea in rotation is relatively low-risk, particularly in Zombwe where moisture
constraints are lowest (Figure 1.6). It is rare for yields in maize-pigeonpea rotation systems to be
lower than those of continuous maize systems in the same year. Risk in rotation systems is
slightly increased in Kasungu, because of increased water stress, and risk remains elevated over
time. Intercrops are more risky, particularly in early establishment years. Mean maize yields in
intercrops are lower than maize following pigeonpea in rotation. However, because a rotation

34

will only produce maize in alternate years, it may be less attractive to farmers than intercrop
systems, depending on the value placed on the legume crop, and the amount of land available to
satisfy maize food requirements.
There are clear differences in maize yield response to legumes due to differences in soil
type. The greatest yield benefit is seen in less fertile, sandy soils, where small increases in
available nitrogen, soil cover and organic carbon are most beneficial. Yields under maize-legume
systems at Zombwe after 10 years were the same across all three soil types (Figure 1.3), despite
continuing differences in continuous maize yields. At Kasungu differences by soil type persist,
particularly between soils HF and LF, the finest and coarsest-textured soil. These differences are
most likely caused by the higher soil water stress in the sandier soil at Kasungu.
1.4.3 Soil water and nitrogen trade-offs
It is clear that a trade-off exists between increased available nitrogen and higher soil
water stress in legume systems. However, in both sites discussed here, the effects of increased
nitrogen greatly outweigh the increases in soil water stress. The two factors combined explain
nearly all of the variation in yield, particularly in the rotation. At Zombwe the nitrogen effect is
clearly the dominant predictor of yield (Figure 1.8). When yields occurring in water-stressed
years are removed at Kasungu, N stress becomes similarly dominant. There is a larger amount of
remaining variability in the intercrop, some of which is most likely caused by competition for
light between maize and pigeonpea.
1.4.4 Long-term impacts of maize-legume systems
The potential long-term yield increases from maize-legume systems are considerably
greater than the benefits seen in the first year. This long-term effect is difficult for farmers to
consider, given the high discount rates they apply to future yields (Waddington and Karigwindi,

35

2001). Long-term impacts on soil carbon are difficult to measure in on-farm trials, because the
huge variability in these soils disguises any changes over short time frames. Despite the potential
models have for contributing to understanding these soil processes, most modeling studies reset
soil variables yearly. This is due to the possibility for compounding errors in nitrogen, carbon
and water from year to year. While longer duration continuous (non-reset) model runs may result
in less precise yield measurements, they allow modeling to contribute to addressing concerns
about long-term impact. Further modeling efforts, especially when combined with long-term
field experiments, would provide important additional information.
1.5 Conclusions
The approach used here to integrate modeling with participatory research shows promise
for scaling out results of such research and for generating new insights. Our results show that
maize-pigeonpea cropping systems are viable, low-risk options for improving crop production
and soil fertility with minimal reliance on external inputs. Soil moisture requirements of the
intercrop systems posed some risk of maize yield reduction, but only for the initial years of
adoption. Trade-offs between increased nitrogen availability and decreased soil moisture may
pose a barrier to adoption at sites drier than those considered here. Pigeonpea can be targeted to
low fertility soils to obtain the maximum benefit, and has potential to ameliorate challenges to
production in these soils.

36

Figure 1.1: Soil characteristics at field sites, topsoil (0-15 cm) depth. Colors indicate groups used
to generate the three soil types used in modeling. HF=high fertility, MF=medium fertility,
LF=low fertility. For interpretation of the references to color in this and all other figures, the
reader is referred to the electronic version of this thesis.

37

Figure 1.2 Schematic depiction of time series model setup. Simulations are run continuously for
ten years, after which soil parameters are reset. Model is run ten times (only three runs are shown
here) with staggered start dates to maintain high levels of replication.

38

Table 1.1: Properties used in APSIM soils input files. See text for field data used in soil
characterization.
Wilting
Field
Saturated
Bulk
Depth
pH Organic
capacity
water
carbon
point
density
(cm)
content
(mm/mm)
(%)
(mm/mm)
(mm/mm)
Soil HF

0.24

0.41

1.50

6.0

0.78

0.18

0.28

0.42

1.47

5.5

0.60

30-60

0.18

0.28

0.42

1.47

5.5

0.42

60-90

0.18

0.28

0.42

1.47

5.5

0.23

90-120

0.18

0.28

0.42

1.50

5.5

0.19

0-15

0.14

0.22

0.44

1.50

6.0

0.58

15-30

0.13

0.21

0.40

1.60

6.0

0.38

30-60

0.13

0.20

0.38

1.70

6.0

0.27

60-90

0.13

0.20

0.35

1.70

6.2

0.20

90-120

0.13

0.20

0.35

1.70

6.5

0.20

0-15

0.076

0.13

0.41

1.53

6.0

0.58

15-30

0.11

0.17

0.40

1.50

6.0

0.38

30-60

0.11

0.17

0.40

1.51

6.0

0.27

60-90

0.11

0.17

0.35

1.60

6.0

0.20

90-120

Soil LF

0.15

15-30

Soil MF

0-15

0.11

0.17

0.35

1.60

6.0

0.20

Properties in all soils
finert
fbiom

Initial NO3
(kg/ha)

0.6

0.03

1.0

0.7

0.02

1.0

0.8

0.02

1.5

0.9

0.01

0.5

0.95

0.01

0.25

39

Figure 1.3: Comparison of field-measured data and results from simulation model used for
measured
evaluation. Means reported for maize biomass and grain yield from 2009-2010 field
2010
experimentation at Ekwendeni, Malawi (columns Error bars are standard deviations from the
,
(columns). rror
mean. Where columns are missing, field data was not available. Points represent model results
from simulations run on three soil types. Treatments are described in methods.
a. Maize grain yield

b. Maize biomass

40

Figure 1.4: Mean maize yields for 10 years from system establishment for three cropping
systems in three soil types. Time series model results averaged over 83 years at Kasungu and 66
years at Zombwe, northern Malawi. Error bars are standard errors of the mean over the same
time period. Variation is based on differences in weather from year to year. Soil types defined in
methods. HF=high fertility, MF=medium fertility, LF=low fertility.

41

Figure 1.5: Cumulative probability distributions of maize yields for three cropping systems in three soil types. Time series model
results from simulations of 83 years at Kasungu and 66 years at Zombwe, northern Malawi. Results plotted separately for each soil
type and period from establishment. Periods early, mid, and late refer to years from system establishment: 2-4, 5-7, and 8-10
respectively. Soil types defined in methods. HF=high fertility, MF=medium fertility, LF=low fertility.

42

Figure 1.5 (cont’d)

43

Figure 1.6: Mean effect of legume systems on maize yields as percent increase over continuous
maize yield. Continuous maize yield represented as zero. Time series model results over 83 years
at Kasungu and 66 years at Zombwe, northern Malawi. Periods early, mid, and late refer to time
from system establishment: 2-4, 5-7, and 8-10 years respectively. Error bars are standard errors
of the mean. Soil types defined in methods. HF=high fertility, MF=medium fertility, LF=low
fertility.

44

Figure 1.7: Risk of yield reduction in maize-legume systems, as percentage of years where maize
yield in maize-legume system is lower than in the continuous maize system.. Time series model
results from simulations of 83 years at Kasungu and 66 years at Zombwe, northern Malawi.
Periods early, mid, and late refer to time from system establishment: 2-4, 5-7, and 8-10 years
respectively. Soil types defined in methods. HF=high fertility, MF=medium fertility, LF=low
fertility.

45

Figure 1.8: Inorganic N in soil at planting for three cropping systems in three soil types. Time
series model results from simulations of 83 years at Kasungu and 66 years at Zombwe, northern
Malawi. Periods early, mid, and late refer to time from system establishment: 2-4, 5-7, and 8-10
years respectively. Soil types defined in methods. HF=high fertility, MF=medium fertility,
LF=low fertility.

46

Figure 1.9: Maize yield and cumulative nitrogen stress in three cropping systems in three soil
types. Cumulative N stress calculated from model factor, see results for more detailed
description. Time series model results from simulations of 83 years at Kasungu and 66 years at
Zombwe, northern Malawi. Periods early, mid, and late refer to time from system establishment:
2-4, 5-7, and 8-10 years respectively. Soil types defined in methods. HF=high fertility,
MF=medium fertility, LF=low fertility.

47

Figure 1.10. Effect of legume diversification on maize yields and cumulative nitrogen stress relative to continuous maize at the same
fertility level. Continuous maize yield is zero. Negative N stress reduction (x axis) values correspond to increases in nitrogen stress.
See Results for further information. Time series model results from simulations of 83 years at Kasungu and 66 years at Zombwe,
northern Malawi. Results displayed are from early period (years 2-4 from system establishment). Regression lines are significant at
2
p<0.05 and shaded region represents standard error of the regression. r values for regression lines range from 0.65 to 0.87. Soil types
defined in methods. HF=high fertility, MF=medium fertility, LF=low fertility.

48

Figure 1.11: Maize yields and in-crop rainfall for three cropping systems in three soil types.
Time series model results from simulations of 83 years at Kasungu and 66 years at Zombwe,
northern Malawi. Results are restricted to in-crop rainfall below 550 mm. Lines represent linear
regression with intercept forced to zero and are significant at p<0.05. Shaded area represents
2
standard error of the regression. r values for regression lines range from 0.82 to 0.95. Soil types
defined in methods. HF=high fertility, MF=medium fertility, LF=low fertility.

49

Figure 1.12: Maize yields and cumulative water stress for three cropping systems in three soil
types. Time series model results from simulations of 83 years at Kasungu and 66 years at
Zombwe, northern Malawi. Points with cumulative soil water stress less than one are not
displayed. Results are from mid period, 5-7 years from system establishment. Soil types defined
in methods. HF=high fertility, MF=medium fertility, LF=low fertility.

50

Figure 1.13: Mean soil water stress for three cropping systems in three soil types. Time series
model results averaged over 83 years in Kasungu, northern Malawi. Error bars show standard
errors of the mean. Periods early, mid, and late refer to time from system establishment: 2-4, 5-7,
and 8-10 years respectively. Soil types defined in methods. HF=high fertility, MF=medium
fertility, LF=low fertility.

51

Figure 1.14: Maize yield and extractable soil water at critical plant growth stages for three cropping systems in three soil types. Time
series model results from simulations of 83 years in Kasungu, northern Malawi. Results displayed are from mid period (3-5 years
from system establishment). Points shown restricted to years with in-crop rainfall less than 550mm. Soil types defined in methods.
HF=high fertility, MF=medium fertility, LF=low fertility.

52

Figure 1.15: Mean extractable soil water at critical plant growth stages for three cropping systems in three soil types. Time series
model results from simulations of 83 years in Kasungu, northern Malawi. Results displayed are from mid (3-5 year) period from
establishment. Error bars are standard errors of the mean. Soil types defined in methods. HF=high fertility, MF=medium fertility,
LF=low fertility.

53

CHAPTER TWO
Modeling maize-legume systems under variable climate scenarios in Northern Malawi
Abstract
Climate change has the potential to impact food security in sub-Saharan Africa, but details about
future weather patterns, particularly rainfall, are uncertain. We develop methods to create climate
change scenarios by combining a variety of rainfall scenarios with increases in temperature. This
allows for exploration of alternative possible future climates, and facilitates greater participation
in the research process, since scenarios generated are transparent and easy to understand. We use
scenarios including temperature increases of up to 4 °C and rainfall changes of +/-10% and +/25%, with different rainfall distributions to evaluate a promising rotation system of maize and
the long duration legume pigeonpea. Its performance compares favorably under future climate
scenarios to continuous maize at 24 kgN/ha and 92 kgN/ha. Maize in rotation outperforms
continuous maize at the same level of fertility, and shows smaller yield declines at high
temperatures than high fertility maize for all the rainfall scenarios tested.
2.1 Introduction
Increasing awareness of the potential impacts of climate change on crop production and
food security in sub-Saharan Africa have led to a number of efforts to quantify and understand
these effects (e.g. Thornton et al., 2010; Doherty et al., 2010; Twomlow et al., 2008). It is clear
that agronomic recommendations must take into consideration the potential response of the
proposed management and technological systems to a changing climate.
The 2007 report of the Intergovernmental Panel on Climate Change (IPCC) outlines the
broad-scale consequences that will likely result from changes in climate. In sub-Saharan Africa,
the report forecasts that by the end of the 21st century, temperatures will rise by over 3 °C.

54

Projected rainfall changes are more variable (Christensen et al., 2007). Both global and regional
climate models (Thornton et al., 2009) project precipitation increases in East Africa and
decreases during some portions of the year in Southern Africa (Christensen et al., 2007). In order
to evaluate the impact of climate change on a local scale, the large-scale projections provided by
global climate models must be downscaled to account for local variability in topography, and to
translate long-term changes in climate to changes in short-term weather patterns (Winkler et al.,
2011). This process is complicated in sub-Saharan Africa because of a relative scarcity of highquality and long-term weather station data.
In addition to the uncertainty of climate predictions, evaluating the outputs from these
models is difficult. For many scientists, and more so for farmers and extensionists, the models
can seem like “black boxes” whose outputs must either be taken at face value or completely
rejected. This dichotomy can be especially problematic when farmers’ perceptions of climate
change are different from those predicted by the models. Farmer surveys (e.g. Rao et al., 2011,
Thomas et al., 2007, Mertz et al., 2009) report common beliefs that rainfall is decreasing or
becoming more variable. While these beliefs may not be confirmed by meteorological data (Rao
et al., 2011), they do influence farmers’ thinking and may contribute to adoption or non-adoption
of technologies.
One way to engage stakeholders in climate change impact analysis is by developing
potential future scenarios and modeling their impacts (Swart et al., 2004). Developing climate
change scenarios that can be easily understood without detailed knowledge of climate science
could be a tool for more engaged research on the subject. Using these types of scenarios along
with crop models can allow testing of current technologies as well as potential adaptation
strategies. Such simulations can also be used to explore interactions among factors such as CO2

55

concentration, temperature, and rainfall (Ludwig and Asseng, 2006). A scenario analysis
approach was used by a group of researchers from the International Crops Research Institute for
the Semi-Arid Tropics (ICRISAT) (Cooper et al., 2009), who evaluated the impacts of 1° and 3°
C increases in temperature as well as a 10% reduction in rainfall on maize production in
Makindu, Kenya, and Bulawayo, Zimbabwe, and on groundnut production in Kasungu, Malawi.
Weather inputs for the Cooper study, as for Ludwig and Asseng’s study on wheat in Australia,
were generated by modifying historical data on each day: increasing temperatures and reducing
each day’s rainfall amount by 10%. For systems with high input use, this resulted in significant
decreases in yield, but for low-input systems the effects were minimal. The effect of temperature
on yield was more significant than that of rainfall, due to the combination of increased
evaporation and more rapid phenological development. This led to alterations in duration of
growth and timing of maturation (Cooper et al., 2009). The simple perturbation method used for
changing rainfall did not account for potential changes in distribution that could lead to different
degrees of impact. Rainfall distribution can have marked impacts on crop and soil response and
these impacts should be further explored (Stern and Cooper, 2011).
Malawi, as it is located in a transitional area between East and Southern Africa, has an
extremely uncertain climate future. Novel cropping systems including the long-duration legume
pigeonpea (Cajanus cajan) have been widely adopted in areas of northern Malawi, particularly
around Ekwendeni, Mzimba district (Snapp et al., 2010). Participatory action research methods
have been key to the development and adoption of these systems, and farmers and researchers
are interested in expanding this effort to evaluate the response of legume-diversified systems to
climate change. In this area, model agreement among GCMs is not good (Christiansen et al.
2007), and downscaled climate forecasting systems fail to accurately estimate current conditions.

56

The Climate Information Portal of the Climate Systems Analysis Group at the University of
Cape Town, South Africa (Climate Systems Analysis Group, 2012) has made available
statistically downscaled results for a number of locations including Lilongwe, in Central Malawi.
At this location all climate models, when used to generate baseline rainfall, overestimated both
numbers of rainy days and total rainfall amounts (Figure 2.1) when compared to observed data.
Model projections for precipitation in Lilongwe in 2090 range from a 23% reduction to a 13%
increase in the A2 scenario, with the majority of models projecting slight decreases (Figure 2.2).
There are slightly smaller deviations in the B1 scenario. This degree of uncertainty, along with
low confidence in model accuracy given difficulties in representing current conditions, does not
provide useful guidance to those interested in developing adaptation strategies. Farmers,
meanwhile, believe that rainfall is becoming more erratic and that rainy seasons are starting later
(Bezner Kerr, personal communication).
The Agricultural Production Systems Simulator (APSIM) has been extensively tested in
Eastern and Southern Africa, and is able to simulate low-fertility smallholder conditions as well
as grain yield response to legumes (Robertson et al. 2005, Shamudzarira and Robertson 2002). It
includes modules for crops including maize and pigeonpea, as well as soils modules for
simulating processes including water, carbon, and nitrogen (N) dynamics (Keating et al., 2003).
APSIM was used previously at the same Malawi sites to explore the effects of current climate
variability on yields in smallholder systems (Chapter 1). That study showed that systems
including pigeonpea had higher yields than continuous maize systems, but some trade-offs exist
between the nitrogen benefits from an N-fixing legume, and increased soil water stress from a
longer-duration crop. Given these trade-offs, it is important to understand the potential risk for
crop failure in a warmer and potentially drier climate.

57

The objectives of the current study are to develop a realistic assessment of climate change
impacts on smallholder maize-based cropping systems through the following:
1. Develop methods for creating climate change scenarios by varying the amount and distribution
of rainfall in a historical record.
2. Use APSIM to evaluate the performance of smallholder maize and maize-legume cropping
systems under climate change scenarios in northern Malawi
3. Determine the degree to which increases in temperature and changes in rainfall increase soil
water stress and impact nitrogen availability in maize-based cropping systems.
2.2 Materials and Methods:
2.2.1 Generation of climate scenarios from current weather data
Rainfall has two basic characteristics: amount, and distribution. The distribution of rainfall —
the number of rainy days, number and distribution of dry spells, length of rainy season — is
often more critical to crop production than the total amount of rainfall (Barron et al., 2003).
Therefore, changes in both amount and distribution should be considered when examining the
sensitivity of a cropping system to changes in rainfall. To do this we have developed a set of
functions for use with the R statistical software package (R Development Core Team, 2011)
which add, remove, or redistribute rainfall in various ways.
Changing rainfall distribution by removing rainy days
This is accomplished by a set of functions which remove a percentage of existing rainfall from a
user-selected fraction of rainy days or rainy spells. Rainy days are defined by a minimum
threshold rainfall amount, with default of 0.85 mm. The existing rainfall record provides, in each
year, a stochastic rainfall pattern, so the removal is deterministic: a count is kept of rainy days
and the nth day is removed each time. It is possible to vary the days selected by varying the start

58

of the count, to check for artificially inflated or reduced rainfall totals based on single large
events. Rainy spells can be of length one or greater. For spells, the count value is increased by
one on the first day with rain, where the previous day is dry, and thus remains the same
throughout the spell.
A secondary set of functions redistributes removed rainfall, by dividing it evenly among
all remaining rainy days or on a fraction of rainy days, to reach a specified goal percentage of
original rainfall. This percentage can be either greater than or less than the original total. By
restricting these functions to a date-defined fraction of the full data set, rainfall changes can be
constrained to calendar months or seasons.
Changing season length
Adjusting growing season length requires first identifying the start and end of the rainy
season. This is done according to criteria that can be set by the user. Stern and Cooper (2010)
define the first period of three days with greater than 20mm total rainfall as the beginning of the
rainy season. They define the end of the rainy season as the first date with less than 10mm total
rainfall in the preceding 10 days and no rainfall in the following 20 days as the end of the rainy
season. We use these definitions. The search domain for these dates will depend on the seasonal
rainfall patterns in the area: for Malawi we use November through January for the start date, and
April through June for the end date. Once start and end dates are defined, rainfall removal and
redistribution functions can be applied around these dates in order to shorten or extend the rainy
season. This can be done either by counting rainy days from start or end dates, or by creating a
new data structure by extracting a period of days to be used as input to the global rainfall
redistribution functions.

59

All functions are designed to be applied over one year, which can be defined by the
calendar year, or by creating a harvest year variable that begins and ends in accordance with
growing seasons. Creating a harvest year variable is helpful for cases like that of Southern Africa
where the growing season begins in one calendar year and ends in the next. Packages previously
developed for R such as plyr (Wickham, 2011) allow functions to be applied individually to
years within a longer data set, or simultaneously to records from multiple stations, streamlining
the creation of model-usable weather data files (Script is available as supplementary material to
the electronic version of this thesis or by contacting the author at mollenburger@gmail.com)
2.2.2 Weather data used
The two sites used for this study are Zombwe (11.33°S, 33.82°E, altitude 143 m.a.s.l.), near
Ekwendeni, and Kasungu (13.03°S, 22.45°E, altitude 1036 m.a.s.l.). The data available had long
daily rainfall records, from 1945-2011 in Zombwe, and from 1927-2010 in Kasungu
Temperature records in Kasungu were available from 1987, in Zombwe only long-term monthly
mean temperatures were available. Weatherman, in the DSSAT 4.0 application suite (Pickering
et al., 1994) was used to fill in missing data in both records and to generate solar radiation
estimates. Annual average temperature at Zombwe was 20.5 °C, and at Kasungu 21.7 °C; mean
annual rainfall is 783 mm/yr at Zombwe and 739 mm/yr at Kasungu.
2.2.3 Scenarios used in simulations
The rainfall scenarios used here were chosen to represent the range seen in climate model
projections for the area. While some models project large percentage changes in rainfall during
the dry season (June through September), rainfall amounts during these months are very low (see
Figures 2.1 and 2.2). Overall, the projected changes in rainfall amounts are within 25% of
current rainfall, so scenarios were developed within this range. A variety of distributions were

60

tested for each rainfall amount, to examine the sensitivity of yield to differing changes in
distribution. Rainfall records were modified in several ways. First, a set of moderate rainfall
reduction scenarios with rainfall reductions of about 10% were generated (Group R90). Second,
a group of increased-rainfall scenarios were generated, with increases of 10% and 25% (Group
R+). Third, a set of worst-case rainfall reduction scenarios were generated. These included 25%
reductions in rainfall across the growing season, as well as reductions targeted to critical earlyseason periods also leading to approximately 25% reductions in total rainfall (Group R75).
Rainfall removal patterns in each scenario are described in Table 1. Each rainfall scenario was
simulated for temperature changes in increments of 1° from 0°C to 4°C increase, and at three
CO2 concentration levels: 350, 500, and 700 ppm. The APSIM baseline CO2 concentration is
350 ppm. Temperature increases of approximately 2 °C are predicted by 2050, and increase of
between 3 and 4 °C are predicted by the end of the century. CO2 concentrations of 500 and
700ppm are consistent with the IPCC A1b scenario for 2050 and 2090 (Christensen et al., 2007).
2.2.4 Simulation setup
Simulations were designed to follow the experimental setup described in Chapter 1. Treatments
simulated were continuous maize with the addition of fertilizer at two levels, 24 and 92 kgN/ha,
and a maize-pigeonpea rotation with 24 kgN/ha applied to maize, simulated with both maize and
pigeonpea entry points to provide a maize yield value in each year. Fertilizer was applied as urea,
at 4 WAP in the treatments with 24kgN/ha, and in the treatment with 92 kgN/ha as a split
application with 24 kgN/ha at planting, 68 kgN/ha at 4 WAP. Simulations were run for the length
of the rainfall record at each site: 66 years at Zombwe and 83 year at Kasungu. Soil carbon and
nitrogen in these simulations were reset on July 15, before planting pigeonpea in the case of the
rotation system, and yearly in all other treatments. Soil water was allowed to carry over between

61

seasons. Planting was set to occur on the first instance between November 15 and January 30
where there is greater than 20mm of rainfall over 3 days, with a minimum soil moisture of 15
mm. This is a common planting condition for the area (Stern and Cooper, 2011). 85% of maize
residues were removed, following farmer practice and to avoid immobilization of soil N by
residues. Pigeonpea residues were incorporated. Each treatment was simulated for two soil types
which are representative of smallholder conditions (Mhango et al., 2012). Soil HF has relatively
higher fertility and finer texture, with 63% sand, 27% clay with 0.78 % organic carbon (OC) in
the topsoil layer, while Soil LF has lower fertility and coarser texture, with 79% sand and 14 %
clay, and has 0.57% OC in the topmost layer. This model was found to simulate yields in onfarm participatory trials at Zombwe to within one standard deviation of the mean. Details of
model parameterization and evaluation can be found in Chapter 1.
2.3. Results:
2.3.1 Rainfall in climate change scenarios
Mean total rainfall (Figure 2.3) in the scenarios tested ranged from 537-926 mm at Kasungu, and
from 578-984 mm at Zombwe, compared to averages of 740 mm and 787 mm respectively in the
base rainfall scenarios. Season start and end dates varied with scenario: even for scenarios where
early-season rainfall was not targeted, start dates tended to be later in reduced rainfall scenarios,
and end dates tended to be earlier (Figure 2.4). These effects were small, less than one week in
all cases. Season start dates are similar at the two sites, near December 1 in base rainfall
scenarios. End dates differed by nearly three weeks between sites, from April 7 in Kasungu to
April 25 at Zombwe.

62

2.3.2 Yield response to increases in CO2 concentration
Maize response to CO2 concentration is minimal. Changes in mean yield with the change
from 350 ppm to 700 ppm are essentially zero for low N maize (continuous maize with 24
kgN/ha). At Kasungu, response is between 20 and 150 kg/ha for high N maize (continuous maize
with 92 kgN/ha), while at Zombwe it is essentially zero. The response is negative at both sites
for rotation, with yield reductions of up to 1000 kg/ha. The negative result in rotation is likely a
result of very high CO2 response for pigeonpea, where yields increase up to 35% in some cases.
When examining overall relationships between maize and pigeonpea there appears to be a
threshold pigeonpea yield of about 600-650 kg/ha above which maize yields begin to decline.
The simulated CO2 response in pigeonpea is very high at both sites. A shift from 350
ppm to 500 ppm resulted in, on average, a 17% increase in yields, and an increase from 350 to
700ppm resulted in a 35% increase. These values are much higher than those observed in
chamber experiments (Saha et al., 2012), which reported a maximum increase of 12% from CO2
fertilization effects at 580 ppm. Clearly, the model requires further adjustments to accurately
simulate CO2 effects on pigeonpea. Since the maize yield response is so low and the pigeonpea
response is questionable, CO2 effect is not considered in further discussion, and figures show
results at baseline CO2 levels
2.3.3 Yield response to increases in temperature
Yield responses to temperature vary by treatment, site, and soil type (Figure 2.5). Rotation with
pigeonpea improves maize yields relative to continuous maize at the same fertility level for all
climate scenarios. At Zombwe, very high pigeonpea yields in the high fertility soil seem to be

63

correlated with lower maize yields in that soil than in the low fertility soil. Low N (24 kgN/ha)
continuous maize shows very little response to temperature at either site, remaining essentially
constant. High N (92 kgN/ha) continuous maize yields respond differently by site and soil types.
Yield decreases are greatest in soil HF, the soil with higher fertility and water holding capacity.
In Kasungu, yields in this soil are nearly constant with temperature increases up to 2 °C. Above
this, yields fall more rapidly reaching levels 15% below baseline yields at +4°C. In Zombwe, all
temperature increases result in substantial yield decreases in high N maize in soil HF, with a
39% decrease at +4°C . Yields in the low-fertility soil are nearly constant at Kasungu, and
decrease 17% at Zombwe. At both sites maize yields in rotation increase slightly with
temperature increases of 1-2 °C and decrease at higher temperatures. At both sites yields in
rotation remain at or slightly above baseline levels even with a 4°C increase in temperature.
Pigeonpea yields increase at Zombwe with 1°C temperature increases and then decline,
returning to near baseline at a 3 °C temperature increase with baseline rainfall (Figure 2.6).
Declines begin immediately and are more pronounced at Kasungu, which has a baseline mean
temperature 1°C higher than Zombwe. Here yields decline by about 25% at a 4 °C temperature
increase with baseline rainfall. One cause of lower yields at higher temperatures is the decrease
in length of the growing season. At Zombwe, season length is reduced by approximately 25 days
for maize and 74 days for pigeonpea with a 4 °C temperature increase. At Kasungu, for the base
rainfall scenario, season length was reduced by approximately 20 days for maize, and 45 days for
pigeonpea. There was more variation in season length with scenario type at Kasungu, especially
in the low fertility soil (Soil LF).

64

2.3.4 Differences in yield response in rainfall scenarios
Overall, maize yield response to altered rainfall amount and distribution was minimal in
in the low N maize system and highest in the rotation system. Yield response was largest at
Kasungu, the more arid of the two sites. Figure 2.5 shows the effect of rainfall changes, grouped
by the percentage of base rainfall: these are most notable at Kasungu in soil LF, the sandy and
low-fertility soil type. Yield responses by individual rainfall distribution scenario are shown in
figure 2.7. Here yields are shown relative to the simulated yield under baseline rainfall. Yield in
each rainfall scenario is divided by the yield under baseline rainfall for each soil type, treatment,
and temperature increase simulated. Thus yield with baseline rainfall is equal to one. This
isolates the effect of rainfall from the effect of temperature, and allows for comparison of this
effect across treatments and soil types. For rainfall scenarios with 90% of baseline rainfall (R90),
the additional effect of changes in rainfall was within +/- 5% of yield with baseline rainfall,
except for the A scenario at Zombwe, which resulted in increases over baseline of 12%. With a
25% reduction in rainfall (R75 scenarios), yields in rotation at Kasungu were as much as 23%
below yields at baseline rainfall, while at Zombwe yields remained within 15%. High N
continuous maize had maximum yield reductions in R75 scenarios of 16% at Kasungu, and 9%
at Zombwe. The largest impacts in both rotation and high N maize occur for scenarios with yield
reductions focused in the first 60 days of the season. The impact of reduced rainfall was greatest
in the sandier soil (Soil LF).
Rainfall increases did not always result in increased yields relative to baseline rainfall.
For most rainfall and temperature combinations at Zombwe, impact of increased rainfall was
negative, resulting in a maximum 10% reduction in scenario BB+, where 25% rainfall increases
are concentrated to one fifth of rainy days. At Kasungu, yields increased in increased-rain

65

scenarios for rotation systems and for high N continuous maize in the higher fertility soil, but
were reduced for the high N maize system in the sandier soil, particularly at low temperatures. In
some A-type cases, where rainfall amount is adjusted on each rainy day, there are spikes in
relative yield measurements at 3°C temperature increases: from figure 2.5 it is clear that these
spikes correspond not to actual yield increases but to smaller decreases than in baseline rainfall
scenarios. Overall, the combined effect of rainfall and temperature changes is to decrease
differences between treatments and between soil types.
Differences in yields among rainfall distribution scenarios at the same total rainfall level
were small. When only 10% of rainfall was removed, differences in yields were within 2-3%
among different scenarios. At higher rainfall removal rates, differences were most notable
between scenarios that covered the full growing season and scenarios focused on early months.
The most extreme case is found in the sandy soil type at Kasungu with 4°C temperature increase.
Here, maize yields in rotation yielded 10% less in removal scenario F (where half of rainy days
were removed over the first 60 days of the season) than in removal scenario AA (where 25% of
rain was removed on each rainy day throughout the year).
Pigeonpea yields follow similar trends, but with slightly larger impacts on yield than in
maize (Figure 2.6). Pigeonpea yields are reduced by up to 26% in low-rainfall scenarios at
Kasungu, and increase up to 17% above yields in base rainfall in increased rainfall scenarios.
Pigeonpea yields at Zombwe were more affected by rainfall, with yield reductions of up to 35%,
but from a higher baseline yield.
2.3.5 Water and nitrogen stress in maize
Water stress in maize increases with both rising temperature and declining rainfall,
contributing to the yield reductions seen in such scenarios (Figure 2.8). APSIM calculates a daily

66

soil water stress value ranging from zero to one, for processes including phenology,
photosynthesis, and biomass expansion. We use the value calculated for photosynthesis, summed
over the full growing season, as a measure of water stress in the crop. Rotation and high-N maize
systems show the greatest effect of soil water stress. Figure 2.8 shows the percentage of years,
for the worst-case R75 scenarios, in which cumulative water stress is greater than 10: at this level
yield reductions due to water stress are about 10%. Yields decrease linearly so that at a
cumulative water stress level of 25, yields are reduced by nearly 50%. Low N maize has very
few instances of water stress, because in all but the most extreme cases, maize remains N limited
and not water limited. Both rotation and high N maize show higher levels of water stress, and
levels in high N maize are highest in the sandier soil (soil LF). Interestingly, in soil LF maize in
rotation has a lower percentage of stressed years than high N maize, while the pattern is reversed
in soil HF.
While soil water stress increases with rising temperature, nitrogen stress is reduced.
Cumulative N stress, as with cumulative soil water stress, is calculated as the sum of daily values
of nitrogen stress for photosynthesis, which range from zero to one, over the growing season.
The reduction in N stress is largest in the rotation system, where nitrogen stress is reduced by
77% in the high fertility soil and 62% in the low fertility soil at Kasungu. Reductions are 84%
and 83% respectively at Zombwe. In low N continuous maize, levels of N stress are reduced by
37% at Kasungu and 58% at Zombwe in the high fertility soil under base rainfall. Levels are
reduced in high N maize as well, however this is from a lower baseline than the other two
treatments and thus has a smaller impact on yields. Trends are similar in all rainfall scenarios at
Kasungu. Decreases in N stress are slightly lower at Zombwe in high rainfall scenarios. For

67

example, in the AA+ rainfall increase scenario (an evenly distributed 25% increase over base
rainfall), the reduction in N stress in rotation is 77%.
2.3.6 Rotation effects under climate change
Maize yields in rotation are between 150% and 200% of yields in continuous maize systems with
the same fertilization rate at base rainfall and temperature, with greater increases seen in the
sandier soil type (Figure 2.5). This simulated rotation response is reduced under lower rainfall
and higher temperature scenarios at the more arid site, Kasungu, but almost no effect is seen at
Zombwe. This is presumably due to differences in water availability and thus maize water stress
at these two sites. Because of the risk-averse nature of smallholder farmers, risk of decreased
yield is important to examine (Figure 2.10). This risk represents the chance that a farmer will see
reduced yields following pigeonpea compared to the yield would have been following maize.
Risk is low at baseline temperature and rainfall. At Kasungu risk generally increases with rising
temperature and diminished rainfall. At Zombwe both increased rainfall—particularly rainfall
concentrated to a few extreme events—and decreased rainfall increase risk relative to baseline
rainfall. Notably, the maize-pigeonpea rotation system is always extremely low-risk in soil LF at
Zombwe, with less than 5% chance of yield reductions in any scenario. At Zombwe, soil LF has
lower risk than soil HF. However, at the more arid Kasungu site, the mean effect of rotation is
similar in the two soils, but the risk in soil LF is higher than that in soil HF.
2.4 Discussion:
2.4.1 Soil fertility remains the limiting constraint in low input systems.
It is clear that the dominant constraint to crop production in typical smallholder systems
is soil fertility, and that will remain the case for low-fertility systems under even relatively
extreme climate change predictions. As discussed elsewhere (Dimes et al., 2008), the challenge

68

of intensification of smallholder crop production remains critical, even as climate change
adaptation receives increased attention. We see here that for continuous maize with low fertilizer
use, productivity remains very low, with minimal impact from climate change, and that generally
in a positive direction. Regardless of changes in climate, yields can be significantly increased by
changing management practices.
2.4.2 Effects on high-input continuous maize
Applications of 92 kgN/ha more than doubled maize yields relative to applications of 24
kg N/ha, to approximately 4000 kg/ha. Yields decreased from this high level as temperature
increased, notably for mean annual gains of 3 °C or more. Effects of reduced rainfall were
generally small in the 10% reduction (R90) scenarios, becoming substantial, especially in
Kasungu, at the 25% reduction level. Soil type influenced the level of yield reduction. In the
sandier soil the magnitude of yield reductions due to temperature increases was lower. This
effect may have been mediated by a positive influence on soil N mineralization in sandy soils
balancing a negative direct effect of temperature on plant growth. At the same time, in sandy
soils rainfall effects were magnified. This is presumably due to more rapid drainage in sandy
soils exacerbating soil moisture stress with lower rainfall. At both Kasungu and Zombwe sites,
yields in high N maize showed the largest declines in scenarios with extreme temperature
increases and where rainfall reduction occurred during the early part of the growing season.
These scenarios also showed the highest levels of water stress in the simulated crops.
2.4.3 Effects on maize in rotation
Rotation systems had improved maize yield performance at higher temperatures relative
to continuous maize (Figure 2.5). At moderate temperature increases, maize yields improved in
rotation systems. Overall, a substantial narrowing of the yield gap was observed for maize

69

between high N continuous and rotated systems with much lower N inputs, particularly in
constant and increased rainfall scenarios, and in the sandier soil type. Improved yields in rotation
systems are likely due to a reduction in nitrogen stress levels in rotation systems with higher
temperatures. This, in turn, may be due to increased mineralization of high nitrogen residues
with higher temperatures. Hence we have a second type of trade-off, between positive and
negative effects of higher temperatures on soil and crop processes. This is clearly a highly
complex interaction and our data highlight trade-offs that otherwise could have been overlooked.
We note that data generated depend on model assumptions and algorithms, and confirmation of
the crop and soil interactions with environment observed here will require further exploration
through field investigations and modeling.
While we were not able to accurately simulate rotation effects including CO2
concentration impacts, we note that if pigeonpea yields increase dramatically this can have a
negative impact on following-year maize yields. However, management plays a central role and
rotation challenges could be avoided by reducing plant density or pruning pigeonpea.
Decreases in rainfall had a larger negative impact on maize yields in rotation than on
continuous maize, particularly for rainfall focused near the beginning of the growing season
(Figure 2.7). This vulnerability to early-season rainfall reductions is particularly important to
note since downscaled climate models generally project rainfall decreases for this area in
November and December. Therefore, early season rainfall reductions similar to those described
by scenarios E and F can be expected. In sandy soils at Kasungu, with the approximately 50%
reductions in early-season rainfall used for E and F scenarios, maize in rotation showed a yield
decrease of more than 20% at the highest temperature levels when compared to baseline rainfall.
Overall, however, the favorable performance of the maize-pigeonpea rotation under temperature

70

increases outweighed its increased sensitivity to rainfall reductions. In general, the effect of all
climate change scenarios was a reduction in the yield gap between maize grown with high N
fertilizer inputs and maize grown in rotation with modest fertilizer inputs.
The strong performance in the rotation occurred despite increases in soil water stress with
increased temperature and decreased rainfall. Water stress provides an explanation for the
relatively better performance of rotation systems in the sandier soil, as stress levels in this soil
tend to be lower for rotation systems than for continuous maize with a high N fertilization rate.
This seems counterintuitive given that the long-duration pigeonpea is extracting water for a
larger fraction of the year, and soil water levels near planting are indeed lowest in these systems.
However, from examination of daily water content in some simulations, it appears that while the
initial soil water content is low in rotation, recharge occurs rapidly, particularly in well-drained
soils. This effect was also seen in field experiments in Zimbabwe by Ncube et al. (2007).
Because of this rapid recharge, the initially low water content does not lead to water stress
conditions throughout the growing season. The rotation systems also include a higher amount of
cover from the long-duration crop, and pigeonpea residues remain in the field until pre-sowing
tillage. Increased soil cover, such as that provided by a leafy long-duration crop like pigeonpea,
has been shown previously to mitigate some of the water stress effects seen under climate change
scenarios (Cooper et al., 2009).
2.4.4 Effects on Pigeonpea
Pigeonpea yields showed smaller temperature-induced decreases than maize yields in
Zombwe despite much larger changes in length of growing period. Because pigeonpea matures
far into the dry season, the temperature-induced reduction in time to maturity is likely beneficial
to the crop (Dimes et al., 2008). At Kasungu, which is both drier and warmer, pigeonpea yield

71

decreases were large: up to 25% in base rainfall over 4 °C. Pigeonpea yield declines due to
decreased rainfall are larger than corresponding maize yield decreases: up to 25% less in the
worst-case rainfall scenarios at 4°C temperature increases than in the baseline rainfall. Pigeonpea
cultivars show a wide range of maturity durations, as well as varying responses to temperature
and photoperiod (Silim et al., 2010). Therefore the temperature response seen here will vary
depending on the pigeonpea variety used.
2.4.5 Implications for use of long-duration legumes
In all cases tested here, maize in the rotation system has higher average yields than
continuous maize at the same (24 kgN/ha) N fertilization level. On average, yields in rotation are
1.5 to 2 times those in continuous maize. However, the risk of poor performance, defined as a
yield below that found in the low N continuous maize system in the same year, increases with
increasing temperature and decreasing rainfall. In extreme cases at Zombwe, this risk reaches
25%, which is problematic. Risk of poor yield on an absolute level remains low here: the risk of
yields below 2000 kg/ha in the rotation system is always below 15%. It is also important to note
that this analysis is based on the immediate response to one year of pigeonpea; longer-term soil
productivity benefits of a rotation system were not explored here. Previously we have shown that
rotations induce changes in soil water holding capacity and soil N status that can be substantial
(Snapp et al., 2010 and Chapter 1). Rotation decreases risk over time, as shown through decadal
simulation scenarios (Chapter 1) and through on-farm establishment of pigeonpea-maize systems
(Snapp et al., 2002), so if a maize-legume rotation is continued over more than two years, greater
yield increases and lower risk could be expected. Given the signs of faster mineralization of
organic residues, however, long-term increases in soil carbon seen in rotation systems with high
residue inputs (Palm et al., 2001) may be reduced with increased temperatures. Risk remains

72

lowest and return highest in low-fertility, sandy soils. This suggests that maize-pigeonpea
systems should be targeted to these types of soils for maximum benefit.
Overall, maize-pigeonpea rotation systems are successful in increasing maize yields, even
under the most extreme climate scenarios and particularly in poorer soils. Pigeonpea yield
responses vary depending on site and rainfall, and may be maintained, increased, or reduced,
with declines most likely at Kasungu. Over moderate temperature increases, maize yields in
these systems can actually increase, and high maize yields are maintained even at high
temperatures, rarely putting farmers as risk relative to continuous maize (Figure 2.10).
Additional complexity is faced by farmers with very small land holdings, as these farmers will
not necessarily be able to devote land to a rotational crop given considerations of meeting maize
food requirements each year. Further study of alternative integrated soil fertility management
strategies is needed to define a range of options for the variety of constraints faced by
smallholder farmers.
2.5 Conclusions:
Maize-pigeonpea systems for integrated soil fertility management are robust to most
potential changes in climate, particularly in low-fertility well-drained soils, where they show
potential for ameliorating constraints to crop production. Given the complexity and uncertainty
created in modeling and downscaling climate change projections, the type of empirically created
scenarios used here are critical for informing the development of climate-aware agricultural
management strategies that can help alleviate fertility constraints on production while
minimizing negative impacts of climate change.

73

Figure 2.1: Baseline (2000) rainfall estimates for Lilongwe, Malawi from statistically downscaled global climate models. Line is mean
of observed data from 1982-2000 (Climate Systems Analysis Group, 2012)
2000

74

Figure 2.2: Rainfall projections for Lilongwe, Malawi for 2050 and 2090 in two scenarios from the Special Report on Emissions
Scenarios of the Intergovernmental Panel on Climate Change Results are from statistically downscaled global climate models
Change.
(Climate Systems Analysis Group, 2012).

75

Table 2.1: Rainfall scenarios used in climate change simulations. As described in Materials and
Methods, rainfall distribution groups describe the level of rainfall relative to baseline. Group R90
has 90% of baseline rainfall, Group R75 has 75% of baseline rainfall, group R+ has additional
rainfall at 110% and 125%. Individual letters correspond to the methods used to change rainfall
distribution. Scenarios using the same letter use the same pattern of rainfall removal or addition
at different intensities.
Group
R90
A

Reduce rainfall 10% on each rainy day

B

Remove all rainfall from 1 in 10 rainy days

C

Remove 1 in 5 rainy days, redistribute rainfall on 1/5 of remaining days to
reach 90% of initial rainfall amount

D

Remove all rainfall from 1 in 10 rainy spells

Group
R75
AA

Reduce rainfall 25% on each rainy day

BB

Remove all rainfall from 1 in 4 rainy days

CC

Remove 1 in 3 rainy days, redistribute rainfall on 1/5 or remaining days to
reach 75% of initial rainfall amount

E

Remove 50% of rainfall from all rainy days for 60 days after start of rainy
season

F

Remove all rainfall from ½ of rainy days for 60 days after start of rainy
season

Group
R+
A+

Increase rainfall 10% on each rainy day

AA+

Increase rainfall 25% on each rainy day

B+

Distribute additional rainfall to 1 in 5 rainy days to reach 110% of initial
total rainfall

BB+

Distribute additional rainfall to 1 in 5 rainy days to reach 125% of initial
total rainfall

76

Figure 2.3: Mean total annual rainfall in empirically generated rainfall scenarios averaged over 83 years at Kasungu and 66 years at
3:
scenarios,
Zombwe, northern Malawi. Colors denote rainfall scenario groups, as described in Table 2. Error bars are standard errors of t mean.
the

77

Figure 2.4: Mean rainy season end and start dates calculated from rainfall in empirically generated climate scenario averaged over 83
4:
scenarios,
years at Kasungu and 66 years at Zombwe, northern Malawi. For definitions of start and end date, see methods. Error bars are standard
errors of the mean.

78

Figure 2.5: Mean modeled maize yields in empirically generated climate scenarios based on
changes in temperature and rainfall amount. Points are mean simulated yields averaged over 83
years in Kasungu and 66 years in Zombwe, northern Malawi. MZ24 is continuous maize with 24
continuous
kgN/ha, MZ92 continuous maize with 92 kgN/ha, and Rot24 a maize pigeonpea rotation with 24
maize-pigeonpea
kgN/ha applied in maize years. Properties of soils HF (high fertility) and LF (low fertility)
described in methods. Error bars in base rainfall are standard errors of the mean, describing yearyear
to-year variability. Error bars in other rainfall groups are standard errors among scenario means
year
within the group, describing group variability. Descriptions of rainfall scenarios are found in
Table 2.1. Soil types defined in methods. HF=high fertility, LF=low fertility.
ypes

79

Figure 2.6: Mean modeled pigeonpea yields in empirically generated climate scenarios based on
changes in rainfall and temperature. Model results over 83 years in Kasungu and 66 years in
Zombwe, northern Malawi. Pigeonpea in rotation with maize, with 24 kgN/ha applied in maize
years. Properties of soils HF (high fertility) and LF (low fertility) described in methods. Error
bars in base rainfall are standard errors of the mean, describing year year variability. Error
year-to-year
bars in other rainfall groups are standard errors among scenario means within the group,
describing group variability. Descriptions of rainfall scenarios are found in Table 2.1. Soil types
.
2.1
defined in methods. HF=high fertility, LF=low fertility.

80

Figure 2.7: Mean maize yields in empirically generated climate scenarios as fraction of maize
yield with baseline rainfall. Yield with baseline rainfall equals one for all treatments, soil types,
and temperatures. Model results over 83 years in Kasungu and 66 years in Zombwe, northern
Malawi. MZ24 is continuous maize with 24 kgN/ha, MZ92 continuous maize with 92 kgN/ha,
and Rot24 a maize-pigeonpea rotation with 24 kgN/ha applied in maize years. Descriptions of
rainfall scenarios are found in Table 2.1. Soil types defined in methods. HF=high fertility,
LF=low fertility.

81

Figure 2.7 (cont’d)
a. Kasungu

82

Figure 2.7 (cont’d)
b. Zombwe

83

Figure 2.8: Percentage of years in which cumulative soil water stress in maize exceeds 10 in R75
8:
group rainfall reduction climate scenarios (see Table 2.1 and Figure 2.3). Model results from 83
3).
years in Kasungu and 66 years in Zombwe, northern Malawi. Expected yield reduction at soil
Zombwe,
water stress of 10 is approximately 10%. MZ24 is continuous maize with 24 kgN/ha, MZ92
continuous maize with 92 kgN/ha, and Rot24 a maize pigeonpea rotation with 24 kgN/ha applied
maize-pigeonpea
in maize years. Descriptions of rainfall scenarios are found in Table 2.1. Soil types defined in
tions
methods. HF=high fertility, LF=low fertility.

84

Figure 2.8 (cont’d)

85

Figure 2.9: Mean cumulative N stress under increases in temperature. Rainfall is maintained at
9:
.
baseline levels. Model results from 83 years in Kasungu and 66 years in Zombwe, northern
Malawi. MZ24 is continuous maize with 24 kgN/ha, MZ92 continuous maize with 92 kgN/ha,
and Rot24 a maize-pigeonpea rotation with 24 kgN/ha applied in maize years. Soil types defined
pigeonpea
d
in methods. HF=high fertility, LF=low fertility.

86

Figure 2.9 (cont’d)

87

Figure 2.10: Risk of yield reduction due to legume diversification in empirically generated climate scenarios. Risk is defined as the
e
percentage of years where rotation yield is lower than continuous maize yield at the same N fertilization rate (24 kgN/ha ). Model
results from 83 years in Kasungu and 66 years in Zombwe, northern Malawi. Descriptions of rainfall scenarios are found in Table 2.1.
Soil types defined in methods. HF=high fertility, LF=low fertility.

88

Figure 2.10 (cont’d)

89

REFERENCES

90

REFERENCES
Barron, J., J. Rockström, F. Gichuki, and N. Hatibu. 2003. Dry spell analysis and maize yields
for two semi-arid locations in east Africa. Agricultural and Forest Meteorology 117(1-2):
23–37.
Bezner Kerr, R., S. Snapp, M. Chirwa (deceased), L. Shumba, and R. Msachi. 2007.
Participatory Research on Legume Diversification With Malawian Smallholder Farmers for
Improved Human Nutrition and Soil Fertility. Experimental Agriculture 43(04): 437–453.
Carberry, P.S. 1989. Testing the CERES-Maize simulation model in a semi-arid tropical
environment. Field Crops Research 20(4): 297–315.
Carberry, P.S., M.E. Probert, J.P. Dimes, B.A. Keating, and R.L. Mccown. 2002. Role of
modelling in improving nutrient efficiency in cropping systems. Plant and Soil: 193–203.
Carberry, P.., R. Ranganathan, L.. Reddy, Y.. Chauhan, and M.. Robertson. 2001. Predicting
growth and development of pigeonpea: flowering response to photoperiod. Field Crops
Research 69(2): 151–162.
Chikowo, R., M. Corbeels, P. Tittonell, B. Vanlauwe, A. Whitbread, and K.E. Giller. 2008.
Aggregating field-scale knowledge into farm-scale models of African smallholder systems:
Summary functions to simulate crop production using APSIM. Agricultural Systems 97(3):
151–166.
Christensen, J.H., B. Hewitson, A. Busuioc, A. Chen, X. Gao, I. Held, R. Jones, R.K. Kolli, W.T. Kwon, R. Laprise, V. Magaña Rueda, L. Mearns, C.G. Menéndez, J. Räisänen, A. Rinke,
A. Sarr, and P. Whetton. 2007. Regional Climate Projections. In Solomon, S., Quin, D.,
Manning, M., Chen, Z., Chen, M., Marquis, K.B., Averyt, K.B., Tignor, M., Miller, H.L.
(eds.), Climate Change 2007: The Physical Science Basis. Contribution of Working Group I
to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.
Cambridge University Press, Cambridge, Unkited Kingdom and New York, NY, USA.
Climate Systems Analysis Group. 2012. Climate Information Portal. University of Cape
TownAvailable at http://cip.csag.uct.ac.za/ (verified 9 July 2012).
Cooper, P.J.M., J. Dimes, K.P.C. Rao, B. Shapiro, B. Shiferaw, and S. Twomlow. 2008. Coping
better with current climatic variability in the rain-fed farming systems of sub-Saharan
Africa: An essential first step in adapting to future climate change? Agriculture, Ecosystems
& Environment 126(1-2): 24–35.
Cooper, P.J.M., K.P.C. Rao, P. Singh, J. Dimes, P.S. Traore, K. Rao, P. Dixit, and S.J.
Twomlow. 2009. Farming with current and future climate risk : Advancing a “Hypothesis
of Hope” for rainfed agriculture in the semi-arid tropics. Journal of SAT Agricultural
Research 7: 1–19.

91

Dakora, F.D., and S.O. Keya. 1997. Contribution of Legume Nitrogen Fixation to Sustainable
Agriculture in Sub-Saharan Africa. Soil Biology and Biochemistry 29(5/6): 809–817.
Dalgliesh, N., M. Foale, and R. McCown. 2009. Re-inventing model-based decision support with
Australian dryland farmers. 2. Pragmatic provision of soil information for paddock-specific
simulation and farmer decision making. Crop and Pasture Science 60(11): 1031–1043.
Dimes, J., P. Cooper, and K. Rao. 2008. Climate change impact on crop productivity in the semiarid tropics of Zimbabwe in the 21st century. p. 189–198. In Proceedings of the Workshop
on Increasing the Productivity and Sustainability of Rainfed Cropping Systems of Poor,
Smallholder Farmers. Tamale, Ghana.
Dimes, J.P., S.J. Twomlow, and P.S. Carberry. 2003. Application of new tools: exploring the
synergies between simulation models and participatory research in smallholder farming
systems. Global Theme 3: Water, Soil and Agrodiversity Management for Ecosystem
Resilience. Report no 5. Bulawayo, Zimbabwe.
Dixit, P., P.J.M. Cooper, J. Dimes, and K. Rao. 2011. Adding value to field-based agronomic
research through climate risk assessment: A case study of maize production in Kitale,
Kenya. Experimental Agriculture 47(02): 317–338.
Doane, T.A., and W.R. Horwith. 2003. Spectrophotometric Determination of Nitrate with a
Single Reagent. Analytical Letters 36(12): 2713–2722.
Doherty, R.M., S. Sitch, B. Smith, S.L. Lewis, and P.K. Thornton. 2010. Implications of future
climate and atmospheric CO 2 content for regional biogeochemistry, biogeography and
ecosystem services across East Africa. Global Change Biology 16(2): 617–640.
Funk, C., M.D. Dettinger, J.C. Michaelsen, J.P. Verdin, M.E. Brown, M. Barlow, and A. Hoell.
2008. Warming of the Indian Ocean threatens eastern and southern African food security
but could be mitigated by agricultural development. Proceedings of the National Academy
of Sciences of the United States of America 105(32): 11081–6.
Garrity, D.P., F.K. Akinnifesi, O.C. Ajayi, S.G. Weldesemayat, J.G. Mowo, A. Kalinganire, M.
Larwanou, and J. Bayala. 2010. Evergreen Agriculture: a robust approach to sustainable
food security in Africa. Food Security 2(3): 197–214.
Gijsman, a. ., S.. Jagtap, and J.. Jones. 2002. Wading through a swamp of complete confusion:
how to choose a method for estimating soil water retention parameters for crop models.
European Journal of Agronomy 18(1-2): 77–106.
Giller, K.E., P. Tittonell, M.C. Rufino, M.T. van Wijk, S. Zingore, P. Mapfumo, S. Adjei-Nsiah,
M. Herrero, R. Chikowo, and M. Corbeels. 2010. Communicating complexity: Integrated
assessment of trade-offs concerning soil fertility management within African farming
systems to support innovation and development. Agricultural Systems 104(2): 191–203.

92

Gowing, J.W., M.D.B. Young, N. Hatibu, H.F. Mahoo, F. Rwehumbiza, and O.B. Mzirai. 2003.
Developing Improved Dryland Cropping Systems for Maize in Semi-Arid Tanzania. Part II.
Use of a Model To Extrapolate and Add Value To Experimental Results. Experimental
Agriculture 39(3): 293–306.
Høgh-Jensen, H., F. Myaka, and W. Sakala. 2007. Yields and qualities of pigeonpea varieties
grown under smallholder farmers’ conditions in Eastern and Southern Africa. African
Journal of Agricultural Research 2(6): 269–278.
Jones, P.G., J. Amador, M. Campos, K. Hayhoe, M. Marín, J. Romero, and A. Fischlin. 2005.
Generating climate change scenarios at high resolution for impact studies and adaptation:
Focus on developing countries. p. 38–56. In Robledo, C., Kanninen, M., Pedroni, L. (eds.),
Tropical forests and adaptation to climate change. Center for International Forestry
Research, Bogor Barat, Indonesia.
Jones, J.W., G. Hoogenboom, C.H. Porter, K.J. Boote, W.D. Batchelor, L.A. Hunt, P.W.
Wilkens, U. SIngh, A.J. Gijsman, and J.T. Ritchie. 2003. The DSSAT cropping system
model. European Journal of Agronomy 18(3-4): 235–265.
Jones, P.G., and P.K. Thornton. 2000. MarkSim: Software to Generate Daily Weather Data for
Latin America and Africa. Agronomy Journal 92(3).
Keating, B.A., P.S. Carberry, G.L. Hammer, M.E. Probert, M.J. Robertson, D.P. Holzworth, N.I.
Huth, J.N.G. Hargreaves, H. Meinke, and Z. Hochman. 2003. An overview of APSIM, a
model designed for farming systems simulation. European Journal of Agronomy 18(3-4):
267–288.
Ludwig, F., and S. Asseng. 2006. Climate change impacts on wheat production in a
Mediterranean environment in Western Australia. Agricultural Systems 90(1-3): 159–179.
Marenya, P.P., and C.B. Barrett. 2009. Soil quality and fertilizer use rates among smallholder
farmers in western Kenya. Agricultural Economics 40(5): 561–572.
Mertz, O., C. Mbow, A. Reenberg, and A. Diouf. 2009. Farmers’ perceptions of climate change
and agricultural adaptation strategies in rural Sahel. Environmental management 43(5):
804–16.
Mhango, W.G. 2011. Nitrogen Budgets in Legume-Based Cropping Systems in Northern
Malawi.
Mhango, W.G., S.S. Snapp, and G.Y.K. Phiri. 2012. Opportunities and constraints to legume
diversification for sustainable maize production on smallholder farms in Malawi.
Renewable Agriculture and Food Systems In press.
Morris, M., V.A. Kelly, R.J. Kopicki, and D. Byerlee. 2007. Fertilizer use in African agriculture:
Lessons learned and good practice guidelines. World Bank Publications, Washington D.C.

93

Mwale, C.D., V.H. Kabambe, W.D. Sakala, K.E. Giller, A.A. Kauwa, I. Ligowe, and D.
Kamalongo. 2011. Model Validation Through Long-Term Promising Sustainable MaizePigeon Pea Residue Management in Malawi. p. 325–333. In Bationo, A., Waswa, B.,
Okeyo, J.M., Maina, F., Kihara, J.M. (eds.), Innovations as Key to the Green Revolution in
Africa. Springer Netherlands, Dordrecht.
Myaka, F.M., W.D. Sakala, J.J. Adu-Gyamfi, D. Kamalongo, A. Ngwira, R. Odgaard, N.E.
Nielsen, and H. Høgh-Jensen. 2006. Yields and accumulations of N and P in farmermanaged intercrops of maize–pigeonpea in semi-arid Africa. Plant and Soil 285(1): 207–
220.
Ncube, B., J.P. Dimes, M.T. van Wijk, S.J. Twomlow, and K.E. Giller. 2009. Productivity and
residual benefits of grain legumes to sorghum under semi-arid conditions in south-western
Zimbabwe: Unravelling the effects of water and nitrogen using a simulation model. Field
Crops Research 110(2): 173–184.
Ncube, B., S.J. Twomlow, M.T. Wijk, J.P. Dimes, and K.E. Giller. 2007. Productivity and
residual benefits of grain legumes to sorghum under semi-arid conditions in southwestern
Zimbabwe. Plant and Soil 299(1-2): 1–15.
Ngongondo, C., C.-Y. Xu, L. Gottschalk, and B. Alemaw. 2011. Evaluation of spatial and
temporal characteristics of rainfall in Malawi: a case of data scarce region. Theoretical and
Applied Climatology 106: 79–93.
Palm, C.A., K.E. Giller, P.L. Mafongoya, and M.J. Swift. 2001. Management of organic matter
in the tropics : translating theory into practice. Nutrient Cycling in Agroecosystems
(1997): 63–75.
Pickering, N.B., J.W. Hansen, J.W. Jones, C.M. Wells, V.K. Chan, and D.C. Godwin. 1994.
WeatherMan: A Utility for Managing and Generating Daily Weather Data. Agronomy
Journal 337: 332–337.
Piha, M.I. 1993. Optimizing Fertilizer Use and Practical Rainfall Capture in a Semi-Arid
Environment with Variable Rainfall. Experimental Agriculture 29: 405–415.
Probert, M.E., J.P. Dimes, B.A. Keating, R.C. Dalal, and W.M. Strong. 1998. APSIM’s water
and nitrogen modules and simulation of the dynamics of water and nitrogen in fallow
systems. Agricultural Systems 56(1): 1–28.
R Development Core Team. 2011. R: A language and environment for statistical computing. R
Foundation for Statistical Computing, Vienna, Austria.
Rao, K., W. Ndegwa, and K. Kizito. 2011. Climate variability and climate change: Farmer
perceptions and understanding of intra-seasonal variability in rainfall and associated risk in
semi-arid Kenya. Experimental Agriculture 47: 267–291.

94

Ritchie, J.T. 1972. Model for predicting evaporation from a row crop with incomplete cover.
Water Resources Research 8(5).
Robertson, M.J., P.S. Carberry, Y.S. Chauhan, R. Ranganathan, and G.J. O’Leary. 2001.
Predicting growth and development of pigeonpea: a simulation model. Field Crops
Research 71(3): 195–210.
Robertson, M.J., P.S. Carberry, N.I. Huth, J.E. Turpin, M.E. Probert, P.L. Poulton, M. Bell, G.C.
Wright, S.J. Yeates, and R.B. Brinsmead. 2002. Simulation of growth and development of
diverse legume species in APSIM. Australian Journal of Agricultural Research 53(4): 429–
446.
Robertson, M.J., W.D. Sakala, T. Benson, and Z. Shamudzarira. 2005. Simulating response of
maize to previous velvet bean (Mucuna pruriens) crop and nitrogen fertiliser in Malawi.
Field Crops Research 91(1): 91–105.
Saha, S., V.K. Sehgal, S. Nagarajan, and M. Pal. 2012. Impact of elevated atmospheric CO2 on
radiation utilization and related plant biophysical properties in pigeon pea (Cajanus cajan
L.). Agricultural and Forest Meteorology 158-159: 63–70.
Sanchez, P.A. 2002. Soil fertility and hunger in Africa. Science 295(5562): 2019.
Sanchez, P.A. 2010. Tripling crop yields in tropical Africa. Nature Geoscience 3(5): 299–300.
Saxton, K.E., and W.J. Rawls. 2006. Soil Water Characteristic Estimates by Texture and Organic
Matter for Hydrologic Solutions. Soil Science Society of America Journal 70(5): 1569.
Saxton, K.E., W.J. Rawls, J.S. Romberger, and R.I. Papendick. 1986. Estimating Generalized
Soil-water Characteristics from Texture. Soil Science Society of America Journal 50(4):
NP.
Shamudzarira, Z., and M.J. Robertson. 2002. Simulating response of maize to nitrogen fertilizer
in semi-arid Zimbabwe. Experimental Agriculture 38(01): 79–96.
Silim, S., E. Gwataa, and R. Coeb. 2010. Response of pigeonpea genotypes of differrent maturity
duration to temperature and photoperiod in Kenya. African Crop Science Journal 15(2): 73
– 81.
Snapp, S. 1998. Soil nutrient status of smallholder farms in Malawi. Communications in Soil
Science and Plant Analysis 29(17): 2571–2588.
Snapp, S., M.J. Blackie, R.A. Gilbert, R. Bezner-Kerr, and G.Y. Kanyama-Phiri. 2010.
Biodiversity can support a greener revolution in Africa. Proceedings of the National
Academy of Sciences of the United States of America: 1–6.

95

Snapp, S., G.Y. Kanyama-Phiri, B. Kamanga, R.A. Gilbert, and K. Wellard. 2002. Farmer and
Researcher Partnerships in Malawi: Developing Soil Fertility Technologies for the NearTerm and Far-Term. Experimental Agriculture 38(04).
Stern, R.D., and P.J.M. Cooper. 2011. Assessing climate risk and climate change using rainfall
data–a case study from Zambia. Experimental Agriculture 47(02): 241–266.
Swart, R.. J., P. Raskin, and J. Robinson. 2004. The problem of the future : sustainability
science and scenario analysis. Global Environmental Change 14: 137–146.
Thomas, D.S.G., C. Twyman, H. Osbahr, and B. Hewitson. 2007. Adaptation to climate change
and variability: farmer responses to intra-seasonal precipitation trends in South Africa.
Climatic Change 83(3): 301–322.
Thornton, P.K., P.G. Jones, G. Alagarswamy, and J. Andresen. 2009. Spatial variation of crop
yield response to climate change in East Africa. Global Environmental Change 19(1): 54–
65.
Thornton, P.K., P.G. Jones, G. Alagarswamy, J. Andresen, and M. Herrero. 2010. Adapting to
climate change: Agricultural system and household impacts in East Africa. Agricultural
Systems 103(2): 73–82.
Twomlow, S.J., F.T. Mugabe, M. Mwale, R. Delve, D. Nanja, P.S. Carberry, and M. Howden.
2008. Building adaptive capacity to cope with increasing vulnerability due to climatic
change in Africa-A new approach. Physics and Chemistry of the Earth, Parts A/B/C 33(813): 780–787.
Vanlauwe, B., A. Bationo, J. Chianu, K.E. Giller, R. Merckx, U. Mokwunye, O. Ohiokpehai, P.
Pypers, R. Tabo, K.D. Shepherd, and others. 2010. Integrated soil fertility management
Operational definition and consequences for implementation and dissemination. Outlook on
Agriculture 39(1): 17–24.
Waddington, S.R., and J. Karigwindi. 2001. Productivity and Profitability of Maize + Groundnut
Rotations Compared with Continuous Maize on Smallholder Farms in Zimbabwe.
Experimental Agriculture 37(01): 83–98.
Whitbread, A.M., M.J. Robertson, P.S. Carberry, and J.P. Dimes. 2010. How farming systems
simulation can aid the development of more sustainable smallholder farming systems in
southern Africa. European Journal of Agronomy 32(1): 51–58.
Wickham, H. 2011. The Split-Apply-Combine Strategy for Data. Journal Of Statistical Software
40(1): 1–26.
Winkler, J.A., G.S. Guentchev, and P. Tan. 2011. Climate Scenario Development and
Applications for Local ⁄ Regional Climate Change Impact Assessments : An Overview for

96

the Non-Climate Scientist Part I : Scenario Development Using Downscaling Methods.
Geography 6: 275–300.
Wood, A.W., L.R. Leung, V. Sridhar, and D.P. Lettenmaier. 2004. Hydrologic Implications of
Dynamical and Statistical Approaches to Downscaling Climate Model Outputs. Climatic
Change 62(1-3): 189–216.
Wortmann, C.S., B.D. Mcintyre, and C.K. Kaizzi. 2000. Annual soil improving legumes:
agronomic effectiveness, nutrient uptake, nitrogen fixation and water use. Field Crops
Research 68: 75–83.

97