INTEGRATED REMOTE SENSING AND CROP SYSTEM MODELING
FOR PRECISION AGRICULTURE ACROSS SPATIAL AND TEMPORAL SCALES
By
Bradley George Peter
2019
A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Geography—Doctor of Philosophy
ABSTRACT
By
Bradley George Peter
INTEGRATED REMOTE SENSING AND CROP SYSTEM MODELING
FOR PRECISION AGRICULTURE ACROSS SPATIAL AND TEMPORAL SCALES
In light of global environmental change, population pressure, and food production demands, there
is considerable value in mapping biogeographic crop niche and characterizing crop productivity
at multiple scales to enhance the impact of agricultural improvement across Africa. Crop system
research has advanced sustainable strategies for intensifying food production; however, questions
regarding where to implement innovative technologies are largely unresolved.
This dissertation focuses on four geographic questions: (1) Where is the fundamental cli-
mate niche of maize, pigeonpea, and sorghum across Africa? (2) Where are marginal lands in
Malawi and what are the underlying drivers of marginality? (3) Based on the drivers of marginal
maize production, what are geographic scaling options for integration of pigeonpea into maize-
based cropping systems? (4) What spatial resolutions are effective for conducting precision agri-
culture at the farm scale in smallholder systems?
Overarching themes within the geographic discipline such as the modifiable areal unit
problem and ecological fallacy problem underpin this research. Marginal areas for maize are high-
lighted at the Africa and Malawi scales and overlain with the optimal climate niche for crops such
as sorghum and pigeonpea that offer multiple ecosystem services (e.g., soil rehabilitation through
nitrogen fixation). Crop productivity is evaluated at scales relative to policy making delineations
in Malawi (i.e., country, district, and extension planning area) to disentangle heterogeneity at local
scales that may appear homogeneous at broader scales. At the Malawi farm scale, this research
included the use of a small unmanned aerial system (sUAS), national government satellites (e.g.,
Sentinel-2), and commercial satellites (e.g., SPOT 6). Spectral measurements of crop status were
evaluated at multiple spatial resolutions (ranging from 0.07–20-m) to determine what spatial res-
olutions and what spectral indices are most effective for estimating crop yields and crop chloro-
phyll.
Results of this research include high spatial resolution maps of maize, pigeonpea, and sor-
ghum suitability across Africa, indicating that pigeonpea and sorghum occupy unique agroecolog-
ical zones throughout the continent (e.g., sorghum in the Sahel region). Similarly, pigeonpea suit-
ability in Malawi occupies a greater land area than the extent to which it is currently cultivated,
demonstrating that integration into maize-based cropping systems, particularly where soil is mar-
ginal, can have beneficial scaling outcomes. For the smallholder farm scale, problems of clouds
and satellite revisit rates have not yet been overcome for precision agriculture. In this regard, sUAS
are a promising option for relating spectral signals to on-farm measurements of crop status. Evi-
dence from drone flights conducted at two experimental farms in the central region of Malawi
(Nyambi and Ntubwi) suggest that spatial resolutions closer to the plant scale (i.e., 14–27-cm) are
most effective for relating spectral imagery to crop status. Moreover, the green normalized differ-
ence vegetation index (GNDVI) and green soil adjusted vegetation index (GSAVI) were consist-
ently correlated with crop chlorophyll and yield, illustrating that a broad range of indices should
be evaluated for precision agriculture.
Copyright by
BRADLEY GEORGE PETER
2019
ACKNOWLEDGMENTS
Many people have been integral to my academic journey in geography. First, I am eternally grateful
to my advisor, Dr. Joseph Messina, for inviting me into his lab, mentoring me as an early career
scholar and scientist, introducing me to numerous collaborative opportunities, and continually
providing support throughout my doctoral program. I would also like to extend my utmost grati-
tude to Dr. Sieglinde Snapp for her regular involvement in my dissertation, offering her expertise
in agronomy and inviting me to collaborate with her lab. Many thanks as well to Dr. Kyla Dahlin
for expanding my knowledge of ecological remote sensing and inviting me to participate in her
lab, and thank you to Dr. Leo Zulu for challenging me to approach remote sensing-based mapping
more holistically by integrating social science philosophies. Additionally, this dissertation required
extensive field work in Malawi for which I am grateful to Mayamiko Kakwera for his expertise
and friendship.
Thanks also to Dr. Nathan Moore and Dr. Randall Schaetzl for their support in advancing
my academic career, and to Dr. Ashton Shortridge for his guidance when I first joined Michigan
State University. Special thanks as well to Jean Lepard and Sharon Ruggles for keeping me on
track throughout my program and during my travels abroad. To my undergraduate geography
professors at The University of Texas at Austin, Dr. Leo Zonn, Dr. Eugenio Arima, and Dr.
Jennifer Miller, I am appreciative of the encouragement and support during my transition to grad-
uate studies, and for exposing me to the breadth of geography as a discipline.
I am also incredibly grateful to the colleagues with whom I shared a research lab for all the
collaborative problem solving and scientific discussions. Special thanks in this regard to April
Frake, Mark DeVisser, and Shengpan Lin. I would also like to extend thanks to my friends in the
Department of Geography, Environment, and Spatial Sciences for their camaraderie during our
v
time at MSU. To my family here in Michigan, Cadi Fung and our cat Edwin, I am grateful for the
moments of relaxation, adventure, and countless hours working together. To my mother (Cathe-
rine Clifford), father (Douglas Peter), and brother (Christos Peter), I am thankful for the unwa-
vering support and patience.
Finally, I would like to thank all of the organizations that offered much appreciated finan-
cial support and partnership throughout my research: USAID, The Bill & Melinda Gates Foun-
dation, The United States Geospatial Intelligence Foundation (USGIF), The American Society of
Photogrammetry and Remote Sensing (ASPRS), MSU Global Center for Food Systems Innova-
tion (GCFSI), MSU Center for Global Change and Earth Observations (CGCEO), MSU Envi-
ronmental Science and Policy Program (ESPP), Owen Gregg Climate Change Research Award,
The Lilongwe University of Agriculture and Natural Resources (LUANAR), Africa RISING, and
The Department of Geography, Environment, and Spatial Sciences at Michigan State University.
vi
TABLE OF CONTENTS
LIST OF TABLES ............................................................................................................................ x
LIST OF FIGURES .......................................................................................................................... xi
INTRODUCTION ........................................................................................................................... 1
1.1 Multiscale Geographic Challenges ............................................................................................... 1
1.2 Multiscale Geographic Solutions ................................................................................................. 6
NATURE-BASED AGRICULTURAL SOLUTIONS:
SCALING PERENNIAL GRAINS ACROSS AFRICA ......................................................... 11
2.1 Abstract ........................................................................................................................................... 11
2.2 Introduction .................................................................................................................................... 12
2.2.1 Adapting Agriculture in View of a Changing Landscape ................................................ 12
2.2.2 Sustainable Land Management and Emergent Nature-Based Solutions ....................... 12
2.2.3 Perennial Grains: A Nature-Based Approach for Africa ................................................ 14
2.3 Methods ........................................................................................................................................... 16
2.3.1 Fundamental Climate Niche: Maize, Pigeonpea, and Sorghum ..................................... 16
2.3.2 Spatial Distribution of Climatic Variables ......................................................................... 18
2.4 Results ............................................................................................................................................. 19
2.4.1 Spatial Organization of Crop Suitability ............................................................................ 19
2.4.2 Scaling Perennial Innovations for Maximum Benefit ...................................................... 23
2.5 Discussion and Conclusions......................................................................................................... 25
2.5.1 Promoting a Sustainable Future .......................................................................................... 25
2.5.2 Conclusions ............................................................................................................................ 27
A MULTISCALAR APPROACH TO MAPPING MARGINAL
AGRICULTURAL LAND: SMALLHOLDER AGRICULTURE IN MALAWI ........... 29
3.1 Abstract ........................................................................................................................................... 29
3.2 Introduction .................................................................................................................................... 30
3.2.1 Mapping Global Marginal Agricultural Lands for the Smallholder Context ................ 34
3.2.2 Malawi as a Bellwether for Identifying Marginal Agricultural Lands ............................ 35
3.2.3 The Malawian Landscape: Cultivation, Climate, and Population ................................... 37
3.3 Methods ........................................................................................................................................... 39
3.3.1 Remotely Sensed Data Products and Preprocessing ........................................................ 39
3.3.2 Locating Long-Term Agricultural Land: Minimizing Errors of Commission .............. 41
3.3.3 Measuring Historic Productivity and Interannual Variability of Agriculture ............... 43
3.3.4 Calibrating the Model to Field Observations .................................................................... 45
3.3.5 Identifying Crop Niche and Mapping Marginal Maize Agricultural Land .................... 47
3.3.6 Devising a Multiscalar Marginal Land Model ................................................................... 50
3.4 Results and Discussion .................................................................................................................. 50
3.4.1 Multiscalar Interannual Productivity and Variability of Agriculture .............................. 50
3.4.2 Map of Marginal Maize Agricultural Land in Malawi ...................................................... 50
3.4.3 Geographic Themes.............................................................................................................. 52
vii
3.4.4 Reclaiming Marginal Lands .................................................................................................. 54
3.4.5 Social Factors Driving Marginal Productivity ................................................................... 55
3.5 Conclusions .................................................................................................................................... 58
SCALING AGRICULTURAL INNOVATIONS: PIGEONPEA IN MALAWI ............ 59
4.1 Abstract ........................................................................................................................................... 59
4.2 Introduction .................................................................................................................................... 59
4.2.1 Prospect for Soil Rehabilitation in Malawi ........................................................................ 62
4.2.2 Horizontal Scaling of Pigeonpea in Malawi ...................................................................... 63
4.3 Methods and Data ......................................................................................................................... 65
4.3.1 Pigeonpea Niche ................................................................................................................... 65
4.3.2 Targeting Development Strategies ...................................................................................... 65
4.3.3 Field Survey and Data Acquisition ..................................................................................... 67
4.4 Results ............................................................................................................................................. 68
4.4.1 Spatial Distribution of Pigeonpea in Malawi ..................................................................... 68
4.4.2 Pigeonpea Niche and Deployment Potential .................................................................... 69
4.5 Discussion ....................................................................................................................................... 74
4.5.1 Targeting Investments in Sustainable Agriculture Technologies ................................... 74
4.5.2 Drivers of Pigeonpea Production ....................................................................................... 75
4.6 Conclusions .................................................................................................................................... 77
MULTI-SPATIAL RESOLUTION SATELLITE AND sUAS IMAGERY FOR
PRECISION AGRICULTURE ON SMALLHOLDER FARMS IN MALAWI ............ 78
5.1 Abstract ........................................................................................................................................... 78
5.2 Introduction .................................................................................................................................... 78
5.2.1 Precision Agriculture and sUAS ......................................................................................... 78
5.2.2 Variable-Rate Technology and Spectral Signals of Farm Productivity .......................... 80
5.2.3 National Government and Commercial Satellites ............................................................ 81
5.3 Methods ........................................................................................................................................... 82
5.3.1 Problems of Clouds and Satellite Revisit Rates ................................................................ 84
5.3.2 Multispectral Indices and On-Farm Measurements ......................................................... 89
5.3.3 Multiple Linear Regression Permutations.......................................................................... 93
5.4 Results and Discussion .................................................................................................................. 94
5.5 Conclusions .................................................................................................................................... 99
CONCLUSIONS ............................................................................................................................... 101
6.1 Overall Contributions ................................................................................................................... 101
6.2 Future Research ............................................................................................................................. 104
6.2.1 Web-Based GIS for Near Real-Time Crop Niche Mapping ........................................... 104
6.2.2 Crop Modeling and Return on Investment from Variable-Rate Technology .............. 105
6.2.3 Crop Biodiversity, Nutrition, and Land Equivalence Ratios .......................................... 107
6.2.4 Crop Differentiation and Further Evaluation of Spectral Indices Using sUAS .......... 108
6.3 Closing Remarks ............................................................................................................................ 109
viii
APPENDIX ........................................................................................................................................ 111
BIBLIOGRAPHY ............................................................................................................................. 123
ix
LIST OF TABLES
Table 2.1 Optimal temperature and precipitation ranges for maize, pigeonpea, and sorghum 17
Table 2.2 Agricultural land area proportions for each category (suitability classifications) ....... 22
Table 2.3 Agricultural land area proportions for each category
(marginal/optimal intersection) ......................................................................................................... 25
Table 3.1 Biogeographic data sets used to determine marginality and suboptimality of
maize agriculture ................................................................................................................................... 36
Table 3.2 Land-use products incorporated in errors of commission-managed agricultural
land-use map ......................................................................................................................................... 39
Table 3.3 Marginal temperature and precipitation parameters for maize .................................... 44
Table 4.1 Optimal pigeonpea climate niche ..................................................................................... 66
Table 5.1 Indices, abbreviations, and equations for each index under study .............................. 91
Table 5.2 SPAD and yield measurements at the Ntubwi and Nyambi trial farms...................... 93
Table 5.3 Top performing single and multiple linear regression models as ranked by AICc .... 97
Table A.1 Field survey data—crop systems and biophysical factors limiting production ......... 113
x
LIST OF FIGURES
Figure 1.1 Climate niche for common bean in Southeastern Africa ............................................. 3
Figure 1.2 Linking the dissertation studies ....................................................................................... 8
Figure 2.1 Diagram of the maize, pigeonpea, and sorghum temperature and precipitation
suitability classiï¬cation thresholds ..................................................................................................... 17
Figure 2.2 Temperature and precipitation suitability for maize, pigeonpea, and sorghum ....... 21
Figure 2.3 Combined suitability of temperature and precipitation suitability for maize,
pigeonpea, and sorghum ..................................................................................................................... 22
Figure 2.4 Intersection of marginal maize and optimal pigeonpea or sorghum.......................... 24
Figure 3.1 Errors of commission-managed agricultural land–use map ........................................ 43
Figure 3.2 Illustrative diagram of the six interannual variability trends........................................ 44
Figure 3.3 Multiscalar interannual variability and marginal agricultural land model ................... 47
Figure 3.4 Multiscalar interannual variability: low-stable (light orange), low-variable (dark
orange), medium-stable (light blue), medium-variable (dark blue), high-stable (light green),
and high-variable (dark green) ............................................................................................................ 49
Figure 3.5 Marginal maize agricultural land (2000–2014) ............................................................... 52
Figure 4.1 Reclassification diagram: Transforming maize marginality into pigeonpea
opportunity ............................................................................................................................................ 66
Figure 4.2 Study area map highlighting the northern, central, and southern regions of
Malawi and the contained districts and extension planning areas ................................................. 68
Figure 4.3 Map depicts the spatial distribution of pigeonpea production and yield in 2005
and 2012 by extension planning area ................................................................................................. 70
Figure 4.4 Regional descriptive statistics: Box and whisker plot of pigeonpea production
and yield in three major regional subdivisions of Malawi for 2005 and 2012 ............................. 70
Figure 4.5 Pigeonpea climate niche: Optimal temperature and precipitation (2000–2014) ...... 72
Figure 4.6 Potential areas for deployment of pigeonpea and market development ................... 73
Figure 5.1 Spatial resolution and cost per square kilometer curve ................................................ 82
xi
Figure 5.2 Heuristic: multi-spatial resolution satellite and sUAS imagery for precision
agriculture on smallholder farms in Malawi ..................................................................................... 83
Figure 5.3 Study area map of the trial farms in Ntubwi and Nyambi, Malawi, with plots of
differing management strategies labeled ........................................................................................... 84
Figure 5.4 Satellite image NDVI acquisition over the trial farm in Nyambi, Malawi ................. 86
Figure 5.5 NDVI of the trial farm in Ntubwi across spatial resolutions ...................................... 88
Figure 5.6 NDVI of the trial farm in Nyambi across spatial resolutions ..................................... 89
Figure 5.7 Nyambi multispectral index panel, 7-cm spatial resolution ......................................... 92
Figure 5.8 Example single variable linear regressions—SPAD and spectral indices .................. 95
Figure 5.9 GSAVI 27-cm spatial resolution grid at Nyambi, scaled to the plot level and
compared to on-farm measurements of SPAD and yield .............................................................. 95
Figure 5.10 Comparing linear regression correlations of spectral indices and on-farm
measurements across spatial resolutions ........................................................................................... 99
Figure 6.1 Linking present and future research goals ..................................................................... 107
xii
INTRODUCTION
1.1 Multiscale Geographic Challenges
Continental Scale: Food Security and Climate Change—Many countries in Sub-Saharan Africa have ex-
perienced food shortages over the last several decades (Maxwell 1999; Buerkle 2005; Banda 2015;
Phiri and Macharia 2016), and smallholder subsistence agriculture is increasingly stressed by pop-
ulation pressures, soil exhaustion, and climate change (Postel 2000; Godfray 2010). Droughts and
unpredictable rainfall patterns have led to substantial fluctuations in annual agricultural output
(Tadross et al. 2009), particularly in areas where biophysical conditions are suboptimal or where
social factors limit productivity (Challinor et al. 2007). Moreover, there are sustainability concerns
associated with the continuous production of maize consequently leading to soil nutrient and or-
ganic matter depletion (Sanchez 2002; Glasson 2010; Ngwira et al. 2012; Thierfelder et al. 2013).
Among the United Nations Sustainable Development Goals are to increase agricultural
production and reduce the cases of hunger and malnutrition globally (United Nations 2018).
Given the existing and anticipated concerns regarding future climate change, impacts on global
food security, and the need to produce more food (Wheeler and Braun 2013), a question looms
for geographers—where can new agricultural technologies be introduced, and at what scales can
geospatial data be leveraged to improve agricultural production?
Elucidating the spatial organization of crop suitability across Africa is critical for resolving
questions of where and how to maximize cropping system efficacy across space. At the continental
scale, Africa contains many agroecological zones that can support a wide variety of crops, yet
many are underrepresented by current global crop suitability models (Kane et al. 2016; Peter et al.
2017a). Promotion of ecosystem service-offering crops has taken hold; however, geospatial prod-
ucts designed to target implementation areas for crop system innovations are essential to
1
successful scaling (Wood et al. 1999), yet currently lacking, particularly at regular temporal inter-
vals. At present, evidence suggests that the production of legumes in Africa does not coincide
with the biophysical extent of suitability (Snapp et al. 2019a), and that perennial grain suitability
may be more geographically extensive than previously perceived (Peter et al. 2017a). In this regard,
there is potential for legume diversification to enhance agroecosystem productivity in many re-
gions of Africa (Snapp et al. 2019a), not only for soil rehabilitation, but also as complementary
nutrition in household consumption (Bezner-Kerr 2007; Jones et al. 2014; Ojiewo 2015). For il-
lustration, the geographic extent of the fundamental climate niche for common bean across South-
eastern Africa is demonstrated here in Figure 1.1.
Importantly, crop technology adoption will only take place where biophysical suitability is
appropriate (Kwesiga et al. 2003), and as global changes continue to occur, geospatial analytics
and map outputs will need to be generated with more frequency. As global climate information is
released to the public in near real-time, crop climate-suitability maps can and should be generated
at the same pace. With cloud-based geographic information system platforms emerging, such as
Google Earth Engine (Gorelick et al. 2017), data acquisition and geospatial analytics can be per-
formed at unprecedented geographic scales and product delivery speeds to transform empirical
spatial data into scientific and actionable knowledge.
2
Figure 1.1. Climate niche for common bean in Southeastern Africa. Temperature and precipitation pa-
rameters collected from Beebe et al. (2011). Data sources: NASA MODIS Land Surface Temper-
ature (MOD11A2) (NASA LP DAAC 2015b) and CHIRPS Precipitation (Funk et al. 2015).
3
Country Scale: Characterizing Marginal Lands and Informing Policy—In densely cultivated Malawi,
the only options for increased production are through agricultural expansion into marginal areas
or intensification of farms currently persisting on marginal lands (Kalipeni 1996; Josephson et al.
2014; Ricker-Gilbert et al. 2014). Consideration of marginal lands for agricultural sustainability is
critical to national food security in Malawi as approximately 88 percent of the population lives in
rural areas and many rely on subsistence farming (Malawi National Statistical Office NSO 2005;
Chinsinga and O’Brien 2008). With this in mind, the focus for increased agricultural productivity
revolves primarily around internal reorganization of existing farms, while simultaneously meeting
goals associated with environmental conservation (Josephson et al. 2014; Ricker-Gilbert et al.
2014; Kremen 2015). However, marginal agricultural lands in need of increased production are
not widely recognized as a land-use class, and despite increasing research efforts to identify mar-
ginal lands (Wood et al. 1999; Gopalakrishnan et al. 2011; Cai et al. 2011), there is not a universal
definition for characterizing marginal land.
Few marginal agricultural land classifications directly address the smallholder farming con-
text, and such studies are often limited to one scale. Problematically, marginal agricultural lands
are typically presented as a singular classification and regularly conflated as unsuitable, which can
divert agricultural investments away from agricultural areas in need of resources (Nalepa et al.
2017). Rather than the typical singular marginal land classification, a multi-category definition that
considers the underlying factors driving marginality can inform what strategies are selected for
improving crop production on smallholder farms. Locating marginal agricultural lands and iden-
tifying the drivers of marginality ensures that agricultural improvements are targeted with respect
to the limiting factors of production.
4
Global satellites and time-series spatial information can aid in evaluating long-term agri-
cultural system resilience (Haworth et al. 2018), as well as provide critical biophysical data for crop
niche mapping. With multispectral satellite imagery, global agricultural productivity can be meas-
ured and characterized (Atzberger 2013); however, at global scales of analysis, local-scale hetero-
geneous landscapes can appear homogeneous (Sexton et al. 1998; Duveiller et al. 2015). This is in
part a product of the modifiable areal unit problem (MAUP), popularly coined by Openshaw
(1983), recognizing that pixel classifications or statistical aggregations can vary depending on the
data distribution within the areal unit selected. At the pixel scale, the ecological fallacy problem
also emerges, where assumptions may be erroneously applied to individual based on aggregate
data (Openshaw 1983). A multiscalar challenge to overcome here is that global- and continental-
scale characterizations of land in Malawi can be [mis]represented as one homogeneous category,
which is ultimately not effective for country- or district-scale policy decisions, let alone an indi-
vidual farm. Notably, Blaikie and Brookfield (1987) argued the importance of geographic scales
and socioeconomic organizations in evaluating human-environment interactions (Nuemann
2009). To this end, policymakers will benefit from a multiscale classification approach that con-
siders the modifiable areal unit problem and elucidates crop production variability at scales relative
to multiple administrative decision levels (e.g., country, districts, and extension planning areas in
Malawi).
Farm Scale: Precision Agriculture and Variable-Rate Technology—High spatial-resolution small
unmanned aerial systems (sUAS) offer unprecedented levels of geospatial analytics at the farm
scale. In-season crop status monitoring with these systems has become popular in industrialized
agriculture, with spatial applications of crop improvement informed by spatial measurements of
crop performance (known as variable-rate technology) (Sawyer 1998; Zhang and Kovacs 2012).
5
Precision agriculture can add value to smallholder environments in semi-arid Africa to inform
timely and efficient sowing and weeding, and spatial allocation of fertilizer (Aune et al. 2017);
however, there are substantial limitations to obtaining sUAS imagery across vast geographic ex-
tents. Advances in spatial and temporal resolution satellite imaging have made precision agricul-
ture in remote areas increasingly feasible; however, cloud cover and revisit rates still pose a con-
siderable hurdle (Mulla 2013).
With a fixed-wing craft such as the senseFly eBee equipped with a Parrot Sequoia multi-
spectral camera, images with a spatial resolution of less than 5-cm can be achieved. The standard
camera collects images at four spectral wavelengths: green, red, red edge, and near infrared, which
can be manipulated to produce an extensive collection of crop status indices. At this scale, intra-
farm variability can be measured and intra-farm recommendations made; however, it is well-
known that the linear correlation between remote sensing-based indices and in situ crop measure-
ments will vary depending on the crop evaluated, index selected, spatial resolution of the imagery,
and interference of clouds, shadows, or soil (Hatfield and Prueger 2010). Two questions of interest
in this research area are: What spatial resolutions (ranging from national government and com-
mercial satellites to sUAS) are effective for conducting precision agriculture at the smallholder
farm scale in Malawi?, and (2) What spectral indices are effective for informing variable-rate tech-
nologies on smallholder farms?
1.2 Multiscale Geographic Solutions
Purpose and Objectives—This research focuses on four overarching questions: (1) Where is the fun-
damental climate niche of maize, pigeonpea, and sorghum across Africa? (2) Where are marginal
lands in Malawi and what are the underlying drivers of marginality? (3) Based on the drivers of
marginal maize production, what are geographic scaling options for integration of pigeonpea into
6
maize-based cropping systems? (4) What spatial resolutions and spectral indices are effective for
conducting precision agriculture at the farm scale in smallholder systems?
Key outputs of these studies include the development of protocols and products that pro-
vide multi-scale decision guidance for improving smallholder agricultural systems in Malawi by
integrating high-resolution satellite imagery, sUAS imagery, field measurements, geospatial analyt-
ics, and variable-rate technology. The project consists of the following four geographic solutions
for improving agricultural production across spatial scales. Study 1: establish crop profiles and
map the fundamental climate niche for maize and select perennial grains across Africa. Study 2:
identify marginal maize agricultural land in Africa and Malawi based on climate niche, terrain char-
acteristics, and measured interannual production variability. Study 3: (a) map the fundamental cli-
mate niche of pigeonpea in Malawi and (b) identify potential areas for the deployment of pigeon-
pea into maize-based cropping systems to illustrate geographic scalability. Study 4: evaluate a range
of satellite and sUAS imagery spatial resolutions and spectral indices to determine effective spatial
resolutions for precision agriculture on smallholder farms (Figure 1.2).
7
Figure 1.2. Linking the dissertation studies. Evaluating agricultural productivity and targeting areas
for enhanced crop production across spatial scales. Study 1—Nature-Based Agricultural Solutions:
Scaling Perennial Grains Across Africa; Study 2—A Multiscalar Approach to Mapping Marginal
Agricultural Land: Smallholder Agriculture in Malawi; Study 3—Scaling Agricultural Innovations:
Pigeonpea in Malawi; Study 4—Multi-Spatial Resolution Satellite and sUAS Imagery for Precision
Agriculture on Smallholder Farms in Malawi.
Nature-Based Agricultural Solutions: Scaling Perennial Grains Across Africa (Study 1)—This study
established crop profiles and mapped the fundamental climate niche for maize, pigeonpea, and
sorghum across Africa using cropping parameters from a literature review of tested crop perfor-
mance. In view of widespread soil organic matter depletion, nature-based solutions such as per-
ennial grain integration into maize-based cropping systems have emerged (Maes and Jacobs 2015).
Pigeonpea, for example, is resilient to marginal environments and offers nitrogen fixation, carbon
sequestration, and has substantial root biomass that can aid in [re]building soil fertility (Fujita et
8
al. 1992; Snapp et al. 2010). Sorghum, like pigeonpea, is resilient to marginal environments (Dicko
et al. 2006) and is commonly used to improve field structure as a ridge crop for erosion mitigation
(Okigbo and Greenland 1976).
To identify the geographic scaling potential of pigeonpea and sorghum across Africa, cli-
mate-suitability maps of each crop were constructed. There are two overarching categories, opti-
mal and suboptimal, which are further subcategorized into a range of suitability. Following the
suitability classifications of maize, pigeonpea, and sorghum, marginal maize areas were intersected
with suitable sorghum and pigeonpea areas to identify areas where maize-based systems may ben-
efit from integration of one or both perennials.
A Multiscalar Approach to Mapping Marginal Agricultural Land: Smallholder Agriculture in Malawi
(Study 2)—In this study, a multiscale heuristic was designed for identifying marginal agricultural
lands and the underlying drivers of marginality. Marginal agricultural lands exist as, or may emerge
from, a complex combination of factors and are defined here as land characterized by variable
interannual production and/or chronically low production, and suboptimal biogeographic condi-
tions. The framework consists of measuring interannual crop productivity (and variability) of long-
term farmed locations and evaluating biophysical land and climate characteristics.
Net primary productivity was used to measure time-series trends of vegetation on agricul-
tural land, classifying pixels based on both the level of production and the productivity variance
over 15 years. A methodology was developed to produce these classifications across, and relative
to, multiple administrative levels (i.e., country, district, and extension planning area); however, the
model devised supports any geographic delineation. This method exposes the modifiable areal
unit problem, showing that classifications can change depending area and data distributions, but
can also serve to manage the problem by disaggregating pixels relative to local scales where
9
agroecologies are more commonly shared. These classifications were paired with marginal bioge-
ographic conditions (temperature, rainfall, and soil) to address the factors that may be limiting
agricultural production. In the absence of a biophysical driver explaining marginal production, it
is possible that some other variable is driving marginality, namely social factors such as fertilizer
availability, labor requirements, or farm management choices.
Scaling Agricultural Innovations: Pigeonpea in Malawi (Study 3)—The purpose of this study was
to identify the geographic scaling potential of integrating perennial pigeonpea into maize-based
cropping systems in Malawi. Since pigeonpea exhibits soil rehabilitation properties, integration of
this legume into maize-based cropping systems shows a range of predictable benefits, particularly
where maize production is marginal due to suboptimal soil status. Historically, pigeonpea has been
cultivated dominantly in the southern region of Malawi; however, we found that climate suitability
for pigeonpea spans the latitudinal extent of the country. This evidence is supported by the fact
that where pigeonpea is grown in the northern and central regions of Malawi, production and
yields are comparable to those in the south.
Multi-Spatial Resolution Satellite and sUAS Imagery for Precision Agriculture on Smallholder Farms
in Malawi (Study 4)—This study examined a range of spatial resolution imagery over two small-
holder farms in Malawi. The imagery under analysis was Sentinel-2 (20-m), SPOT 6 (7-m), Planet
(3-m), Pléiades (2.5-m), and sUAS (7-cm, 14-cm, and 27-cm). A heuristic was developed to (1)
automatically generate a collection of spectral indices from the initial set of spectral bands (e.g.,
normalized difference vegetation index [NDVI] and green soil adjusted vegetation index
[GSAVI]), and (2) evaluate multiple linear regression permutations of the selected indices and
proximal measurements in the field (crop chlorophyll content and crop yield).
10
NATURE-BASED AGRICULTURAL SOLUTIONS: SCALING PERENNIAL
GRAINS ACROSS AFRICA1
2.1
Abstract
Modern plant breeding tends to focus on maximizing yield, with one of the most ubiquitous im-
plementations being shorter-duration crop varieties. It is indisputable that these breeding efforts
have resulted in greater yields in ideal circumstances; however, many farmed locations across Af-
rica suffer from one or more conditions that limit the efficacy of modern short-duration hybrids.
In view of global change and increased necessity for intensiï¬cation, perennial grains and long-
duration varieties offer a nature-based solution for improving farm productivity and smallholder
livelihoods in suboptimal agricultural areas. Speciï¬c conditions where perennial grains should be
considered include locations where biophysical and social constraints reduce agricultural system
efficiency, and where conditions are optimal for crop growth. Using a time-series of remotely-
sensed data, we locate the marginal agricultural lands of Africa, identifying suboptimal temperature
and precipitation conditions for the dominant crop, i.e., maize, as well as optimal climate condi-
tions for two perennial grains, pigeonpea and sorghum. We propose that perennial grains offer a
lower impact, sustainable nature-based solution to this subset of climatic drivers of marginality.
Using spatial analytic methods and satellite-derived climate information, we demonstrate the scala-
bility of perennial pigeonpea and sorghum across Africa. As a nature-based solution, we argue that
perennial grains offer smallholder farmers of marginal lands a sustainable solution for enhancing
1 This is an Accepted Manuscript of an article published by Elsevier in Environmental Research on 18 August
2017, available online at doi.org/10.1016/j.envres.2017.08.011. Reference: Peter, B.G., Mungai, L.M.,
Messina, J.P. and Snapp, S.S., 2017. Nature-based agricultural solutions: Scaling perennial grains across
Africa. Environmental Research, 159, pp.283-290.
11
resilience and minimizing risk in confronting global change, while mitigating social and edaphic
drivers of low and variable production.
2.2
Introduction
2.2.1 Adapting Agriculture in View of a Changing Landscape
Globally, there is a call to increase food production through sustainable practices, to counter the
effects of climate change, rising population, land-use change, and deterioration of natural re-
sources (Postel 2000; Godfray et al. 2010). Subsistence agriculture in Africa in particular is becom-
ing increasingly stressed by population pressures, soil exhaustion, and climate change. Since the
1990's, the population of Africa has been growing at a rate faster than the global average (Reilly
and Schimmelpfennig 1999), contributing to the conversion of land for both agricultural produc-
tion and urbanization. Climate variability, such as unpredictable timing and quantity of rainfall,
affect smallholder farmer crop production and farm management practices (Nelson et al. 2009),
particularly in marginal environments (Reilly and Schimmelpfennig 1999). These changes, among
others, have prompted increased research for agricultural intensiï¬cation (Josephson et al. 2014;
Ricker-Gilbert et al. 2014). Sub-Saharan African agricultural systems are among the most vulner-
able systems facing these challenges (Challinor et al. 2007), and ongoing research that is both
innovative and participatory will foster biodiversity and increased productivity (Snapp et al. 2010).
2.2.2 Sustainable Land Management and Emergent Nature-Based Solutions
Sustainable land management is at the foundation of societal goals to produce food, fuel, and
fodder in an environmentally sound and supportable manner over the long-term. How to achieve
sustainable intensiï¬ed agriculture is contested, with advocates promoting such various approaches
as organic agriculture, agroecology, and conservation farming (Petersen and Snapp 2015). At the
same time, there are a number of sustainability principles about which consensus is emerging that
12
can be followed regardless of the type of agriculture deployed (Robertson et al. 2014). Notable
among these is the case for perennial vegetation (Pimentel et al. 1986; Glover et al. 2010). This
type of plant trait ensures greater capture and conservation of resources, as living plants are con-
tinuously present to photosynthesize and foster biological processes such as nutrient recycling
(Jackson 2002). Thus, a range of alternative management practices, including organic and conser-
vation agriculture can potentially derive considerable beneï¬t from integration of perennials. In
contrast, agriculture as currently practiced involves crops with overwhelmingly annual growth
habits. This leads to managed lands left bare or unproductive for much of the year, as most food
crops are grown for a scant four to ï¬ve months (Jackson 2002). There are sustainable alternatives
to such annual-intensive production, which include mixed cropping systems that incorporate per-
ennial plants as companion species and the development and deployment of crop plants that have
perennial properties. Sorghum and pigeonpea are two such crops with perennial growth habits
that allow regrowth and multiple harvests through an agronomic practice called ratooning (Rogé
et al. 2016), which involves cutting the main stem(s) after reproductive maturity.
Based on agroecology and conservation agriculture principles, nature-based solutions offer
sustainable qualities, such as resilience to disturbance over variable time scales and minimal input
requirements, as well as an aim toward general improved human well-being (European Commis-
sion 2015; Bauduceau et al. 2015). Nature-based solutions for sustainable urbanization emerged
in the European Union as a research initiative with the explicit aim of addressing environmental,
social, and economic challenges sustainably (European Commission 2015; Bauduceau et al. 2015).
Similarly, smallholder agricultural systems will beneï¬t from adoption of technologies that support
sustainable principles and the complex system processes of nature. At the same time, the “net
beneï¬t of nature-based solutions depends on how much non-renewable energy can be replaced
13
without decreasing total production of ecosystem services†(Maes and Jacobs 2015: 3). Attention
to resilience and resource amelioration is often overlooked in agricultural discourse in favor of the
intensiï¬cation of high-yield annuals, which dominates the development lexicon. In contrast to
nature-based solutions that focus on economic solutions (e.g., fertilizer subsidy), nature-based
solutions might be better expressed in the smallholder agricultural context through sustainable
soil management strategies, ï¬eld/crop organization, and local market development.
2.2.3 Perennial Grains: A Nature-Based Approach for Africa
Pigeonpea (Cajanus cajan (L.) Millspaugh) is a perennial leguminous shrub commonly cultivated
and consumed by smallholder farmers in the tropical and often marginal environments of Africa
(Kumar Rao and Dart 1987). Pigeonpea is biologically suited to production as an intercrop, as a
slow initial growth habit minimizes competition for resources with a primary crop (Sakala et al.
2000). Farmers typically grow pigeonpea as a mixed cropping system, producing two or more food
products while simultaneously building soil fertility through biological processes (e.g., nitrogen
ï¬xation and carbon sequestration) (Fujita et al. 1992; Snapp et al. 2010). Sorghum (Sorghum bi-
color (L.) Moench) is a grass with perennial properties commonly grown by smallholder farmers
in semi-arid regions of Africa (Doggett 1988; Paterson et al. 2009). The cultivation of sorghum
occurs primarily in dryland areas and hot, semi-arid tropical environments (Dicko et al. 2006),
though there are sorghum species that grow in wetter regions (Harlan and Stemler 1976). Sorghum
has deep and spread roots and a solid stem (Buchanan, 1885), and it is common to grow sorghum
along ï¬eld ridges, as this helps prevent soil erosion (Okigbo and Greenland 1976). Pigeonpea and
sorghum are generally drought-tolerant and resilient in dryland environments, and in perennial
forms maintain these desirable traits (Parr et al. 1990).
14
A research gap that has not been addressed is the spatial assessment of suitability for di-
versiï¬cation of nature-based solutions. In this paper, we consider the phenology of two perennial
crops, pigeonpea and sorghum, both of which offer soil rehabilitation properties and are tolerant
to marginal environments (Okigbo and Greenland 1976; Snapp et al. 2010). The unique properties
of these crops enhance sustainability of cropping; however, they face biophysical constraints and
require an appropriate socio-economic context to achieve successful adoption. Our overall objec-
tive is to identify appropriate integration properties and delineate the climate niche for deployment
of these sustainable perennials within existing maize-based smallholder farms. From an agrocli-
matology perspective, we explore areas suitable for the cultivation of perennial sorghum and pi-
geonpea, and highlight the geographic potential for perennial grains to scale across Africa. Re-
motely-sensed imagery such as those presented here allow for global observation of the environ-
mental and biophysical factors that influence perennial development and adaptation and their
spatial organization. Identifying biogeographic conditions in which sorghum and pigeonpea can
prosper is crucial for effective integration and scaling of these perennial grains (Wood et al. 1999).
This work builds on other global and continental suitability maps (e.g., the Global Agro-
Ecological Zones (GAEZ) database (Fischer et al. 2008; IIASA/FAO 2012)) by highlighting the
intersection between marginal dominant crop conditions (in this case maize) and optimal perennial
crop conditions (in this case pigeonpea and sorghum), ultimately targeting areas where maize-
based systems are likely to beneï¬t from perennial integration. More, the data presented here are
unaltered and at the remote sensing pixel level, providing a direct link to the farmer experience
and offering a level of transparency (in terms of background variables) commonly masked by data
aggregations and complex zoning classiï¬cations. We use geospatial technologies and techniques
to (1) explore suitable (and optimal) climate conditions for pigeonpea and sorghum across Africa,
15
(2) assess marginal maize conditions and historic agricultural productivity, and (3) identify areas
where existing maize-based agricultural systems are likely to beneï¬t from integration of sorghum
or pigeonpea.
2.3 Methods
2.3.1 Fundamental Climate Niche: Maize, Pigeonpea, and Sorghum
The methodology presented here consists of three major components. First, we identify the fun-
damental climate niches for maize, pigeonpea, and sorghum. Second, we present a range of suita-
bility (and optimality) for each crop based on temperature and precipitation conditions. Third, we
locate the intersection between marginal maize areas and the optimal pigeonpea or sorghum cli-
mate niche.
The fundamental climate niche (i.e., temperature and precipitation) used for maize, pi-
geonpea, and sorghum is based on a literature review of tested crop performance (Table 2.1).
Pigeonpea is tolerant to variable rainfall, from very low to high amounts ranging from 350 to 2000
mm during a growing season (Kimani 2001; Valenzuela and Smith 2002; Wood and Moriniere
2013; Houérou, n.d.), and produces grain even during dry spells, unlike other legumes (Okiror
1986). The crop prefers temperatures ranging between 20.0 °C and 35.5 °C (Omanga et al. 1996;
Carberry et al. 2001; Silim and Omanga 2001; Sardana et al. 2010; Wood and Moriniere 2013).
Suitable growing season conditions for sorghum consist of rainfall between 150 and 950 mm
(Chipanshi et al. 2003; Du Plessis 2008; Mishra et al. 2008; Wood and Moriniere 2013) and tem-
peratures between 20.0 °C and 35.0 °C (Du Plessis 2008; Wood and Moriniere 2013; FAO, n.d.-
a).
16
Temperature (°C)
23.8–32.2
22.7–30.9
22.1–33.7
Crop
Maize*
Pigeonpea**
Sorghum***
Table 2.1. Optimal temperature and precipitation ranges for maize, pigeonpea, and sorghum. Sources: *(Pingali
2001; Wood and Moriniere 2013; Sanchez et al. 2014; Infonet 2016; FAO, n.d.-b). **(Omanga et
al. 1996; Kimani 2001; Carberry et al. 2001; Silim and Omanga 2001; Valenzuela and Smith 2002;
Sardana et al. 2010; Wood and Moriniere 2013; Houérou, n.d.). ***(Chipanshi et al. 2003; Du
Plessis 2008; Mishra et al. 2008; Wood and Moriniere 2013; FAO, n.d.-a)
Precipitation (mm)
750–1217
544–1263
317–833
Figure 2.1. Diagram of the maize, pigeonpea, and sorghum temperature and precipitation suitability classiï¬cation
thresholds. Classiï¬cations include optimal and suboptimal, which were further subcategorized into
a range of suitability: superoptimal, optimal, suitable 1 (S1), suitable 2 (S2), suitable 3 (S3), and
unsuitable. Maize, circle; pigeonpea, square; sorghum, diamond. Suitability classiï¬cation color
scheme used in Figs. 3 and 4.
For the optimal range, we calculated a mean low-value and mean high-value from all opti-
mal temperature and precipitation ranges collected in our review. From the values presented in
Table 2.1, there are two basic categories, optimal and suboptimal, which were further subcatego-
rized into a range of suitability: superoptimal, optimal, suitable level 1 (S1), suitable level 2 (S2),
suitable level 3 (S3), and unsuitable (Figure 2.1). We used equal intervals from the mean low-value
and high-value outward, depending on the range between the minimum or maximum value, to
17
designate thresholds for levels of suitability. We extrapolated one step inward (toward superopti-
mal) and one step outward (toward unsuitable) in order to highlight an additional level of spatial
variability. The same protocol was applied for classifying precipitation. The categories were con-
structed in this manner so that precipitation and temperature tolerance (or intolerance) would be
considered when moving outward from optimal.
2.3.2 Spatial Distribution of Climatic Variables
The spatial distribution of climate variability was acquired using two remotely-sensed NASA prod-
ucts. Temperature data were gathered from NASA MODIS Land Surface Temperature (LST—
MYD11B3) (NASA LP DAAC 2015a) and precipitation from NASA/JAXA Tropical Rainfall
Measuring Mission (TRMM—3B43) (NASA/JAXA TRMM 2016). Mean annual temperature and
accumulated annual rainfall values were calculated between 2003 and 2014 and a conditional state-
ment applied to classify the ranges of suitability as shown in Table 2.1. In order to generate suita-
bility maps that consider both temperature and precipitation, the individual suitability maps of
temperature and precipitation were overlain and reclassiï¬ed so that each pixel was represented by
the most unsuitable factor.
Marginal agricultural lands are those places with suboptimal temperature and precipitation
conditions for the dominant crop (e.g., maize), as well as places with chronically low production
levels and/or high interannual variability in production (Peter et al., in preparation). Marginal
maize agricultural land is identiï¬ed here using suboptimal temperature and precipitation ranges
for maize (Table 2.1, Figure 2.1), in combination with an interannual variability model that iden-
tiï¬es areas where agricultural production is historically low and/or highly variable (Peter et al., in
preparation). The model uses time-series Net Primary Productivity (NPP—MOD17A3) (NASA
LP DAAC 2015c) to categorize individual pixels based on regional productivity trends,
18
highlighting three levels of production and two types of interannual variability. Six results are
returned: Low-Stable (LS), Low-Variable (LV), Medium-Stable (MS), Medium-Variable (MV),
High-Stable (HS), High-Variable (HV) (Peter et al., in preparation). From these classiï¬cations, we
considered areas that are historically low and/or variable (LS and LV) to be marginal, as well as
areas that are historically medium and variable (MV). Combining suboptimal temperature and
precipitation conditions with marginal historic productivity reveals the spatial organization of var-
ious combinations of marginality and suboptimality, and serves to disentangle possible drivers of
underproduction (Figure 2.4). In order to delineate areas by agricultural land, we used a global
cropland percentage map developed by Fritz et al. (2015).
Areas likely to receive maximum beneï¬t from perennial deployment exist at the intersec-
tion between marginal dominant crop conditions (i.e., maize) and the optimal climate niche for
perennial grains (i.e., pigeonpea and/or sorghum) (Peter et al. 2017b). We identify this intersection
by overlaying optimal perennial pigeonpea or sorghum conditions with the marginal maize areas
identiï¬ed above (Figure 2.4). These are areas where perennial properties may serve to improve
existing maize-based agricultural systems.
2.4
Results
2.4.1 Spatial Organization of Crop Suitability
Suitable temperature and precipitation conditions for maize, pigeonpea, and sorghum are spatially
organized across Africa, following well-established agroecological zones (e.g., IIASA/FAO 2012)
(Figure 2.2). Marked temperature suitability differences exist in the Southern and Central regions
of the continent. Suitable precipitation ranges vary signiï¬cantly by crop, and clear regional delin-
eations are revealed when mapped. In terms of precipitation, sorghum, for example, is highly
suitable in Southern Africa, Northern Africa (i.e., Northern Algeria, Morocco, and Tunisia), parts
19
of Eastern Africa (i.e., Kenya and Tanzania), the Horn of Africa (i.e., Somalia, Ethiopia, Djibouti,
and Eritrea), and a horizontal stretch along the Sahel. Precipitation in these areas is uniquely suit-
able for sorghum and is largely suboptimal or unfavorable for pigeonpea and maize. Pigeonpea
has a breadth of rainfall suitability that covers much of Africa south of the Sahara. Unlike sorghum
and maize, pigeonpea suitability, particularly in terms of precipitation, covers a large land area in
the Central region of Africa (i.e., the Democratic Republic of Congo and adjacent countries).
Since suitable temperature ranges are similar among the crops under study, precipitation
is the primary determinant of continental-scale suitability; however, temperature does reveal its
importance when combined with precipitation, particularly at ï¬ne spatial scales (Figure 2.3). The
most notable distinction is that temperature restricts the suitability of sorghum in the Southern
tip of Africa, as well as the North western portion of the continent (i.e., Northern Algeria, Mo-
rocco, and Tunisia); maize and pigeonpea are affected in these areas, as well.
20
Figure 2.2. Temperature and precipitation suitability for maize, pigeonpea, and sorghum. Kenya is high-
lighted in order to show local spatial variability.
21
Figure 2.3. Combined suitability of temperature and precipitation suitability for maize, pigeonpea, and sorghum.
Kenya is highlighted in order to show local spatial variability.
Crop
Superoptimal Optimal
Maize
8.1%
Pigeonpea 11.3%
Sorghum
12.6%
Table 2.2. Agricultural land area proportions for each category (suitability classifications). Global cropland
percentage map by Fritz et al. (2015) used to delineate areas by agriculture.
Unsuitable
12.6%
5.7%
49.1%
20.0%
33.4%
11.7%
Suitable
Level 1
21.9%
26.4%
8.0%
Suitable
Level 2
22.0%
15.5%
9.5%
Suitable
Level 3
15.4%
7.7%
9.1%
Overall, we found that maize production is optimal on approximately 28.1 percent of ag-
ricultural land across Africa, suitable on 59.3 percent, and unsuitable on 12.6 percent (Figure 2.3,
Table 2.2). Pigeonpea is optimal on 44.7 percent of agricultural land across Africa, suitable on 49.6
percent, and unsuitable on 5.7 percent. Sorghum is optimal on 24.3 percent of agricultural land
across Africa, suitable on 26.6 percent, and unsuitable on 49.1 percent. Sorghum is the least suit-
able in terms of land area; however, conditions are optimal for its cultivation where maize and
pigeonpea are not. Optimal pigeonpea conditions occupy the greatest land area, and optimal con-
ditions for maize are surprisingly low given the extent to which it is cultivated.
22
2.4.2 Scaling Perennial Innovations for Maximum Beneï¬t
Within the optimal pigeonpea niche, 24.9 percent exhibits marginal agricultural productivity and
suboptimal rainfall conditions for maize, 41.6 percent exhibits sole marginal agricultural produc-
tivity, and 20.9 percent exhibits optimal conditions for maize (Figure 2.4, Table 2.3). Within the
optimal sorghum niche, 71.4 percent exhibits marginal agricultural productivity and suboptimal
rainfall conditions for maize, 18.2 percent exhibits sole marginal agricultural productivity, and only
4.6 percent exhibits optimal conditions for maize. Areas that exhibit sole marginal agricultural
productivity are areas where marginality may be driven by social or other biogeographic factors
(e.g., soil quality and land management practices). In areas where maize conditions are suboptimal
and/or agricultural productivity is marginal, and pigeonpea or sorghum climate conditions are
optimal, integration of these crops is likely to beneï¬t existing maize-based agricultural systems.
23
Figure 2.4. Intersection of marginal maize and optimal pigeonpea or sorghum. Areas presented in color
exhibit optimal temperature and precipitation conditions for pigeonpea or sorghum. Suboptimal
temperature and precipitation legend classiï¬cations are associated with the fundamental niche for
maize: T = suboptimal temperature; TP = suboptimal temperature and marginal productivity; R
= suboptimal rainfall; RP = suboptimal rainfall and marginal productivity; TR = suboptimal tem-
perature and suboptimal rainfall; P = marginal productivity; TRP = suboptimal temperature,
suboptimal rainfall, and marginal productivity; O = optimal; A = other agriculture (i.e., suboptimal
for pigeonpea or sorghum); N = non-agriculture. Global cropland percentage map by Fritz et al.
(2015) used to delineate areas by agriculture.
24
T
TP
1.2%
R
5.2%
5.4%
RP
24.9%
71.4%
Crop
Pigeonpea 4.3%
Sorghum <0.1% <0.1%
Table 2.3. Agricultural land area proportions for each category (marginal/optimal intersection). Proportions
presented only within the optimal pigeonpea or sorghum niche; other agricultural land not in-
cluded. Global cropland percentage map by Fritz et al. (2015) used to delineate areas by agricul-
ture.
TR
1.0%
<0.1%
O
20.9%
4.6%
P
41.6%
18.2%
TRP
1.0%
0.3%
2.5 Discussion and Conclusions
2.5.1 Promoting a Sustainable Future
Historically, most smallholder farmers in Africa have depended on intensively managed maize
systems as a cash crop and staple food (Byerlee and Heisey 1996); however, taking a closer look
reveals frequent and continual cultivation of, and dependence upon, pigeonpea and/or sorghum
for food security (Ruï¬no et al. 2013). Many African households are familiar with cultivation needs
of perennial grains through generational knowledge (Ruï¬no et al. 2013; Rogé et al. 2016), yet there
is little academic attention given to perennial management practices (e.g., ratooning) of perennial
grains (Kane et al. 2016). In Africa, there is potential to revitalize perennial crops such as pigeon-
pea and sorghum through innovative practices that offer multiple beneï¬ts (e.g., ecosystem im-
provements, crop water-use efficiency, and improved cultivars that enhance nutrition) (DeVries
and Toenniessen 2001; Cox et al. 2006; DeHaan et al. 2007).
Perennial grains offer beneï¬ts that encompass rural and marginal environment smallholder
farming contexts, and should be promoted in locations where labor constraints reduce agricultural
system efficiency, or where topography and farming practices increase erosion or depletion of soil
organic matter. Integration of perennial pigeonpea or sorghum into maize-based farming systems
shows wide spatial applicability across Africa. Though the optimal land area for sorghum cultiva-
tion appears of moderate size, low rainfall requirements permit it to ï¬ll a spatial niche that
25
pigeonpea does not. This regional delineation of crop suitability between sorghum and pigeonpea
is particularly evident along a horizontal stretch of the Sahel (Figure 2.4). This is a marginal area
for which there are few suitable crop options, and sorghum is often overlooked in a maize-centric
agricultural development agenda (Hellin et al. 2012). Overall, conditions are optimal for pigeonpea
cultivation on approximately 45 percent of agricultural land across Africa, and optimal for sor-
ghum on 24 percent (Figure 2.4). Between the two crops, approximately 53 percent of African
agricultural lands are likely to beneï¬t from their integration. There are certainly other perennials
that may be promoted in Africa (e.g., rice), and there is considerable potential to improve maize-
based farming systems through integration of semi- and fully perennial crops that provide multiple
ecosystem services (Glover et al. 2010).
Pigeonpea diversiï¬cation offers unique environmental beneï¬ts, particularly in locations
where soil fertility decline is limiting crop yields. The ability of pigeonpea to biologically ï¬x nitro-
gen has been proven to function in a smallholder farm context across a range of soil types, reha-
bilitating soils through input of 70–100 kg of nitrogen per hectare annually (Myaka et al. 2006).
Given the overlap between the pigeonpea and maize niche across large swaths of agricultural
lands, this could potentially address a substantial portion of the nitrogen deï¬cit that is a primary
yield limiting factor on African smallholder farms (Vanlauwe et al. 2011). Sorghum is more
drought-resistant than maize, yet it has been replaced by maize in areas such as Northern Malawi
(Bezner-Kerr 2014). Where it is still grown, farmers in Malawi ratoon by cutting the stem to extend
production for more than one season, and have reported that ratooning also saves labor and seeds
(Rogé et.al 2016). Smallholder farmers typically cultivate sorghum in mixed cropping systems
where the spatial arrangement of crops plays a major role. In Malawi, for example, farmers have
reported growing sorghum on soils that are marginal for maize (Rogé et al. 2016). In North-
26
Central Nigeria, sorghum has been reported to be intercropped in combinations with millet,
groundnut, and/or cowpea (Okigbo and Greenland 1976). Due to current global climate impacts
on crops, there have been partial efforts for breeding of sorghum (ICRISAT, n.d.), and research
shows that breeding sorghum can produce varieties adaptable to climate variability and pests
(Haussmann et al. 2012).
Despite the legacy cultivation of maize across much of Africa, and recent expansion into
new areas of West and North Africa (Blackie and Dixon 2016), there are large areas that are
suboptimal environments for maize growth. The decision to cultivate maize in these areas reflects
socioeconomic factors such as market demand and farmer preference, and may also be influenced
by the superior calories per unit input that maize offers (Snapp and Silim 2002). Interestingly, our
analysis shows that the Horn of Africa is largely unsuitable for maize production and optimal for
sorghum, yet maize is being widely promoted. One study found that one-third of farmers ratoon
sorghum for perennial production in lowland areas of Ethiopia (Mekbib 2009), yet surprisingly
little research focuses on this drought-adaptation option, even in the center of origin for the crop.
Overall, our ï¬ndings highlight the role that both sorghum and pigeonpea could play in improving
food security and resilience across marginal swaths of Africa. This is further evidence that the
crop has been understudied, particularly for novel perennial genotypes and historically important
ratooned types (Paterson et al. 2014).
2.5.2 Conclusions
Sustainable land management is at the forefront of agronomic research, and nature-based solu-
tions are emerging as long-term subsistence strategies. Following nature-based models, sustainable
agricultural solutions for smallholder farms are innovations inspired by nature. Though many na-
ture-based solution initiatives in the European context have been urban-centric (Nesshöver et al.
27
2016), the core sustainable intensiï¬cation concepts have broad potential. In rural Africa, where
land availability and fertilizer access are often limited, smallholder farming systems may beneï¬t
from the sustainable cropping methods and synergistic crop choices that nature-based solutions
promote. As arable, fertile land continues to diminish and population growth continues to drive
production demand, proactive strategies for increased and sustainable production need to be con-
sidered. There are certainly other perennials that may be promoted in Africa, and the potential to
improve maize-based farming systems through perennial integration is substantial, and underap-
preciated. Though we present the spatial suitability of maize, pigeonpea, and sorghum in Africa,
the heuristic proposed is generalizable and not limited to any particular set of crops or region. We
recommend that policy- and decision-makers utilize information and techniques such as these to
intelligently target areas that may beneï¬t from perennial integration. As pigeonpea, sorghum, and
other perennials are reintroduced to parts of Africa where perennial production has been ne-
glected in preference for maximizing maize yields, it is crucial to promote cultivation of these
environmental options in areas where climate conditions are conducive to prosperous and syner-
gistic growth.2
2 This research was supported by the Perennial Grain Crops for African Smallholder Farming Systems
project, Grant #OPP1076311, funded by the Bill & Melinda Gates Foundation, and through support of
the United States Agency for International Development, AID-OAA-A-13-00006. The opinions expressed
herein are those of the authors and do not necessarily reflect the views of the US Agency for International
Development or the US Government. We thank the organizers of the special issue for the invitation and
for highlighting the importance of this topic.
28
A MULTISCALAR APPROACH TO MAPPING MARGINAL AGRICULTURAL
LAND: SMALLHOLDER AGRICULTURE IN MALAWI3
3.1
Abstract
Marginal agricultural lands are defined here by suboptimal biophysical conditions and historically
variable or low agricultural production. We characterize these areas using remotely sensed infor-
mation to disentangle the biophysical and possible social factors driving marginality. Considering
both the modifiable areal unit problem and the ecological fallacy problem, the heuristic we pro-
pose is generalizable across geographies and scales and provides information at multiple decision-
making levels through a multiscalar interannual variability model. We present results from our
study of Malawi, where the landscape is densely cultivated and smallholder farmers frequently
occupy marginal lands, to illustrate the potential of a multiscalar analysis in a place where food
insecurity alleviation is needed and where remote sensing can provide necessary information. Our
framework for identifying marginal agricultural lands consists of (1) locating long-term agricultural
land, (2) measuring interannual productivity of long-term farmed locations, and (3) assessing mar-
ginal biophysical land characteristics and the fundamental climate niche for the dominant crop (in
this case maize). Productivity and marginality in Malawi are spatially organized, and an assessment
of productivity at multiple scales highlights the importance of presenting both global and local
spatiotemporal variability for managing agroecological variance. By disaggregating broad classes
of historically marginal production and the underlying drivers of marginality, different interven-
tion efforts can intelligently target areas most likely to receive maximum benefit. These
3 This is an Accepted Manuscript of an article published by Taylor & Francis in Annals of the American
Association of Geographers on 18 January 2018, available online at doi.org/10.1080/24694452.2017.1403877.
Reference: Peter, B.G., Messina, J.P. and Snapp, S.S., 2018. A Multiscalar Approach to Mapping Marginal
Agricultural Land: Smallholder Agriculture in Malawi. Annals of the American Association of Geogra-
phers, 108(4), pp.989-1005.
29
methodologies can be applied by both policy-makers and scholars to identify and target marginal
agricultural areas for improved productivity and for the support of smallholder farmer livelihoods.
3.2
Introduction
Marginal agricultural lands are dynamic and in many places and contexts represent the dominant
form. Whether through cause or consequence, sustainability of smallholder farming systems is
inextricably linked with agricultural marginality. Assuming a normative case for sustainable pro-
duction of most smallholder farms, marginal agricultural lands become so either by bringing mar-
ginal land into production or by adopting management models that degrade the respective poten-
tial of existing agricultural land.
Marginal agricultural lands exist as, or might emerge from, a complex combination of fac-
tors and rarely have a singular driver. The Consultative Group for International Agricultural Re-
search (CGIAR) defined land marginality as dependent on biophysical characteristics, proximity
to infrastructure, and population pressure (Nelson et al. 1999). Kang et al. (2013) characterized
marginal land as areas with low productivity and reduced economic return. Lantican et al. (2003)
referred to marginal land as “the world’s [unfavorable] dry and difficult cropping environmentsâ€
(353). Others consider marginal land to be entirely unsuitable and too fragile for cultivation (James
2010). Stimulating controversy, some national governments (e.g., Philippines, Cambodia, and
Ethiopia) conflate marginal land classifications with wastelands or barren land, partitioning them
off for corporations or agribusinesses that benefit the state and potentially negatively affect local
rural populations (e.g., displacement and reduced resource allocation; Borras et al. 2011; Nalepa
and Bauer 2012; Nalepa et al. 2017). Notably, “big data†methodologies have been used to identify
and classify global marginal lands as prospects for agrofuel development (e.g., Cai et al. 2011).
30
Pursuits to maintain and reclaim ideas about marginal land as a category for smallholder
agricultural improvement are present, and research efforts to identify and locate marginal lands
have proliferated in recent years (e.g., S. Wood et al. 1999; Gopalakrishnan et al. 2011). Still, mar-
ginal agricultural lands, particularly those in the smallholder rural context, are in need of more
research wherein the focus on marginal land classifications is more closely tied to smallholder
agricultural improvement, rather than partitioning for other activity. This is especially important
in rural Africa, where much of the smallholder agriculture takes place in marginal environments.
In Malawi, for example, a survey conducted by the Malawi National Statistical Office (NSO 2005)
estimated that approximately 88 percent of the population lives in rural areas with the majority
smallholder subsistence farmers (Chinsinga and O’Brien 2008). In sub-Saharan Africa, the Inter-
national Fund for Agricultural Development (IFAD 2001) estimated that 73 percent of the pop-
ulation were smallholder farmers (Morton 2007). Farming on marginal lands warrants particular
attention because (1) marginal land is frequently occupied by smallholder farmers dependent on
consistent production for subsistence, (2) considerable resources are required to transform mar-
ginal land out of marginality, and (3) marginal areas are particularly vulnerable to abrupt changes
in climate and at market (Morton 2007). Moreover, in densely populated and cultivated countries
such as Malawi, internal reorganization and occupation of marginal and protected lands remain
the only expansion or intensification options (Kalipeni 1996).
Despite increasing attention, marginal agricultural land is not a widely recognized land-use
class and is not identified as such on any of the commonly used land-use/land-cover (LULC)
maps. Even locating agriculture, particularly in low-intensity areas and smallholder farming con-
texts, presents a challenge. There are numerous land-use datasets readily available (e.g., global,
continental, and country specific), yet there are clear and nontrivial inconsistencies among them
31
(Messina et al. 2014). Haack et al. (2015) discussed inadequacies associated with both traditional
LULC classifications and remote sensing models. Inevitably, errors of commission (i.e., inclusion
of non-agriculture features within an area classified as agriculture) will propagate in most LULC
models.
Remote sensing has been used in myriad ways to identify and map marginal agricultural
lands. Gopalakrishnan et al. (2011) used remote sensing to map marginal agricultural lands in
Nebraska on the basis of soil health, current land use, and environmental degradation. S. Wood
et al. (1999) used remote sensing and other spatial data to identify marginal agricultural lands,
categorized by levels of suitability, with the explicit aim of targeting agricultural development. Cai
et al. (2011) used remote sensing to map cross-continent marginal agricultural lands by integrating
global data (i.e., soil properties, terrain characteristics, temperature, and humidity) in a fuzzy logic
model; in the final marginal agricultural land map, the majority of Malawi is classified as marginal.
Yield gap models have been used to identify marginal production areas (e.g., Neumann et al. 2010),
but crop simulation products often do not provide access to the biophysical and social drivers of
marginality.
Following on this research, there is a need to classify marginal land for the smallholder
farming context. Few studies have considered definitions that disentangle the underlying biophys-
ical factors (i.e., temperature, precipitation, and soil) driving agricultural marginality. Rather than
the typical singular marginal land classification, a multi-classification approach allows considera-
tion of all possible marginal combinations of the variables under study. This approach is critical
to inform policy and provide insights of value to sustainable development of smallholder agricul-
ture. For example, a parcel of land simply classified as marginal might be perceived as fit for
biofuel production (Cai et al. 2011), whereas the same parcel of land classified as marginal with
32
regard to temperature or precipitation might be ideal for promotion of climate-resilient varieties
of crops. Areas marginal in terms of soil fertility could be viewed as a prospect for intensification
of crop production through nutrient replenishment.
Moreover, the model presented here provides a heuristic to measure agricultural produc-
tivity at multiple scales to enhance decision making across administrative levels and geographies
and allows for land management across spatially variable agroecosystems. At global and country-
wide aggregations, remotely sensed measures of crop productivity could be indicative of global
processes and variable agroecosystems. By aggregating remotely sensed measures of crop produc-
tivity at local levels, marginality of agriculture could be assessed relative to shared agroecological
conditions; that is, areas of homogeneity at the country level become heterogeneous when ob-
served at a local scale (Sexton et al. 1998). With multiscalar information presented, national and
local administrative units (e.g., districts and extension planning areas in Malawi) can direct re-
sources and technologies in an informed manner, relative to actionable levels. The multiscalar
methodology presented here also allows for assessment of marginal agriculture across (agro) eco-
systems, where LULC (e.g., crop choices, tree type, tree density) is highly variable and inevitably
included in remote sensing spectral signals of agriculture.
To summarize, there are two overarching themes driving this research and two fundamen-
tal questions asked. First, classifications of marginal agricultural lands that include the underlying
drivers of marginality can serve to emphasize the smallholder farming context. Second, the scale
of observation matters for making decisions at actionable levels. Regarding these themes, we pose
two research questions: (1) What measurable, biophysical conditions constitute marginal agricul-
tural land across space, time, and scale? (2) Where are the variable types of marginal agricultural
land located?
33
3.2.1 Mapping Global Marginal Agricultural Lands for the Smallholder Context
Here, we define marginal agricultural land as land that has variable interannual production or is
chronically low-producing, driven by poor biogeographic conditions or social factors. This defi-
nition is directly translatable to the farmer experience and can be measured using remote sensing.
For subsistence farmers, low productivity translates to insufficient food supply and variable pro-
duction translates to unpredictable harvests. Marginality is further characterized here by crop cli-
mate niche (in this case maize) and soil suitability. The definition of marginal agricultural land
outlined is one that we argue can be measured using remotely sensed time-series information and
is generalizable across geographies and scales.
In this study, we identify the spatial organization of marginal agricultural land in Malawi
by disentangling the biophysical factors driving marginality, as well as possible social factors by
deduction (i.e., measured marginal productivity but no explanatory variable). We approach iden-
tifying these marginal areas through a strictly biophysical perspective, and the data we employ are
primarily remotely sensed products. Moving beyond the either–or debate, we present all possible
marginal combinations of the variables under study. Combinations of marginality presented here
include (1) suboptimal temperature, (2) suboptimal precipitation, (3) marginal soil suitability, and
(4) marginal interannual productivity. Additionally, we propose that measures of productivity and
interannual variability are relative to the scale and geography under observation. Our framework
consists of locating long-term agricultural land, measuring interannual productivity (and variabil-
ity) of long-term farmed locations, and assessing marginal biophysical land characteristics and the
fundamental climate niche for the dominant crop, in this case maize.
Because our definitions and measures of productivity are relative to the scale and geogra-
phy under observation and the data we employ are global, the methodology herein can be used to
34
reveal the spatial organization of marginality at any decision-making or administrative level. By
presenting information at multiple scales and administrative boundaries, we aim to manage some
issues manifested by the modifiable areal unit problem (MAUP) and the ecological fallacy prob-
lem. The MAUP arises when classifications or aggregations change based on differences in areal
selection and scale (Openshaw 1983). The ecological fallacy problem occurs when inferences are
made about an individual instance based on aggregate data (Openshaw 1983). The five major
components of the model are as follows: (1) locating long-term agricultural land and minimizing
errors of commission, (2) measuring historic productivity and interannual variability of agriculture,
(3) calibrating the model to field observations, (4) identifying crop niche and mapping marginal
land, and (5) extending the interannual variability model to possess multiscalar capabilities.
3.2.2 Malawi as a Bellwether for Identifying Marginal Agricultural Lands
Malawi is presented here as a bellwether for identifying marginal lands in the smallholder, rural
context. Historically, marginal agricultural lands in countries with low population densities move
into and out of production as prices change, labor shifts, and when climate signals encourage
planting and land clearing. Malawi faces different challenges. The country is largely settled with
very little suitable land available for new agriculture, and dense population reduces farm size and
limits fallowing options (Ricker-Gilbert and Jumbe 2014). Moreover, the Malawian landscape is
mosaicked with mixed land use, regional land-use variance, and relatively small fields, making ag-
ricultural land difficult to discern from other land-use types in most singular global land-cover
products. This long-term use of marginal lands and the spatial arrangement of cultivation is
unique, and marginal agricultural lands in Malawi are particularly important because the current
intensive use reflects population pressures that will emerge in many other regional countries. In
neighboring Zambia, for example, the coupling of large-scale investments in agricultural land
35
(Nolte 2014) and population increases will restrict fallowing options for smallholder farmers, de-
manding more food production on already continually harvested fields (Schneider et al. 2011).
Information-driven targeting for new or different agricultural practices is essential to suc-
cessful scaling (S. Wood et al. 1999), and adoption will only take place where biophysically feasible
(Kwesiga et al. 2003). To date, efforts to identify marginal agricultural lands specific to Malawi are
thinly scattered. Remotely sensed data offer a solution to this problem, particularly in areas where
fine-resolution crop information is unavailable. Depending on the platform, such data can be cost-
effective and cover entire landscapes, improving on regional- or administrative-level development
strategies. Remotely sensed imagery can also be used to measure crop response to various devel-
opment strategies. The Fertilizer Input Subsidy Programme (FISP) in Malawi, for example, was
popularly recognized as initiating an “African green revolution†(Sanchez 2015), but recent studies
suggest that it might not have been effective at reaching those most rural, and production might
have been overestimated (Holden and Lunduka 2012; Chinsinga and Poulton 2014; Messina et al.
2017). Moreover, remote sensing offers a partial solution to targeting resource delivery. Recurring
food insecurity in Malawi and the country’s position in the global market have prompted regular
aid initiatives from both nongovernmental organizations and international governments; however,
resource allocation is the subject of much scrutiny and resource need might not always be a major
driver in resource allocation across all sectors (Dionne et al. 2013).
Precipitation Soil suitability Temperature
Source
NASA/JAXA Li et al. 2017a
Product
TRMM 3B43 —
Spatial resolution
~27-km
Temporal resolution Monthly
Table 3.1. Biogeographic data sets used to determine marginality and suboptimality of maize agriculture.
NASA MODIS NASA MODIS
MOD11A2
1-km
8-day
90-m
—
Productivity
MOD17A3
1-km
Annual
36
3.2.3 The Malawian Landscape: Cultivation, Climate, and Population
In Malawi, “maize is lifeâ€; it is the most frequently cultivated and consumed cereal crop in the
country, commonly taking form as nsima, a dried, finely ground, and reconstituted dense por-
ridge—a staple of the Malawian diet. Production of maize is widely distributed throughout Malawi
and, according to the United Nations World Food Programme (WFP 2016), 80 percent of Mala-
wians are smallholder subsistence farmers. Of the farming population, approximately 97 percent
of farming households cultivate maize, and its consumption comprises 60 percent of the total
caloric intake (Denning et al. 2009). According to Kassie et al. (2012), 70 percent of smallholder
agricultural land is cultivated for maize production. Smallholder farmers rely on the success of the
maize harvest for subsistence, and national food security depends on consistent domestic agricul-
tural productivity. Presently, pressures from rising population, climate change, unsustainable agri-
cultural practices, and deterioration of natural resources all pose major threats to Malawi’s maize
production (Postel 2000; Godfray et al. 2010).
There are underlying agricultural sustainability concerns associated with widespread, con-
tinuous production of maize with limited use of soil amendments (e.g., fertilizer; Thierfelder et al.
2013). Malawi’s long history of unsustainable agricultural practices (e.g., monocrop maize systems,
residue mismanagement, and extensive cultivation; Glasson 2010; Thierfelder et al. 2013; Mungai
et al. 2016) has contributed to widespread soil nutrient and organic matter depletion (Sanchez
2002; Ngwira et al. 2012). Monocropping of maize with little or no inputs is common throughout
Malawi, resulting in diminishing yields over the last several decades (Thierfelder et al. 2013). Res-
idue management is variable across Malawi, depending on the presence of livestock and access to
other biomass sources, and legumes are commonly neglected as a source for residue biomass
(Valbuena et al. 2015; Mungai et al. 2016). Intercropping (e.g., maize with pigeonpea) and other
37
rehabilitation strategies (e.g., conservation agriculture) are frequently offered through extension
efforts, but adoption is generally poor or might be rejected outright (Schulz et al. 2003; Bezner-
Kerr et al. 2007; Snapp et al. 2010). Despite implementation of the FISP, an initiative put forth by
the Malawian government in response to record low production, soil degradation remains a wide-
spread concern and food insecurity persists (Ecker and Qaim 2011; Messina et al. 2017).
Demands for increased production continue as Malawi’s population continues to grow.
On average, the national population increased by approximately 2.9 percent each year between
2000 and 2014 (World Bank 2016). Concerns related to population growth in Malawi are not new.
In fact, House and Zimalirana (1992) warned of Malawi’s rapid growth, lack of resources (e.g.,
access to fertilizer, seed, capital, and labor), and pressures on the land (and human needs) before
the turn of the century, when the population was only 9.7 million (World Bank 2016). In 2014,
Malawi reached a total population of 16.7 million a 3.1 percent increase from the previous year
and up from 11.2 million in 2000 (World Bank 2016). Such population pressures and increased
production demands have contributed to the exhaustion of agricultural lands in Malawi, largely
due to over-cultivation of maize (Manyonda 1998; Thierfelder et al. 2013).
Unpredictable climate events can lead to significant fluctuations in agricultural output, af-
fecting both smallholder livelihoods and national food security. During the last two decades, Ma-
lawi has been afflicted by both severe droughts and floods, threatening the nation’s ability to sus-
tain its large, growing population (Devereux 2007; Pauw et al. 2011). In 2002, Malawi was stricken
with a record drought (Syroka et al. 2010), resulting in 32 percent less reported production than
the previous maize harvest, placing the country in a food insecure “state of disaster†(Devereux
2002; Frize 2002). In 2005, Malawi “faced its worst food crisis in more than a decade,†with more
than 34 percent of the population in need of food aid (Buerkle 2005). In 2015, record floods
38
displaced nearly 70,000 people (Banda 2015). In 2016, Malawi’s President Peter Mutharika de-
clared a national disaster in response to prolonged drought and a devastated maize harvest (Phiri
and Macharia 2016). Twenty-five of Malawi’s twenty-eight districts were affected, leaving the
country with a 1.2 million metric ton maize deficit (Kandaya 2016; Phiri and Macharia 2016).
Overall, droughts and floods contribute to an approximately 1.7 percent loss in gross domestic
product each year (Pauw et al. 2011). Moreover, seasonal shifts and changes in rainfall pattern
leave the optimal timing of planting and fertilizing uncertain each year (Tadross et al. 2009; Ex-
tension Planning Area Officers, personal communication 2015). Smallholder farmers occupying
marginal agricultural areas are most vulnerable to abrupt climatic and biophysical changes, as well
as socioeconomic shifts (e.g., fertilizer and seed access, poverty, and health). These issues are not
expected to subside or resolve with imminent global environmental change and steady population
growth.
Product
Global Land Cover
GlobCover
Temporal range Spatial resolution Source
2000
2005–2006, 2009
986-m
920-m
ESA (2005)
ESA (2008);
ESA (2010)
IFPRI (2002)
FAO (2013)
2000 & 2010
World land use–land cover 2002
Land Cover and Land
Cover Change of Malawi
NASA MODIS Land
Cover Type (MCD12Q1)
Table 3.2. Land-use products incorporated in errors of commission-managed agricultural land-use map. Tem-
poral range under study: 2000–2014.
NASA LP
DAAC (2015d)
2001–2012
920-m
30-m
500-m
3.3 Methods
3.3.1 Remotely Sensed Data Products and Preprocessing
Precipitation data were acquired from the NASA/ JAXA Tropical Rainfall Measuring Mission
(TRMM–3B43) via the NASA Mirador Web portal (Table 3.1; NASA/JAXA TRMM 2016).
39
TRMM data require a precipitation rate conversion to retrieve monthly rainfall amounts. We con-
verted monthly precipitation rates to accumulated growing season rainfall (November–April;
Jayanthi et al. 2013; FAO 2016b) for each year between 2000 and 2014. For both temperature and
precipitation, we opted to use only growing season data so that postharvest measures were not
included (Zhang and Oweis 1999).
Soil suitability was assessed using a product developed by Li et al. (2017)—a multimethod
approach utilizing a suite of landform characteristics and soil properties (i.e., slope, erosion hazard,
organic matter, cation exchange capacity, texture, pH, and drainage) to score and rank soil suita-
bility. There are two marginal categories—marginally suitable and marginally unsuitable. Other
categories include highly suitable, moderately suitable, and permanently unsuitable (e.g., forest
reserves or urban areas; Li et al. 2017a). Land surface temperature (LST) data were acquired from
the NASA Moderate Resolution Imaging Spectroradiometer (MODIS–MOD11A2) via the NASA
Reverb Web portal (NASA LP DAAC 2015b). Annual productivity (net primary productivity
[NPP]) data were acquired from NASA MODIS (MOD17A3) via the NASA Reverb Web portal
(NASA LP DAAC 2015c). For the MODIS products, we selected the Terra satellite for its daytime
temperature accuracy in southeast Africa (DeVisser and Messina 2013). We elected to use the
annual NPP product because the trendline tracks reported yields more closely than growing season
gross primary productivity (GPP) (Messina et al. 2017).
To identify long-term farmed locations, we developed a composite agricultural land-use
map utilizing a suite of widely recognized products (GLC2000, GlobCover 2005/06 and 2009,
IFPRI 2002, FAO 2000 and 2010, and NASA MODIS LC 2001–2012; Table 3.2). We elected to
use the MODIS Land Cover Type product (MCD12Q1) because the temperature and productivity
measures used here come from the MODIS sensor and it covers a large temporal range (2001–
40
2013; NASA LP DAAC 2015d); GlobCover because it is well recognized and covers two time
steps within the temporal range under study (2000–2014; ESA 2008 2010); GLC2000 because it
is well documented and widely used (ESA 2005); IFPRI because it has a relatively unconstrained
definition of agriculture (i.e., we know from field experience that Malawi is cultivated extensively,
and this product reflects a high amount of agricultural land area; IFPRI 2002); and FAO because
it offers a fine spatial resolution Malawi-specific data set and covers two time steps (2000 and
2010) within the temporal range under study (2000– 2014; FAO 2013; see Table 3.2).
3.3.2 Locating Long-Term Agricultural Land: Minimizing Errors of Commission
Before identifying marginal agricultural land, all agricultural lands across the production spectrum
must be identified. The approach taken here shares conceptual and methodological similarities
(i.e., integration of a suite of land-cover products) with those of a global cropland percentage map
developed by Fritz et al. (2015), as well as a global cropland extent map produced by Pittman et
al. (2010); however, we developed our methodology independently and the output categorizations
are of a different type. The logic here is that locating areas where multiple land-cover products
agree (across space and time) will illuminate areas that are routinely and intensively cultivated,
exhibit minimal land-use mixing, and have been historically used for agriculture.
Each land-cover classification system (Table 3.2) was reclassified into a binary system (ag-
riculture or non-agriculture), resampled (nearest neighbor resampling method) to the finest reso-
lution of the inputs (30 m), and aggregated (overlain and summed) to create one land-use map.
The resulting product is an ordinal ranking of values ranging from zero to five; zero represents
areas where none of the classification systems classify an area as agricultural land and five repre-
sents an area where all five classification systems agree that a particular area is agriculture. Finally,
majority resampling was used to generate rankings at a 1-km spatial resolution, coinciding with
41
the resolution of many MODIS products (e.g., LST and NPP; Figure 3.1). To represent the actual
land area devoted to agricultural production, we selected a number of pixels equal to the average
FAO-reported agricultural area between 2000 and 2013—approximately 5.25 million hectares
(52,510 km2 pixels or 55.7 percent of the total land area; FAOSTAT 2016).
We used all pixels from ranks 3, 4, and 5, and added in 7,072 unconstrained random pixels
from rank 2 so that the total agricultural area would equal 52,510 km2. We refer to the final product
as our errors of commission-managed agricultural land-use map. This grid was used to demarcate
agriculture in the marginal agricultural land model. There are undoubtedly marginal agricultural
lands that exist within ranks 1 and 2, and biophysical suitability could be assessed; however, his-
toric productivity in these areas cannot be measured confidently because of changes in land use
over time and the resolution of the NPP product used here (1-km). These areas commonly exhibit
mosaicked land use, particularly mixed forest and agriculture (e.g., Kasungu National Forest in
central-western Malawi).
42
Figure 3.1. Errors of commission–managed agricultural land–use map. 1-km spatial resolution.
3.3.3 Measuring Historic Productivity and Interannual Variability of Agriculture
Marginal agricultural lands are defined here by suboptimal biophysical conditions and historically
variable or low agricultural production. Using this definition, we created a model that categorizes
time-series data, classifying individual pixels relative to spatially structured distributions and ad-
ministrative delineations. Our methodology considers three levels of production (low, medium,
or high) and two types of interannual variability (stable or variable). We use NPP (kg C m-2) as our
indicator of production because it has a direct relationship with agricultural output (Heinsch et al.
2003; Reeves et al. 2005; Lobell 2013a; Messina et al. 2017).
First, to generate levels of production, we calculated the average value of each pixel using
the time-series NPP (2000–2014). Any pixel with a value of zero for any year between 2000 and
43
2014 was excluded so that only land under regular cultivation would be considered; this involves
excluding pixels that were not productive for any range of years within the time frame. The level
of production is derived from a quantile (equal area) classification algorithm; we use three catego-
ries to represent low, medium, and high levels of production. Optionally, the model allows for
selection of alternative classification systems (e.g., Jenks [1967] natural breaks). Second, to meas-
ure interannual variability, we calculated the variance of each pixel using the time-series NPP
(2000–2014). We used the quantile classification algorithm with two categories to disaggregate the
two temporal trends (variable or stable). Again, the classification system here is optional. The
resulting six categories are low-stable (LS), low-variable (LV), medium-stable (MS), medium-vari-
able (MV), high-stable (HS), and high-variable (HV) (Figure 3.2).
Figure 3.2. Illustrative diagram of the six interannual variability trends. Net primary productivity trends
measured between 2000 and 2014 for all of Malawi and individual select depictive pixels. The
regional trend (shown as a dotted line) is the average of all pixels within the selected region (Malawi
in this case). Gray and black continuous lines are select pixels that exhibit the six types of trends.
Temperature (°C)
23.8 > Mm > 32.2
Crop
Maize (marginal)
Table 3.3. Marginal temperature and precipitation parameters for maize. Note: Temperature data acquired
from NASA MODIS LST (MOD11A2). Precipitation data acquired from NASA/JAXA TRMM
(3B43). Precipitation requirements are represented as accumulated growing season rainfall. Pingali
(2001), L. Wood and Moriniere (2013), Sanchez et al. (2014). L. Wood and Moriniere (2013),
Infonet (2016), FAO (n.d.-b).
Precipitation (mm)
750 > Mm > 1217
44
3.3.4 Calibrating the Model to Field Observations
In 2015, prior to the development of the marginal land model, we visited 200 sites in thirty-three
farming regions across the extent of Malawi to conduct a land-cover assessment. We devised a
stratified random sampling scheme binned by distance to roads and access to local authorities (i.e.,
extension planning area officers). Extension planning areas (EPAs) are local administrative zones
situated within districts. We visited sites ranked between 3 and 5 on the errors of commission–
managed agricultural land-use map to increase the likelihood of sampling on agriculture under
regular cultivation. Interviews were held with farmers and EPA officers for inquiry of the primary
drivers (and limiters) of production, as well as crops grown and land-use history.
Agriculture dominates the Malawian landscape. Of the 200 sites we visited, we found that
only ten were either nonagricultural or experienced dominant land-cover change since 2000. Of
these ten, only five were misclassified (one barren or forest, one village center, two pastures, and
one reed field), two were sparsely cultivated (one mountainous region, one rapidly urbanizing),
and three experienced major land-use change (two became maize agriculture, one became govern-
ment housing for police; Messina et al. 2017). Between ranks 3 and 5 on the errors of commission–
managed agricultural land-use map, 95.0 percent (190/200) of locations visited were long-term
farmed locations; 91.5 percent (183/200) were long-term maize dominant. The seven fields not
dominated by maize were sugarcane (three) and tobacco (two) estates and rice irrigation (two).
During our interviews, we found that the majority of responses for primary drivers of
production were related to soil quality or suboptimal precipitation. Temperature and agricultural
management practices were seldom reported as limiting factors for production (<1 percent and 1
percent, respectively). The most commonly driven narrative of yield limitations in Malawi is soil
degradation (i.e., loss of nutrients and soil organic matter). Approximately 50 percent of
45
interviewees reported that soil quality (i.e., soil degradation, lack of fertilizer, soil infertility, soil
suitability, or soil type) was the primary driver limiting yield, with occasional secondary and tertiary
factors; 75 percent overall reported soil quality as one of the drivers of low production. Another
common observation in Malawi, which is gaining increasing attention in the literature, is that the
rainy season is shortening (i.e., starting later and ending earlier; Tadross et al. 2009; Vizy et al.
2015; EPA Officers, personal communication 2015; Sutcliffe et al. 2016) and that rainfall patterns
are erratic, yet the overall amount of precipitation has not changed drastically (EPA Officers,
personal communication 2015). A sole mean descriptive statistic does not capture this phenome-
non. Approximately 50 percent of interviewees reported erratic rains or changing rainfall patterns
as one of the drivers limiting yields; 40 percent of interviewees reported rainfall (erratic rains or
precipitation amount) as the primary driver limiting production. For this reason, we include rain-
fall variability in our definition of suboptimal precipitation (refer to the appendix for more discus-
sion on this topic).
46
Figure 3.3. Multiscalar interannual variability and marginal agricultural land model. *The zones input
provides an optional multiscalar calculation. **The precipitation variance was calculated here
uniquely for Malawi. NPP = net primary productivity; LULC = land use–land cover.
3.3.5 Identifying Crop Niche and Mapping Marginal Maize Agricultural Land
Marginal maize agricultural land is defined here, in part, by suboptimal LST and suboptimal pre-
cipitation. We adopted widely accepted temperature and precipitation ranges for maize to deter-
mine suboptimality (Table 3.3). We calculated a singular average LST value (per pixel) across all
2000 to 2014 growing seasons and, using the parameters from Table 3.3, created a binary grid for
temperature—either suboptimal or not. Our definition of suboptimal precipitation is a coupling
of variable precipitation (determined using our interannual variability model; quantile classifica-
tion, two classes—either variable or not) and accumulated rainfall outside the optimal range of
values (Table 3.3).
In terms of productivity, we consider LS, LV, and MV (gathered from the interannual
variability model) as marginal categories. The soil suitability grid provides various levels of
47
suitability, two of which are marginal (marginally suitable and marginally unsuitable; Li et al.
2017a). Interannual productivity and soil suitability were reclassified into binary grids—either mar-
ginal or not. All marginal and suboptimal grids were combined so that all possible combinations
of marginal maize agriculture (temperature, precipitation, soil suitability, and productivity) would
be revealed (Figure 3.3). Input parameters are time-series spatial data (raster data structure), clas-
sification method, and boundary data; worth mentioning is that the model is not restricted to NPP
and any spatial data could be considered (e.g., rainfall and temperature).
48
Figure 3.4. Multiscalar interannual variability: low-stable (light orange), low-variable (dark orange), medium-
stable (light blue), medium-variable (dark blue), high-stable (light green), and high-variable (dark green). Fritz et
al.’s (2015) global cropland percentage map used to delineate areas by agriculture on the global
and continental maps; the errors of commission–managed agricultural land-use map was used for
Malawi. Scale bar applies to Malawi inset maps. EPA = extension planning area.
49
3.3.6 Devising a Multiscalar Marginal Land Model
We adapted the productivity and interannual variability methodology into a dynamic, multiscalar
model. Decision making often takes place at varying administrative levels and performing mul-
tiscalar analyses can provide additional levels of insight. The model is generalizable across all ge-
ographies, scales, and temporal ranges and is equipped to automatically generate classifications for
any geographical boundary and subset boundaries (i.e., country, district, or EPA). The zones input
is an optional selection that will generate the six-classification (LS, LV, MS, MV, HS, and HV)
map for each subset region (e.g., country, district, or EPA; Figure 3.4).
3.4
Results and Discussion
3.4.1 Multiscalar Interannual Productivity and Variability of Agriculture
Agricultural productivity across Malawi is spatially organized and relative to the scale under ob-
servation. There is a clear regional delineation between relative high primary productivity in the
north and low productivity in the south. The central region of Malawi is a mix of low and medium
productivity. The district- and EPA-level maps reveal additional levels of information (i.e., areas
that are homogenous on the country-wide map exhibit heterogeneity at the district level; Figure
3.4). Important to acknowledge is that productivity on a global and continental scale looks expect-
edly similar to precipitation maps. This is due to global processes driving productivity (precipita-
tion is a calculation input for NPP) and the homogenization of data based on scale. Modifying the
scale to the country, district, and EPA levels reveals local processes driving variability (e.g., topog-
raphy, precipitation, temperature, and soil).
3.4.2 Map of Marginal Maize Agricultural Land in Malawi
Marginal conditions for maize are spatially structured in Malawi. Presented are all possible com-
binations of marginality and the amount of land area each category occupies (Figure 3.5).
50
Suboptimal temperatures and marginal productivity are prominent in the central region, marginal
productivity and marginal soil in the south, and marginal soil and suboptimal temperatures in the
north. Clusters of other combinations are scattered throughout. We found that marginal soils
(pertaining to any agricultural activity) occupy 40 percent of Malawi’s agricultural land; suboptimal
temperatures (pertaining to maize) occupy 52 percent of agricultural land; and suboptimal precip-
itation (pertaining to maize) occupies 59 percent of agricultural land (approximately 19 percent of
the suboptimal precipitation is the result of too much rainfall). Forty-seven percent of agricultural
land is characterized by marginal interannual variability and marginal production levels. Despite
maize being the dominant crop, only 5 percent of Malawi’s total agricultural area is optimal for
maize. Also uncovered in this map are possible regions where social factors are driving marginality.
These are areas that have marginal productivity not explained by one of the biophysical drivers
under study, classified solely as marginal productivity (category P, 5 percent of the agricultural
land area; Figure 3.5). These areas might benefit from extension and market development.
51
Figure 3.5. Marginal maize agricultural land (2000–2014). All possible marginal combinations of the
variables under study: productivity (P), soil (S), temperature (T), and rainfall (R). Percentages in-
dicate the proportion of agricultural land that each combination occupies. Letter codes for the
various combinations of marginality are listed below the legend. Colors presented are semi-arbi-
trary and nonhierarchical.
3.4.3 Geographic Themes
There are two distinctly geographic themes underlying this work: the MAUP and the ecological
fallacy problem. To recount, the MAUP arises when classifications or aggregations change based
on differences in areal selection and scale, and the ecological fallacy problem occurs when infer-
ences are made about an individual instance based on aggregate data (Openshaw 1983). Consider
52
the Malawi country-level map, for example (Figure 3.4). If there are resources available for distri-
bution (e.g., fertilizer or seed subsidies), the southern region with LV interannual variability seems
to be an ideal area for focus. Indeed, the southern region warrants particular focus; however, we
know that in Malawi, the rural poor across the country are facing similar challenges at local levels.
The district- and EPA-level maps manage the country-level homogenization and identify the mar-
ginal agricultural lands with respect to variable regional agroecosystems.
Regional land-cover and land-use variation accounts for much of the country-level varia-
tion in the productivity signal. For example, fields in the northern region of Malawi are largely
composed of maize and cassava, whereas the central region is dominated by maize supplemented
with groundnut and beans. In the southern region, maize-based systems commonly include pi-
geonpea (VAC 2005). The multiscalar map presented here accounts for these regional variations
in LULC by presenting measures relative to the differing regional conditions and agroecosystems.
In terms of policy, if there is to be an equitable spatial distribution of resources, it is critical to
understand variability at multiple aggregations (i.e., country, district, and EPA). If resources (e.g.,
fertilizer or seed packets) are available for distribution in each district, the question then would be
where to deliver the resources within each district. A district-level aggregation provides the nec-
essary information that the country level does not. Without maps like these, policy decisions would
be made with incomplete information. Because much agriculture development in Malawi takes
place at the EPA level, information at this scale can equip extension officers with additional tar-
geting capabilities.
The MAUP is indeed illuminated here, and we argue that taking a multiscalar approach is
one way of managing the problem. There is no single global standard or threshold that would
effectively describe low, medium, and high productivity across geographies or scales. If we were
53
to consider the Africa-wide map on its own, nuanced productivity within Malawi would not be
observed. Similarly, if observations are only made at the country level, within-district heterogeneity
would be missed. Ultimately, no scalar presentation of marginality is better than another irrespec-
tive of context, but depending on resource allocation or research purpose, one map (or a suite of
maps) might be better equipped for the task.
3.4.4 Reclaiming Marginal Lands
To date, the concept of marginal land has been elaborated largely in the context of biofuel advo-
cacy within both academia and government, with some attention given to assessing where biofuel
crops might be grown without disrupting agricultural subsistence and income-generating crops
(Cai, Zhang, and Wang 2011). Biofuel advocacy ignores the pivotal role that marginal lands play
in agricultural production across sub-Saharan Africa and beyond. As global population continues
to increase and land under cultivation is exhausted, more attention must be paid to how to manage
marginal agricultural land for improved productivity and sustainability.
Agrobiodiverse, sustainable agricultural practices suited to conserving marginal agricultural
lands in a productive manner have become an active area of research in the African development
context (Khumalo et al. 2012). Specific examples of how marginal land mapping can help target
development strategies are illustrated here for Malawi. Technologies for soil rehabilitation include
pigeonpea-based systems, for example, as a tool to foster more resilient maize production (Snapp
et al. 2010; Ngwira et al. 2012; Mhango et al. 2013). Illuminating areas that are marginal for maize
agriculture (driven by poor soils) but that are suitable for pigeonpea is a novel approach for iden-
tifying locations where targeted policy and education would have maximum benefit. Areas where
marginality is driven by temperature or precipitation might benefit from climate-resilient crop
varieties. Although we focus on maize here, the model is equipped to assess marginal agricultural
54
land for any crop. Marginal lands identified here might also be candidates for promotion of agro-
forestry, particularly those where human population density is sufficient enough to support the
labor requirements of planting, pruning, and managing trees within agricultural landscapes. The
protocols presented here can identify the intersection of marginal drivers and suitable conditions
for select intervention efforts and crops.
More attention must be given to the potential for remote sensing to characterize marginal
land as a prospect for increased yields, improved smallholder farmer subsistence, and overall long-
term agricultural system improvement in Malawi. Intervention targeting (i.e., bringing develop-
ment initiatives to places in need of support and where benefits are predictable) is often over-
looked and seldom mapped. As mentioned, one prospect in Malawi is targeted integration of pi-
geonpea in maize-based cropping systems for soil rehabilitation in locations where conditions are
optimal for its cultivation. Remotely sensed spatial information and maps such as those presented
here can provide necessary information to target areas in need, particularly in areas most rural
where other forms of climate information are unavailable. Moreover, we see considerable value in
addressing the biophysical and possible socioeconomic constraints at multiple scales, as illustrated
here. In doing so, we aim to overcome some of the hurdles associated with coarse data and char-
acterization of marginal lands by applying big data to a smallholder farming context.
3.4.5 Social Factors Driving Marginal Productivity
The marginal agricultural land map presented here highlights areas where social factors might be
playing a role in marginal productivity. Remotely sensed data do not explicitly capture social driv-
ers of farm productivity (e.g., exogenous economic factors or endogenous agricultural manage-
ment practices); however, we hypothesize that areas where interannual productivity is low or var-
iable but biophysical conditions are optimal are areas where social factors might be dominant
55
factors in marginal productivity. These areas (category P) represent 5 percent of agricultural land
on the marginal maize agricultural land map (Figure 3.5). Soil suitability is invariably linked to
management practices; however, they are categorized separately here because (1) the soil suitability
map used here does not cover a temporal range, making management practices difficult to untan-
gle or verify; (2) the product itself is largely based on terrain characteristics and gross single meas-
urement period soil characteristics; and (3) soil degradation or soil quality (e.g., soil texture) might
be beyond the management capacity of the individual farmer. Moreover, it is often the case that
farmers are cultivating and managing fields in concordance with agricultural extension-recom-
mended practices, but soil suitability is a production driver often operating on timescales beyond
the seasonal or soil degradation is embedded in the soil suitability feedback system. Bringing mar-
ginal land out of marginality or degrading status requires resources that are nontrivial to obtain.
Somewhat surprising temperature and agricultural management practices were rarely re-
ported as drivers of marginality (<1 percent and 1 percent, respectively). The most commonly
driven narrative of yield limitations in Malawi was soil degradation (i.e., loss of nutrients and soil
organic matter). Importantly, there is a prevailing narrative of soil degradation and changing rain-
fall patterns that is delivered from the top down in Malawi (country/district/EPA). EPA officers
carry out missions in concordance with government objectives and soil and precipitation re-
sponses might have been given when the primary limiting factor of production was not known. It
can certainly be argued that current and recommended management practices are suboptimal, thus
contributing to continual soil degradation; however, this situation would not elicit an interviewee
response for inadequate management practices. Agricultural extension officers typically referred
to management practices as a primary driver of low and variable yields when a farmer did not
implement advice from extension, although sometimes insufficient labor or capital were the
56
reasons. Still, it is possible that another factor not presented here is driving marginality. We rec-
ommend that areas where interannual productivity is marginal but all biophysical conditions under
study are optimal (i.e., temperature, precipitation, and soil suitability) receive further research at-
tention to unpack the social variables (or other biophysical drivers) that might be contributing to
marginal production in these areas. In this case, market conditions might be one possible avenue
for future research. Areas where interannual productivity is marginal but temperature and precip-
itation are optimal (and soil suitability is marginal) could also be considered for research on farm
management.
Despite widespread maize agriculture in Malawi, conditions are not optimal for its cultiva-
tion. This phenomenon is widespread, as many crops are grown far outside of their optimal niche,
in terms of both environment and socioeconomic contexts. An increasingly narrow range of crops
is dominant in the region and across the African continent. Maize is a clear example, and it con-
tinues to expand in production area, becoming a larger presence in semiarid areas of West Africa
and Ethiopia, which historically did not grow maize due to local food preferences. This raises
questions related to which agronomic traits support this dominance, such as biological properties,
which we hypothesize might be related to a superior ability to convert photosynthate into calories,
and robust tolerance to weeds and other pests. Socioeconomic conditions might also favor some
crops over others, such as market price support and the fertilizer subsidies that are critical for
production of cereals. Resilience of farming systems and the environment are supported by agro-
biodiversity, and research attention is urgently needed to understand the drivers and geographical
patterns that appear to support a countertrend of narrowing of crop diversity.
57
3.5
Conclusions
Identification of marginal agricultural land is essential for resource conservation and the successful
deployment and scaling of innovative technologies and development strategies. Our model pro-
vides a generalizable and multiscale heuristic for locating marginal agricultural land and disentan-
gling the factors driving marginality. Further, we propose that areas exhibiting marginal produc-
tivity are regions where attention is most needed and indicate the underlying drivers of marginality
to reveal where different intervention efforts would prove most effective. The results presented
here illuminate the role that spatiotemporal mapping can play in shifting contentious debates to
more informed, constructive engagement. Policymakers and scholars can use such mapping exer-
cises to elucidate impacts of agricultural development efforts, and plan effective ways to conserve
the environment while supporting farmer livelihoods and services for societal benefits. As pointed
out by Kremen (2015), moving beyond the either–or debate to a better understanding of vulner-
able lands is key to informing policies that protect natural areas while enhancing food production
in a conservation-sensitive manner.4
4 This research was supported by the Perennial Grain Crops for African Smallholder Farming Systems
project, Grant #OPP1076311, funded by the Bill & Melinda Gates Foundation, and through support of
the U.S. Agency for International Development, AID-OAA-A-13-00006. The opinions expressed herein
are those of the authors and do not necessarily reflect the views of the U.S. Agency for International
Development or the U.S. government. We thank the Lilongwe University of Agriculture and Natural Re-
sources (LUANAR) in Malawi for facilitating research in the field and M. N. Kakwera for his assistance
and expertise, as well as the EPA officers who offered their valuable insight. Additional thanks go to the
Environmental Science and Policy Program at Michigan State University for their continued support.
58
SCALING AGRICULTURAL INNOVATIONS: PIGEONPEA IN MALAWI5
4.1
Abstract
Successful scaling of agricultural development strategies is fundamental to increased production
and yields, yet targeting efforts frequently fail to fully consider the underlying biophysical drivers
of agricultural marginality, particularly at ï¬ne spatial resolutions. We present a heuristic for intel-
ligent targeting, utilizing remotely sensed information to identify the intersection between mar-
ginal conditions for performance of a staple crop and the optimal niche for technologies that
improve crop performance. Here, we explore the geographic potential of maize diversiï¬cation
with pigeonpea, a crop with soil productivity enhancing properties. Overall, 79 percent of agricul-
tural land in Malawi exhibits climate conditions optimal for pigeonpea cultivation and, in total,
approximately 51 percent of Malawian maize-based farming is expected to receive some beneï¬t
from pigeonpea integration, with 9 percent receiving predictable and substantial beneï¬ts. These
ï¬ndings illustrate the geographic scaling potential of pigeonpea in Malawi and provide direction
for informed pigeonpea deployment and market development across the country.
4.2
Introduction
In view of widespread food shortages and rural poverty across Africa, scaling innovations to boost
agricultural productivity is a global priority and essential to meeting many of the United Nations
Millennium Development Goals (Sachs 2005). Although deï¬nitions of scaling vary, achieving ef-
fective agricultural development involves innovations that have positive impacts on productivity.
Scaling in development is both vertical and horizontal. Vertical scaling includes institutionalization
5 This is an Accepted Manuscript of an article published by Taylor & Francis in The Professional Geographer
on 10 August 2017, available online at doi.org/10.1080/00330124.2017.1347798. Reference: Peter, B.G.,
Messina, J.P., Frake, A.N. and Snapp, S.S., 2018. Scaling agricultural innovations: pigeonpea in Malawi. The
Professional Geographer, 70(2), pp.239-250.
59
or decision making at higher levels and often involves sectors and stakeholder groups in the ex-
pansion process (Pachico and Fujisaka 2004). Horizontal scaling, also known as scaling out, refers
to geographical spread (Snapp and Heong 2003; Pachico and Fujisaka 2004). In practice, horizon-
tal scaling translates to the widespread spatial adoption of new behaviors or technologies through
expansion, replication, and adoption of projects, programs, or policies (Linn 2012). Adoption of
new technologies, both inputs and practices, along with the innovative use of existing technolo-
gies, is vital to achieving agricultural sector growth. These developments aim to increase land and
labor productivity, the effective use of natural resources, and farmer income potential, with market
solutions (USAID 2014).
Understanding the drivers that prompt farmers to adopt a new product, process, or prac-
tice is an essential part of designing a successful scaling strategy. Adoption is a dynamic process,
one that is location speciï¬c, and influenced by a wide range of biophysical and socio-economic
factors (Feder, Just, and Zilberman 1985). Not only does technology performance matter, but
there is also a need to consider social context, including input and output market opportunities,
farmer priorities, and perceptions of performance. This is illustrated by adoption and disadoption
of drought-tolerant maize varieties by smallholder farmers in Malawi, as nationally representative
survey data illustrate that this varies with location, as well as with farmer perceptions of drought
risk, and yields of modern maize varieties (Fisher and Snapp 2014). Overall, adoption of sustain-
able agriculture technologies in Africa has been shown to depend on biophysical performance,
which influences proï¬tability, and a complex milieu of land and labor availability, knowledge,
extension, technology availability, and policy (Hazell and Wood 2008; Muyanga and Jayne 2014).
Approaches to scaling agricultural productivity commonly address and sometimes mix in-
tensiï¬cation and extensiï¬cation. In intensiï¬ed systems, efforts are made to generate more product
60
on land currently under cultivation. Increasing production is achieved through changes in system
inputs (i.e., seed, fertilizer, land, labor, time, or feed; Food and Agriculture Organization of the
United Nations [FAO] 2004). Other land-use decisions include the displacement of one commod-
ity for another, where the economic outcome is comparable or greater (Govereh and Jayne 2003).
As consideration is given to the importance of agricultural systems in a global context, emphasis
is increasingly being given to sustainable intensiï¬cation practices. Thirty-eight percent of the
world’s agricultural area has become degraded through poor natural resource management, par-
ticularly in Africa, where up to 65 percent of agricultural land suffers from degradation (Feed the
Future 2015). With little room to expand agriculture into new areas due to poor suitability, access,
or limited availability, there is an increasing focus on intensiï¬ed agricultural productivity (USAID
2016).
Extensiï¬cation involves introducing agricultural production into land areas that have been
previously unused or used for less intensive purposes (Impact Assessment Group 2000). These
land areas are often marginal and might require substantial inputs to bring them into production.
High population density and limited land availability, however, is widespread in Malawi, limiting
extensiï¬cation options (Ricker-Gilbert and Jumbe 2014). Between 1990 and 2013, forest cover in
Malawi decreased by 23 percent (FAO 2016a), attributed in part to agricultural extensiï¬cation by
smallholder farmers and population increase (Government of Malawi 1998; Chibwana, Jumbe,
and Shively 2013). As of 2013, agriculture occupied approximately 61 percent of Malawi’s total
land area, with forest cover diminishing to 34 percent of the total land area (FAO 2016a). Conse-
quently, options for extensiï¬cation have been largely reduced to protected or marginal areas, nei-
ther of which are viable or sustainable for long-term production demand. Thus, intensiï¬cation
has become the primary focus for soil rehabilitation and increased production.
61
4.2.1 Prospect for Soil Rehabilitation in Malawi
Malawi is facing recurrent food security crises due in part to long-term unsustainable agricultural
practices (Glasson 2010; Mungai et al. 2016). There are other contributing factors, such as histor-
ical inequities in resource distribution and minimal investments in infrastructure, research, and
education. From a biophysical standpoint, the continuous production of maize over the last sev-
eral decades has fostered widespread soil nutrient and organic matter degradation (Sanchez 2002;
Ngwira, Aune, and Mkwinda 2012; Thierfelder et al. 2013). In the past, emphasis has been placed
on fertilizer substitution (e.g., Denning et al. 2009); however, the success and long-term viability
of such a solution is widely debated (Chinsinga and Poulton 2014; Messina et al. 2017), and seem-
ingly ineffective for the rural poor in Malawi (Holden and Lunduka 2013).
Pigeonpea has been proposed as a mitigation strategy for combating soil degradation with
its large root system, copious vegetative biomass, and superior ability to ï¬x nitrogen and enhance
phosphorus solubilization for soil rehabilitation (Snapp et al. 2010; Ngwira, Aune, and Mkwinda
2012; Mhango, Snapp, and Phiri 2013). The beneï¬ts of pigeonpea, grown in combination with
maize, or as a doubled-up legume system (i.e., pigeonpea and an understory of soybean or ground-
nut), rotated with maize, have been proven in country-wide trials (Snapp et al. 2010; Snapp et al.
2014). Perennializing agriculture through growing perennial legumes complementary to existing
maize-based farming systems is at the foundation of many such sustainable practices, yet adoption
across Africa has been limited (Schulz et al. 2003; Bezner-Kerr et al. 2007; Snapp et al. 2010). In
Malawi, uptake of legume biodiversity is commonly hindered by proï¬tability and farmer prefer-
ence; however, trials in northern Malawi show promise for semi-perennial legume adoption
(Snapp et al. 2010).
62
Pigeonpea is commonly integrated in maize systems in the southern region of Malawi, yet
farmers in the northern and central regions have historically elected not to grow pigeonpea, choos-
ing other legumes (e.g., groundnut, soybean, common bean, and cowpea) and other forms of crop
diversity (e.g., cassava; Malawi Vulnerability Assessment Committee [VAC] 2005; Simtowe et al.
2010). This regional delineation of crop choice does not appear to be dictated explicitly by bio-
physical conditions. In terms of temperature and precipitation, 79 percent of southeast Africa
(Kenya, Tanzania, Malawi, Mozambique) is suitable for pigeonpea cultivation (Snapp et al. 2019b).
Rather than climate, pigeonpea presence in Malawi might be dictated largely by social factors (e.g.,
market conditions, culture, extension, labor, preference, or pressure to produce cash crops; Snapp
and Silim 2002).
4.2.2 Horizontal Scaling of Pigeonpea in Malawi
Targeting is crucial to the successful deployment and scaling of development strategies and inno-
vation technologies (S. Wood et al. 1999). Although agricultural improvement efforts frequently
take place at the local level (e.g., Giller et al. 2006), scaling is typically targeted by region and absent
the local context (e.g., Millar and Connell 2010). More regional approaches to scaling agricultural
technologies are often unilateral (i.e., focus is given to a singular metric—e.g., sustainability), ne-
glecting to fully address the underlying combinations of biophysical and social factors driving
marginality, particularly at ï¬ne spatial resolutions. For intervention initiatives to succeed, they need
to be delivered in areas where beneï¬ts would be substantial and predictable (e.g., biophysical con-
ditions must be considered), and where farmer adoption is realistic (Kwesiga et al. 2003). Here,
we focus primarily on the biophysical suitability of pigeonpea across Malawi. This means promot-
ing pigeonpea and encouraging market development in regions and sites where (1) cli-mate con-
ditions are optimal for pigeonpea, and (2) maize suitability and maize production are marginal.
63
Remotely sensed data offer cost-effective, ï¬ne spatial and temporal resolution data to un-
cover the complex dynamics of agricultural production across the landscape and in remote places
on earth. These data allow for intelligent targeting of development strategies beyond administra-
tive levels and beyond often arbitrary decision-making processes unrelated to in situ conditions.
Other agricultural improvement initiatives have employed remotely sensed information for tar-
geting deployment of new technologies (e.g., Bellon et al. 2005; Muthoni et al. 2016). Bellon et al.
(2005) used remotely sensed climate data in combination with demographic information to target
agricultural advancements for poverty alleviation of farmers in Mexico. Muthoni et al. (2016) used
a suite of biophysical and socioeconomic variables to recommend zones for sustainable inten-
siï¬cation efforts in Tanzania. Our work complements studies such as these and is differentiated
by providing a framework to disentangle the complex combinations of the underlying biophysical
and possible social factors driving marginality, supplying information and recommendations at a
ï¬ne spatial resolution, and on a pixel basis. Highlighted here are areas and sites where pigeonpea
integration will provide predictable and substantial beneï¬ts in a classiï¬cation format readily usable
by policymakers, decision makers, and scholars.
We propose a heuristic for assessing the horizontal scaling potential of pigeonpea in Ma-
lawi using a comprehensive suite of remotely sensed measures of agricultural productivity, climate
conditions, and land suitability. First, we identify locations where climate conditions are optimal
for the cultivation of pigeonpea, based on fundamental niche. Second, we evaluate areas where
biophysical conditions for maize are suboptimal and productivity is historically marginal. Finally,
the intersection of soil-driven maize marginality and the optimal pigeonpea niche reveals “better
bet†locations for predictable and beneï¬cial integration outcomes. We present a range of scaling
64
outcomes and propose areas where pigeonpea will provide positive results based on biophysical
suitability, as well as areas that might beneï¬t from extension and market development.
4.3 Methods and Data
4.3.1 Pigeonpea Niche
To identify locations suitable for pigeonpea cultivation, we used two remotely sensed products:
(1) NASA MODIS Land Surface Temperature (LST— MOD11A2) for temperature (NASA LP
DAAC 2015b), and (2) NASA/JAXA Tropical Rainfall Measuring Mission (TRMM—3B43) for
precipitation (NASA/JAXA TRMM 2016). Thresholds for optimal pigeonpea conditions are
based on widely accepted and tested temperature and precipitation ranges (Table 4.1). We calcu-
lated the average value for each pixel across all years under study (2000–2014) for the November
through April growing seasons (Jayanthi et al. 2013; FAO 2016b). Only areas where temperature
and precipitation were both optimal were considered for the optimal niche.
4.3.2 Targeting Development Strategies
To identify areas where agricultural production is historically poor and soil is likely driving under-
production, we used a Malawi maize suitability map that disaggregates broad classes of suboptimal
biophysical conditions and agricultural productivity (Peter et al. 2018). The map depicts all com-
binations of suboptimal temperature, suboptimal precipitation, soil suitability, and agricultural
productivity. The dominant focus here is soil quality and pigeonpea integration. Legumes are gen-
erally tolerant to low soil fertility (Snapp and Silim 2002) and are commonly grown in marginal
soil environments (Kumar Rao and Dart 1987), whereas maize growth is impaired by marginal,
resource-poor soils (Heisey and Edmeades 1999). Pigeonpea can grow effectively on marginal
soils and its integration with maize cropping systems can provide soil nutrient enrichment, resili-
ence, and improved maize yields (Snapp et al. 2010). Because pigeonpea has these soil
65
rehabilitation properties, a reclassiï¬cation of the marginal maize map tailored to soil marginality
reveals areas where pigeonpea deployment would prove most effective. This reclassiï¬cation is one
that can be easily interpreted and readily used by policymakers and academics to intelligently target
research and development scaling initiatives.
Temperature (°C)
22.7–30.9
Crop
Pigeonpea (optimal)
Table 4.1. Optimal pigeonpea climate niche. Note: Temperature data acquired from NASA MODIS
LST (MOD11A2). Precipitation data acquired from NASA/JAXA TRMM (3B43). Precipitation
requirements are represented as accumulated growing season rainfall. L. Wood and Moriniere
(2013); Sardana, Sharma, and Sheoran (2010); Valenzuela and Smith (2002); Carberry et al. (2001);
Silim and Omanga (2001); Kimani (2001); Omanga, Summerfield, and Qi (1996); Houérou (n.d.).
Precipitation (mm)
544–1263
Figure 4.1. Reclassification diagram: Transforming maize marginality into pigeonpea opportunity.
Potential areas for pigeonpea deployment are deï¬ned as the intersection between soil-
driven marginal areas for maize and the optimal climate niche for pigeonpea. Locations where
marginality is related to soil (solely or in combination with marginal productivity) were reclassiï¬ed
to a “better bet†option for maximum beneï¬t from pigeonpea integration. In this case, because
soil is the only factor (of those under study) driving marginality, positive outcomes from pigeonpea
66
integration are expected. Locations with marginal production, but where low productivity is not
explained by any of the explanatory drivers, are areas where marginality might be driven by social
factors and would likely beneï¬t from extension and market development. It is also possible, how-
ever, that another factor not under study is driving marginality. Marginal productivity and subop-
timal temperature regions were also considered potential areas for extension and development.
Other reclassiï¬cations include locations that are suitable for pigeonpea, but beneï¬ts are
somewhat unpredictable because marginality is associated with other factors in combination with
soil and might require alternative solutions (e.g., climate-resilient crop varieties). Areas where pi-
geonpea is suitable, but soil is not a driver of marginality, are classiï¬ed as highly unpredictable.
The following are the resulting categories: (1) Better Bet—maximum beneï¬t from pigeonpea in-
tegration; (2) substantial beneï¬t from pigeonpea integration; (3) suitable for pigeonpea— likely
beneï¬ts from extension or market development; (4) suitable for pigeonpea—beneï¬ts somewhat
unpredictable; and (5) suitable for pigeonpea—beneï¬ts highly unpredictable (Figure 4.1). The last
resulting category (6) includes areas unsuitable for pigeonpea or nonmarginal for maize.
4.3.3 Field Survey and Data Acquisition
In 2015, we conducted a country-wide ï¬eld survey in which we spoke with extension ofï¬cers in
thirty-three farming regions (extension planning areas [EPAs]) across the extent of Malawi. We
visited nine EPAs in the north, eight in the south, and sixteen in the central region, conducting
interviews and land-cover/land-use assessments at 200 sites. At each site, we inquired about crops
grown and cropping system patterns (e.g., sole crop and intercrop). In addition to our interviews,
we acquired EPA-level crop production metrics from the Malawi Ministry of Agriculture and
Food Security (MoAFS), which covers the years between 2005 and 2012. These data, along with
the interviews, allow us to cross-reference reports of regional crop delineation (VAC 2005;
67
Simtowe et al. 2010), as well as reveal the spatial distribution of pigeonpea production and yield
at the local scale. Regional and administrative boundaries are presented here for reference (Figure
4.2).
Figure 4.2. Study area map highlighting the northern, central, and southern regions of Malawi and the contained
districts and extension planning areas.
4.4
Results
4.4.1 Spatial Distribution of Pigeonpea in Malawi
Here we present data on pigeonpea production and yield by administrative unit (EPA) across
Malawi for 2005 and 2012; these years represent the greatest temporal range in the data available.
68
There is a clear regional pattern of pigeonpea presence in the southern region of Malawi (Figure
4.3). In 2005, the southern region of Malawi produced 92 percent of the national pigeonpea total
(compared with less than 1 percent in the north). In 2012, the southern region of Malawi produced
89 percent of the national pigeonpea total (compared with less than 1 percent in the north). For
instances where pigeonpea is grown in the northern and central regions, however, yields are com-
parable to those observed in the south (Figure 4.4). In 2005, pigeonpea yields in the southern
region averaged approximately 430 kg ha-1, and in the northern region, pigeonpea yields averaged
approximately 440 kg ha-1. In 2012, pigeonpea yields in the southern region averaged approxi-
mately 980 kg ha-1, and in the northern region, pigeonpea yields averaged approximately 700 kg
ha-1. Pigeonpea is scarce in the central region; however, in 2005 pigeonpea yields averaged approx-
imately 470 kg ha-1 and in 2012 pigeonpea yields averaged approximately 900 kg ha-1. Coinciding
with government reports, our ï¬eld survey shows an overwhelming presence of pigeonpea in the
southern region of Malawi. Of our 200 sites visited (in thirty-three farming regions), we were able
to conï¬rm pigeonpea cultivation at forty sites. Of these sites, pigeonpea was grown at only three
sites in the north, four in the central region, and thirty-three in the south.
4.4.2 Pigeonpea Niche and Deployment Potential
Malawi’s climate is highly suitable for the cultivation of pigeonpea. In terms of both temperature
and precipitation, 79 percent of agricultural land exhibits climate conditions optimal for pigeonpea
growth (Figure 4.5). Approximately 3 percent is optimal only for temperature and 17 percent is
optimal only for precipitation, so only 2 percent of Malawi’s agricultural area is entirely suboptimal
for pigeonpea.
69
Figure 4.3. Map depicts the spatial distribution of pigeonpea production and yield in 2005 and 2012 by exten-
sion planning area.
Figure 4.4. Regional descriptive statistics: Box and whisker plot of pigeonpea production and yield in three major
regional subdivisions of Malawi (north, central, and south) for 2005 and 2012. Production in metric tons
and yield in kg/ha. Note that y-axes vary.
The intersection of marginal maize and optimal pigeonpea covers 74 percent of agricultural
land in Malawi (Figure 4.6). Within this intersection, 46 percent of maize marginality is attributed
70
to soil (solely or in combination with other drivers). Our ï¬ndings suggest maximum beneï¬t from
pigeonpea integration occurring on 2 percent of agricultural land. These are areas where soil suit-
ability is the sole driver of marginality. Seven percent of marginal agricultural land is expected to
receive substantial beneï¬t from pigeonpea integration. These are areas where soil suitability is a
primary driver of marginality, along with suboptimal temperature conditions. Another 7 percent
is suitable for pigeonpea and might beneï¬t from extension and market development. These are
areas where productivity is measurably low, yet there is either no observable limiting driver of
productivity (of those under study), or temperature is the sole factor. We found that 37 percent
of the country is suitable for pigeonpea, but soil is not the sole driver of marginality and beneï¬ts
are less predictable. Twenty-one percent is suitable for pigeonpea but soil is not a driver of mar-
ginal maize; therefore beneï¬ts are highly unpredictable. Twenty-six percent is unsuitable for pi-
geonpea or nonmarginal for maize. Overall, approximately 51 percent of Malawian agriculture is
likely to receive beneï¬ts from pigeonpea adoption, with varying degrees of predictability and ef-
fectiveness, at least 9 percent of which should receive highly predictable and beneï¬cial outcomes
from pigeonpea integration.
71
Figure 4.5. Pigeonpea climate niche: Optimal temperature and precipitation (2000–2014). Data sources:
NASA MODIS LST (MOD11A2) and TRMM (3B43). Inset maps selected to highlight local spa-
tial variability.
72
Figure 4.6. Potential areas for deployment of pigeonpea and market development. Inset maps selected to
highlight local spatial variability.
73
4.5 Discussion
4.5.1 Targeting Investments in Sustainable Agriculture Technologies
It is urgent to reverse the soil degradation trend in Malawi and support sustainable agricultural
practices and higher production levels. Agricultural development will not be successful unless
sustainable management of the underlying soil resource is addressed, nor will investments in agri-
cultural subsidies achieve proï¬table returns. To build soil organic matter requires sufï¬cient or-
ganic matter inputs, delivered efï¬ciently from leguminous plant roots (Puget and Drinkwater
2001; Kong and Six 2010). In Malawi, sustained effort to improve soil organic carbon is also
required. There is a growing body of evidence that the surest way to achieve soil organic matter
gains, in a manner that ensures ease of farmer adoption, is to promote crops that are shrubby,
such as pigeonpea, in contrast to growing exclusively annual crops that have limited life spans,
limited aboveground biomass, and meager root systems (Snapp et al. 2010; Glover, Reganold, and
Cox 2012).
Over the long term, technologies such as residue mulch systems, agroforestry, and inten-
siï¬ed livestock systems that transfer manure to crop ï¬elds are all expected to play a role in building
soil productivity in Malawi. Some of these systems, however, have been shown to require sub-
stantial labor investments, as well as food production opportunity costs, at least in the short term
(Sirrine et al. 2010). The combination of near-term options, such as diversiï¬cation with pigeonpea,
and more long-term, radical options all need to be considered as means to build soil organic mat-
ter. This is essential to ensure crop response and proï¬table use of investments in improved seed
and fertilizer. We recommend the Malawi government extension consider innovative approaches
to targeting intervention efforts, such as the example presented here that assesses where pigeonpea
can be most effectively deployed. To complement long-term investments, we propose immediate
74
research and extension attention be given to multipurpose legumes, such as pigeonpea, in the
“better bet†areas we have highlighted.
It is clear that pigeonpea is preferred in the southern region of Malawi, based on the quan-
tity of production; however, in instances where pigeonpea is grown in the northern and central
regions, it performs quite well and yields are comparable to those in the south (Figure 4.4 and
Figure 4.5). These ï¬ndings support the spatial extent of pigeonpea optimality and the potential
for widespread integration (Figure 4.3). Pigeonpea is largely absent in most of the central region;
however, the conï¬rmation of overlapping optimal niche and reported high productivity leads to
a reasonable assumption that pigeonpea could be grown successfully in the northern and central
region EPAs. In Figure 4.3, we show that much of the northern region is suitable for pigeonpea,
and, based on the drivers of marginality for maize, would receive maximum or substantial beneï¬t
from pigeonpea integration. These areas in the northern and central regions are high-priority areas
for the scaling and market development of pigeonpea.
4.5.2 Drivers of Pigeonpea Production
This article documents the spatial extent of soil and temperature properties that condition pigeon-
pea growth and compares this to the actual pigeonpea distribution. It is clear that pigeonpea is
underrepresented in terms of biophysical suitability, but is an expanding crop in Malawi. Produc-
tion area has increased at a rate of about 5.5 percent per year between 2006 and 2014; over the
same time period, production has increased at a rate of about 5.6 percent per year, and yield at a
rate of about 11.4 percent per year (FAO 2016a). The maps of pigeonpea production and yield
presented (MoAFS) are consistent with FAO-reported trends of continually increasing production
and productivity in Malawi. Indeed, 2005 was a particularly low producing year due to severe crop
response from poorly timed rains (Buerkle 2005); however, there is still a considerable annual yield
75
increase (8.1 percent on average, 2000–2014; FAO 2016a). The maps and reported ï¬gures also
demonstrate that pigeonpea responds similarly to rainfall stress across regions. Somewhat surpris-
ingly, this rapid upward production and yield trend holds true for many of Malawi’s primary crops
(e.g., maize, rice, groundnut, soybean, and tobacco). Despite reported production and yield gains
across crops, the fact remains that widespread food insecurity persists throughout rural Malawi
(Messina et al. 2017).
A driving factor of increasing pigeonpea production might be the growing international
market opportunities occasioned by large and consistent pigeonpea import demands from the
Indian subcontinent (Simtowe et al. 2010). Emergent pigeonpea success in neighboring Mozam-
bique and Tanzania (Walker et al. 2015) make it increasingly important for Malawi to situate itself
within the pigeonpea export arena. There is somewhat contradictory evidence regarding market
demand, which appears to be highly variable in Malawi, as several reports have highlighted farmer
concerns with pigeonpea market access limitations outside of southern Malawi (Rogé et al. 2016;
Waldman et al. 2017). The central region of Malawi, where the country capital of Lilongwe is
situated, is largely undiversiï¬ed, with maize and tobacco dominating the agricultural landscape
(VAC 2005), and pigeonpea is largely absent (Snapp et al. 2003). Conversely, southern districts
such as Zomba have a market structure that encourages pigeonpea production and integration
(Ortega et al. 2016; Waldman et al. 2016). In Malawi, variability of the pigeonpea market over
space and time might drive increased production in some areas and years, and limit it in others,
with an overall positive impact on pigeonpea production. Socioeconomic contexts (e.g., market
structure and farmer preference) are impeding pigeonpea scaling across much of Malawi. Another
barrier to pigeonpea production is the extent of free-ranging livestock in northern and central
Malawi. Compared to southern Malawi, there are few community norms that ensure year-round
76
livestock control, which is key to growth and survival of long-lived semi-perennial crops such as
pigeonpea (Rogé et al. 2016). Taken together, this is suggestive that agricultural development ef-
forts that involve pigeonpea promotion will need to pay attention to livestock control and market
policies, as well as biophysical suitability for pigeonpea.
4.6
Conclusions
Maize produced in optimal areas with recurring high yields will never be replaced by pigeonpea—
the calorie yield difference is simply too great. The strengths of the model proposed here, how-
ever, lie in its ability to transform marginality into opportunity with predictable outcomes. Small-
holder farmers are generally risk averse. Here, we provide evidence that pigeonpea is a promising
and scalable option for soil rehabilitation, nitrogen ï¬xation, and improved maize yields across
many speciï¬c locations in Malawi. The maps and methodology herein are resources that might be
used to intelligently target locations in need of intervention, where beneï¬ts will meet expectations
and minimize risk. The ï¬ne spatial and temporal resolution data we employ allow policymakers
and decision makers to focus efforts beyond the regional level, removing some uncertainty in
extension delivery and promotion of new practices. Although we focus on maize and pigeonpea
here, the model is not limited to a particular crop and could be generalized across geographies,
crops, scales, and innovation strategies. Reclassifying the underlying drivers of marginality, tailored
to speciï¬c technologies, will allow for a wide array of potential for agricultural innovations to scale
and improve agricultural systems across Malawi.6
6 This research was supported by the Perennial Grain Crops for African Smallholder Farming Systems
project, Grant #OPP1076311, funded by the Bill & Melinda Gates Foundation, and through support of
the U.S. Agency for International Development, AID-OAA-A-13-00006. The opinions expressed herein
are those of the authors and do not necessarily reflect the views of the U.S. Agency for International
Development or the U.S. government. We would like to thank the Lilongwe University of Agriculture and
Natural Resources and Africa RISING for facilitating research in the ï¬eld, as well as the Global Center
for Food Systems Innovation at Michigan State University.
77
MULTI-SPATIAL RESOLUTION SATELLITE AND sUAS IMAGERY FOR
PRECISION AGRICULTURE ON SMALLHOLDER FARMS IN MALAWI7
5.1
Abstract
A collection of spectral indices derived from a range of remote sensing imagery spatial resolutions
are compared here to on-farm measurements of crop yield and crop chlorophyll at two trial farms
in central Malawi. The primary objective of this analysis was to evaluate what spatial resolutions
are most effective for relating multispectral images with crop status. Single and multiple linear
regressions were tested for spatial resolutions ranging from 7-cm to 20-m using a small unmanned
aerial system (sUAS) and satellite imagery from Planet, SPOT 6, Pléiades, and Sentinel-2. Evidence
presented here suggests that imagery with spatial resolutions nearest the plant scale (i.e., 14–27-
cm) are most effective for relating spectral signals with crop status on smallholder farms in Malawi.
Moreover, green-band indices were more consistently correlated with crop chlorophyll and yield
than conventional red-band indices such as NDVI, and multivariable models often outperformed
single variable models.
5.2
Introduction
5.2.1 Precision Agriculture and sUAS
Precision agriculture has become a prominent subject of research for remote sensing of cropping
systems since the advent of individual-use small unmanned aircraft systems (sUAS) (Zhang and
Kovacs 2012). While national governments have been monitoring agricultural production via sat-
ellite since the 1970s, coarse spatial resolutions, cloud cover, and infrequent collection has ren-
dered much of these image data unfit for the needs of small-scale farmers (Mulla 2013). Now, the
7 This is an Unpublished Manuscript. Reference: Peter, B.G., Messina, J.P., Carrol, J., Lin, S., Chikowo, R.,
Zhi, J., and Snapp, S.S. 2019, In Preparation. Multi-spatial resolution satellite and sUAS imagery for preci-
sion agriculture on smallholder farms in Malawi.
78
use of quadcopters and fixed-wing drones for precision farming continues to grow as sUAS tech-
nologies become increasingly cost effective and programmatic operations are developed that
streamline the process of collecting imagery and generating map outputs (Floreano and Wood
2015). Many drones, such as the senseFly eBee fixed-wing aircraft with eMotion, no longer require
manual piloting and can be launched with pre-programmed flight plans. Moreover, associated
software such as Pix4D (Pix4D SA 2017) can handle many of the previously arduous image pre-
processing tasks, such as image mosaicking, geometric correction, orthorectification, and radio-
metric calibration. These advancements have made precision agriculture accessible to users with
minimal to no experience with aerial vehicle piloting and remote sensing-based crop analytics.
Continuing the precision autonomous farming trends, large-scale agribusinesses have begun to
leverage GPS enabled systems to deliver fertilizer and water relative to spatial need as measured
via sUAS (Eaton et al. 2008).
Typical agricultural-use sUAS multispectral cameras will acquire red, green, and blue
(RGB), red edge, and near-infrared (NIR) imagery (e.g., Parrot Sequoia multispectral camera) (Par-
rot Drones SA 2017), and more sophisticated cameras with a broader range of spectral bands can
record shortwave and thermal infrared (SWIR and TIR) (Stark et al. 2015), which are critical for
monitoring crop water stress (Ceccato et al. 2001). With the visible spectrum (RGB) and NIR, a
broad range of crop status indices can be calculated. Most notable is the normalized difference
vegetation index (NDVI) (Tucker 1979). This metric is widely used as a measure of crop health—
a healthy plant will absorb visible light (especially blue and red), while the fortified leaf structure
will reflect a high amount of NIR. NDVI calculation is simply (NIR–red)/(NIR+red), returning
values between -1 and 1, ranging from non-vegetated (water or barren) to healthy crops and plants
with a high leaf area index (LAI). Beyond being a metric for crop health, NDVI is frequently
79
correlated with crop production and yield, thus many farmers find value in mapping NDVI across
their fields (Wall et al. 2008).
5.2.2 Variable-Rate Technology and Spectral Signals of Farm Productivity
Perhaps the most common use of drone imagery on-farm is the application of variable-rate tech-
nology. This is the term coined to describe farm productivity improvement applications that are
informed by spatiotemporal measurements of crop performance (Sawyer 1994). In the case of
NDVI being used as an indicator of crop productivity, fertilizer may be applied proportionately
across a field depending on the spatial organization of measured crop status. Similarly, maximum
efficiency applications may be deployed that consider the nitrogen use efficiency (NUE) at various
stages of crop growth (i.e., fertilizer applied to a moderately healthy plant may have a greater return
than fertilizer applied to an unhealthy plant) (Cassman 2002). NDWI (normalized difference water
index), a ratio of the SWIR and NIR bands (Gao 1996), can be used to inform water distribution
across a field, supplying water to areas where crops are experiencing water stress and restricting
the amount delivered to areas at risk of waterlogging. Depending on the spatial resolution, an
image interpreter may be able to visually detect areas that are experiencing leaf damage from pests
and identify invasive plants and weeds between row crops (Hunt and Daughtry 2017). Further,
elevation point clouds derived from drone imagery can be used to identify sloping terrain prone
to erosion and areas susceptible to pooling, which can be managed by manipulating the field struc-
ture or by integrating crops with substantial root biomass (Pineux et al. 2017).
While NDVI is an extensively used vegetation index, it is well known that it does not
always emerge as having the strongest correlation with on-farm yields or leaf chlorophyll and is
often outperformed by other remote sensing indicators (Hatfield and Prueger 2010). Spectral in-
dices in relation to crop status can vary substantially due to differences in crop leaf structure,
80
planting density, soil and shadows, and image spatial resolution. Other indices that are often used
to correlate spectral signals with crop status include GNDVI (green normalized difference vege-
tation index), SAVI (soil adjusted vegetation index), GSAVI (green soil adjusted vegetation index),
GCI (green chlorophyll index), RVI (ratio vegetation index), CVI (chlorophyll vegetation index),
and MCARI1 (modified chlorophyll absorption ratio index 1). Zhu et al. (2012), for example,
found that the soil adjusted vegetation index (SAVI) was effective for estimating leaf nitrogen
content of wheat. Shanahan et al. (2001) found the green normalized difference vegetation index
(GNDVI) to be highly correlated with corn grain yields, particularly during midgrain filling. Sim-
ilarly, Panda et al. (2010) found the red- and NIR-based perpendicular vegetation index (PVI)
reliable for predicting corn yields. Mulla (2013) supplies a comprehensive list of remote sensing-
based indices used in precision agriculture contexts.
5.2.3 National Government and Commercial Satellites
While drones are an increasingly viable option for industrial farms and many small-scale opera-
tions, there is still a considerable financial and infrastructural barrier to entry in remote areas. As
such, national government and commercial satellites play an integral role in scaling agricultural
technologies and improving crop production worldwide. MODIS (250-m), Landsat (30-m), and
Sentinel-2 (20-m) are national government satellites that offer global coverage at regular temporal
intervals. Finer spatial resolution governmental and commercial satellites collect quasi-global im-
agery and can be commissioned for specific missions. Notably, SPOT 6/7 (7-m), Planet (3-m),
WorldView-2 (2-m), and Pléiades (2.5-m) satellite constellations offer regional, inter-farm, and
intra-farm multispectral imagery to measure crop status, though none at the individual maize plant
scale with multispectral bands. In contrast, sUAS can provide intra-farm measurements for indi-
vidual plants down to the leaf scale. For maize, a 30-cm spatial resolution image will capture a
81
complete plant and 7-cm is fine enough to evaluate individual leaves. Though MODIS, Landsat,
and Sentinel-2 are free to the public, finer spatial resolution images require purchase. A cost per
resolution curve of multispectral imagery from the Apollo Mapping repository (Apollo Mapping
2018) is plotted here for comparison in Figure 5.1.
Figure 5.1. Spatial resolution and cost per square kilometer curve. Costs listed are for the multispectral
products available from the Apollo Mapping archives. Prices were retrieved from the 2018 mono
high and medium-resolution satellite imagery price list (archive ≤ 90 days) (Apollo Mapping 2018).
5.3 Methods
The objective of this study was to evaluate the linear relationship between in situ farm measure-
ments (crop chlorophyll and yield) and spectral indices of crop status across multiple spatial res-
olutions. Spatial resolutions included in this analysis are 7-cm, 14-cm, and 27-cm (sUAS), 2.5-m
(Pléiades), 3-m (Planet), 7-m (SPOT 6), and 20-m (Sentinel-2). The workflow consisted of (i) sUAS
image collection and satellite image acquisition, (ii) image mosaicking, radiometric calibration, or-
thorectification, and geometric correction, (iii) index calculation (NDVI, GNDVI, SAVI, GSAVI,
GCI, RVI, CVI, and MCARI1), (iv) mean zonal statistic calculations by plot for each index, (v)
82
single and multiple linear regression permutations of all selected indices compared to crop chlo-
rophyll and crop yield, (vi) regression model rankings using Akaike information criterion (AICc),
and (vii) comparison of correlation results across all spatial resolutions under analysis (Figure 5.2).
The performance of each spectral index in relation to on-farm crop measurements was also eval-
uated.
Figure 5.2. Heuristic: multi-spatial resolution satellite and sUAS imagery for precision agriculture on small-
holder farms in Malawi.
The study took place on two trial farms in the central region district of Machinga in the
villages of Ntubwi and Nyambi (Figure 5.3). The farms are approximately 0.084 hectares (ha) each,
separated into 3 columns (referred to herein as replicates [R]) each containing 9 unique treatments
(T). There are 27 total plots per farm, with 6 rows of monocrop maize in each plot; surrounding
land cover was a complex mosaic of agriculture (e.g., groundnut, soybean, cassava, and finger
83
millet), tree cover, shrubs, barren land, and housing structures. The Ntubwi farm is approximately
0.092 ha (918 square meters) with an average plot size of 34.0 square meters and the Nyambi farm
is approximately 0.077 ha (769 square meters) with an average plot size of 28.5 square meters.
These trial farms are under experimentation by a research group at Michigan State University in
partnership with Africa RISING evaluating maize response to variable fertilizer treatments.
Figure 5.3. Study area map of the trial farms in Ntubwi and Nyambi, Malawi, with plots of differing manage-
ment strategies labeled.
5.3.1 Problems of Clouds and Satellite Revisit Rates
One substantial challenge associated with precision agriculture using remote sensing satellites is
the abundance of clouds during the crop production season. This is particularly problematic in
84
Malawi, where the unimodal rainy season introduces regular cloud cover. Persistent cloud cover,
coupled with infrequent satellite image revisit rates, often results in few clear images available for
farm observation. In the peak of the Malawi growing season, this may mean that only a handful
of commercial satellite images free of clouds will be available. In the case presented here, for two
farms in the central region of Malawi (Ntubwi and Nyambi) during peak growing season (Febru-
ary/March), only one clear image was available from Pléiades (January 28, 2018) and only one
clear image was available from SPOT 6 (March 25, 2018). Cloud-free Planet images were available
for both sites in close temporal proximity to our drone flights (February 17 at Nyambi and Feb-
ruary 20 at Ntubwi); drone flights were conducted February 19–20, 2018. The regular revisit rate
of national government satellites (e.g., Sentinel-2) minimizes this hurdle; however, the spatial res-
olution is insufficient for intra-farm precision agriculture in the Malawian smallholder context,
where a 0.5-hectare farm may contain only 9 complete Sentinel-2 20-m pixels. To demonstrate
and visualize the cloud-cover problem, we assembled NDVI calculations at the Nyambi trial farm
from 5 global monitoring satellites—AVHRR, VIIRS, MODIS, Landsat 8, and Sentinel-2 (Figure
5.4). Each open circle on the time-series represents an image collection and/or image delivery. In
the case of VIIRS and MODIS, composite images are delivered that ‘smooth’ out the effects of
cloud cover. In the Sentinel-2 time-series, for example, the impact of clouds on the spectral sig-
nature is made apparent through frequent drops in NDVI. A clear advantage of sUAS technolo-
gies here is that imaging occurs below the cloud layer.
85
Figure 5.4. Satellite image NDVI acquisition over the trial farm in Nyambi, Malawi. Hollow circles indi-
cate image collection or image delivery.
86
Every cloud free image over the Ntubwi and Nyambi trial farms that was available on the
Apollo Mapping web service (Apollo Mapping 2018) was purchased and processed for dates as
close to February 20th, 2018 as possible (the date of the sUAS image collection flights). Those
satellites included SPOT 6 (March 25, 2018 for Ntubwi) and Pléiades (January 28, 2018 for
Nyambi); however, the SPOT 6 image was collected after harvest had already taken place and was
excluded from analysis, but is presented here to further demonstrate unexpected timing issues
(Figure 5.5). Planet images were downloaded via the Planet API for February 17, 2018 (Nyambi)
and February 20, 2018 (Ntubwi) (Planet Team 2017). Sentinel-2 images were acquired using
Google Earth Engine (Gorelick et al. 2017) at both sites for February 20, 2018. Orthorectification
and geometric corrections of the SPOT and Pléiades imagery were performed using the recom-
mended parameters in the metadata from the imagery suppliers. The senseFly eBee drone flights
at Ntubwi and Nyambi were conducted at flight heights ranging from 79-m to 303-m for spatial
resolutions of 7-cm, 14-cm, and 27-cm. Image collection took place at peak overhead sun hours
(between 1000 and 1400) to minimize shadow. Flight plans were prepared using eMotion and
ground control points were set with a Topcon HIPER SR GPS base station, rover, and tripod.
The drone images were mosaicked, radiometrically calibrated, geometrically corrected, and or-
thorectified using Pix4D (Pix4D SA 2017). Figures 5.5 and 5.6 show the range of imagery resolu-
tions and collection dates for both sites using NDVI as an example.
87
Figure 5.5. NDVI of the trial farm in Ntubwi across spatial resolutions. Note that min (x1) and max (x2)
NDVI are unique to each map.
88
Figure 5.6. NDVI of the trial farm in Nyambi across spatial resolutions. Note that min (x1) and max (x2)
NDVI are unique to each map.
5.3.2 Multispectral Indices and On-Farm Measurements
Using spectral bands consistent across all imaging platforms under study (i.e., green, red, and
NIR), eight vegetation indices were selected to relate spectral signals with on-farm plant metrics
(Table 5.1). Spectral wavelengths of each product are listed below Table 5.1. The indices selected
included the normalized difference vegetation index (NDVI), green normalized difference vege-
tation index (GNDVI), soil adjusted vegetation index (SAVI), green soil adjusted vegetation index
89
(GSAVI), green chlorophyll index (GCI), ratio vegetation index (RVI), chlorophyll vegetation in-
dex (CVI), and modified chlorophyll absorption ratio index 1 (MCARI1) (Figure 5.7).
Index selection was informed using a priori knowledge of demonstrated relationships be-
tween crop status (e.g., crop chlorophyll and yield) and remote sensing spectral signals. The indices
selected are among the most commonly used in precision agriculture contexts, and consideration
was given to representation of both red- and green-based spectral indices, as well as two indices
(CVI and MCARI1) that use both the green and red bands in their calculation (Table 5.1). NDVI
is one of the most commonly used indices for evaluating healthy vegetation (Tucker 1979) and is
ubiquitous across precision agriculture applications; however, NDVI is sensitive to high leaf area
index (LAI). GNDVI was popularized by Gitelson and Merzlyak (1998) as a metric sensitive to
chlorophyll content in leaves and has been used similarly to NDVI as a decision support metric
in precision agriculture mapping tools (Zhang et al. 2010; Candiago et al. 2015); its calculation is
the same as NDVI, but using the green band in place of red. SAVI is similar in formulation to
NDVI, but uses an adjustment factor to minimize the effects of soil on the spectral signal (Huete
1988); this can be critical in areas where vegetation is sparse or where cropping arrangements (or
crop types) allow visibility of the soil beneath the canopy. GSAVI follows the SAVI formula
structure, but uses the green band in calculation instead of red; this index has shown to be effective
for determining early in-season nitrogen requirements for corn (Sripada 2005). CVI and MCARI1
are both indices that use NIR, red, and green in combination. CVI was designed to estimate leaf
chlorophyll at the canopy scale while minimizing the effects of LAI variation (Vincini et al. 2008).
Conversely, MCARI1 has an increased sensitivity to leaf area while also suppressing the sensitivity
to chlorophyll (Haboudane et al. 2004).
90
Abbreviation
NDVI
SAVI
GSAVI
1.5 × [(NIR – R)/(NIR + R + 0.5)]
GNDVI
(NIR – G)/(NIR + G)
Equation
(NIR – R)/(NIR + R)
Index
Normalized difference
vegetation index
Green normalized
difference vegetation
index
Soil adjusted vegetation
index
Green soil adjusted
vegetation index
Green chlorophyll index GCI
RVI
Ratio vegetation index
Chlorophyll vegetation
CVI
index
Modified chlorophyll
absorption ratio index 1
Table 5.1. Indices, abbreviations, and equations for each index under study. Parrot Sequoia multispectral
camera specifications: green 530–570 nm, red 640–680 nm, and NIR 770–810 nm. Planet surface
reflectance multispectral specifications: green 500–590 nm, red 590–670 nm, and NIR 780–860
nm. Pléiades-1A multispectral specifications: green 490–610 nm, red 600–720 nm, and NIR 750–
950 nm. Sentinel-2 MSI Level-1C multispectral specifications: green 560 nm, red 665 nm, and
NIR 835 nm. NIR = near infrared; R = red; G = green.
(NIR/G) – 1
(NIR /R
(NIR/G) × (R/G)
1.5 × [(NIR – G)/(NIR + G + 0.5)]
MCARI1
1.2 × [2.5 × (NIR – R) – 1.3 × (NIR – G)]
91
Figure 5.7. Nyambi multispectral index panel, 7-cm spatial resolution.
After computing index grids, mean zonal statistics were calculated for each plot to match
the scale of the on-farm measurements. Satellite images were resampled down to 50-cm so that
mean calculations of coarse resolution images were relative to the amount of each pixel contained
within each plot; no soil or shadow masking was performed. To link the imagery with in situ meas-
urements, we used a handheld SPAD (Soil-Plant Analyses Development) 502 Plus Chlorophyll
Meter, which returns an index of relative chlorophyll content (calculated using the 650 nm and
940 nm spectral bands) (Spectrum Technologies, Inc., 2019) and can be used to estimate leaf
92
nitrogen concentration in maize (Bullock and Anderson 1998). We collected the average SPAD
reading from the center two rows of each plot. The rationale for selecting plants near the center
of each plot was to minimize influence from neighboring plots that underwent different fertilizer
treatments. Yield measurements were also collected for each plot (Table 5.2).
N
T
U
B
W
I
N
Y
A
M
B
I
R1 SPAD
Yield
R2 SPAD
Yield
R3 SPAD
Yield
R1 SPAD
Yield
R2 SPAD
Yield
R3 SPAD
Yield
T1
33.2
1426
39.6
2033
42.3
911
24.6
467
25.9
467
25.5
488
T2
54.7
2133
51.1
3169
53.1
2113
36.8
3129
36.3
3105
42.7
6616
T3
55.7
3494
56.9
2600
53.3
3433
38.9
3869
39.9
3798
46.4
5422
T4
51.9
1848
53.6
3067
51.5
2275
43.8
4514
45.8
4347
46.6
6019
T5
56.6
2966
54.4
2336
54.6
3250
35.7
3415
46.7
6950
54.2
6067
T6
53.5
1808
54
2986
54.6
1727
41.6
3153
44.9
4228
53.9
5876
T7
56
3433
56
2377
56.5
4063
39
1696
49.5
9172
50.8
5780
T8
52
3128
51.6
2397
55.5
4834
49.9
6067
48.2
5876
58.2
9482
T9
54.1
3880
53
3006
50.9
2945
34.2
2723
43
2651
47
6473
Table 5.2. SPAD and yield measurements at the Ntubwi and Nyambi trial farms. T = treatment; R =
replicate.
5.3.3 Multiple Linear Regression Permutations
Hatfield and Prueger (2010) argue that using a collection of indices to quantify agricultural status
can be more effective than selecting a single index. To correlate the remote sensing indices with
crop status, we tested every possible multiple linear regression (MLR) permutation between the
remote sensing indices selected (independent variables) and the on-farm measurements (SPAD
and yield as dependent variables). A principal component analysis was also conducted, but no
significant groupings emerged due to the inherent collinearity among the variables under study.
In addition to multiple regression, tests were conducted using stepwise regression, partial least
squares regression, and a classification and regression tree (CART) approach. Ultimately, we
93
elected the multiple regression approach to maximize potential linear relationships while main-
taining model parsimony. The following is the global MLR model:
Ŷ = b0 + b1NDVI + b2GNDVI + b3SAVI + b4GSAVI + b5GCI + b6RVI + b7CVI + b8MCARI1,
where Ŷ is SPAD or yield. To generate these permutations, we used the MuMIn package in R
(Barton 2016), which produced 256 multiple linear regression equations per imagery spatial reso-
lution. We then ranked the models according to the Akaike Information Criterion (AICc), which
aims for a parsimonious balance of predictive power and model simplicity to adjust for model
overfitting or underfitting (Akaike 1974). Additionally, the Bonferroni Correction (Bonferroni
1935) was applied to the multiple regression permutation p-values to adjust for model overfitting.
The Bonferroni Correction method selected here involved simply multiplying each p-value by the
number of permutations (N = 256), which was done using p.adjust in R. The top performing
models from these tests are listed in Table 5.3.
5.4
Results and Discussion
The first test consisted of single linear regressions of all indices under study. One example of the
single variable linear regressions is presented here in Figure 5.8 to illustrate the relationships be-
tween proximal farm measurements and spectral signals derived from the sUAS images. This ex-
ample is from the senseFly eBee/Parrot Sequoia multispectral camera at a spatial resolution of 27-
cm at Nyambi. The index with the highest R2 in this particular case is GSAVI (R2 = 0.607), fol-
lowed by GNDVI (R2 = 0.554), GCI (R2 = 0.510), and SAVI (R2 = 0.489). Contrary to popular
use, NDVI does not rank among the top four, but ranks fifth with an R2 of 0.423. Interestingly,
the green band is in each of the top three index calculations. This evidence is consistent with other
studies that have found green-band indices critical for accurately predicting crop health (Gitelson
1998; Shanahan 2001; Sripada 2005). Overall, these R2 values fall below those produced with the
94
MLR models, echoing the argument from Hatfield and Prueger (2010) that a multivariable ap-
proach to predicting farm yields from remote sensing spectral indices is the optimal route. Given
these findings and the abundance of literature reiterating the value of a broader range of spectral
indices, precision agriculture and sUAS developers would benefit from diversifying their marketed
indices and integrate green-based indices alongside NDVI. The 27-cm spatial resolution GSAVI
image at Nyambi is mapped in Figure 5.9, scaled to the plot level, and compared to the on-farm
measurements of SPAD and yield.
Figure 5.8. Example single variable linear regressions—SPAD and spectral indices. Nyambi, senseFly eBee
Parrot Sequoia 27-cm spatial resolution.
Figure 5.9. GSAVI 27-cm spatial resolution grid at Nyambi, scaled to the plot level and compared to on-farm
measurements of SPAD and yield.
95
The second test consisted of all possible MLR permutations of the indices under study.
From the collection of multiple regression permutations, we filtered the top performing models
across all of the imaging platforms and compared spatial resolutions with their correlative strength
(Table 5.3). Each regression model presented has a p-value less than 0.05. Multiple linear regres-
sion models are only presented if the single variable model was outperformed. At Ntubwi, the
highest R2 (0.553) for SPAD (single variable linear regression) was at a spatial resolution of 14-cm
with GNDVI. The highest R2 (0.790) overall for SPAD was at a spatial resolution of 14-cm in a
multiple linear regression with CVI, GNDVI, NDVI, and RVI. Single variable regressions for
SPAD at a 7-cm spatial resolution at Ntubwi did not perform well, with the top performing model
returning an R2 of 0.196 using RVI. The highest R2 (0.349) for yield at Ntubwi was at a spatial
resolution of 14-cm using CVI; 0.27-cm spatial resolution ranked second (R2 = 0.302) with
GSAVI, and there was no statistically significant relationship between yield and spectral indices at
a 7-cm spatial resolution at this site (p > 0.05). Somewhat surprisingly, the top performing MLR
at 7-cm spatial resolution for SPAD at Ntubwi returned a high R2 of 0.750; however, this MLR
equation required 7 out of the 8 selected indices. At Ntubwi, neither Sentinel-2 nor Planet returned
significant relationships with SPAD or yield.
At Nyambi the highest R2 (0.607) for SPAD (single variable linear regression) was at a
spatial resolution of 27-cm with GSAVI. The highest R2 (0.720) overall for SPAD was at a spatial
resolution of 27-cm in a multiple linear regression with GNDVI, MCARI1, and RVI. Single vari-
able regressions for SPAD at a 7-cm spatial resolution at Nyambi also did not perform well, with
the top performing model returning an R2 of 0.283 using GNDVI. In contrast to Ntubwi, there
was no significant MLR model for SPAD at Nyambi at the 7-cm spatial resolution. At Nyambi,
Planet (3-m spatial resolution) GSAVI emerged as statistically significant (p < 0.05), but with an
96
R2 of only 0.165. At Nyambi, neither Sentinel-2 nor Pléiades returned statistically significant rela-
tionships with SPAD. At Nyambi, there were more statistically significant relationships between
spectral indices and yield than at Ntubwi. The highest R2 (0.637) for yield (single variable linear
regression) at Nyambi was at a spatial resolution of 27-cm using GSAVI. The top performing
model overall was at a 27-cm spatial resolution using MLR with CVI and MCARI1 (R2 = 0.677).
Planet and Sentinel-2 emerged with statistically significant relationships to yield at Nyambi (p <
0.05), but both with R2 < 0.2. The results of these tests are plotted in Figure 5.10.
Site Var.
Res. (m) R2 p-value Regression equation
0.07
0.14
0.27
0.196 0.021
0.553 8.7E-6
0.495 4.3E-5
N
T
U
B
W
I
N
Y
A
M
B
I
SPAD
0.07
Yield
SPAD
Yield
0.14
0.27
0.14
0.27
0.07
0.14
0.27
3
0.14
0.27
0.07
0.14
0.27
3
20
0.27
0.750 0.033*
S = 37.3 + 2.5(RVI)
S = -19.4 + 112.6(GNDVI)
S = -1.4 + 94(GNDVI)
S = 491.3 – 76.2(CVI) + 202.7(GCI) –
2314(GNDVI) + 1715.2(GSAVI) –
1010.6(MCARI1) + 939(NDVI) – 64.9(RVI)
0.790 8.7E-5* S = -252.3 – 48(CVI) + 1498.6(GNDVI) –
0.580 0.008*
0.349 0.001
0.302 0.003
0.283 0.004
0.476 6.9E-5
0.607 1.7E-6
0.165 0.035
0.546 0.020*
0.720 3.8E-4* S = -50.8 + 235.6(GNDVI) + 69.2(MCARI1) –
0.226 0.012
0.464 9.1E-5 Y = -4944.7 + 19526(NDVI)
0.637 6.1E-7 Y = -12112.7 + 45275.7(GSAVI)
Y = -17753.2 + 69123.8(SAVI)
0.157 0.041
Y = -2103.8 + 31495(MCARI1)
0.147 0.049
0.677 3.2E-4* Y = -18049.5 + 6263.3(CVI) +
622.8(NDVI) – 7.7(RVI)
S = -31.3 + 179.8(GNDVI) – 4.3(RVI)
Y = -4694.2 + 1820.5(CVI)
Y = -3599.7 + 16880.1(GSAVI)
S = 3.8 + 74.6(GNDVI)
S = 7.9 + 71.6(NDVI)
S = -16.2 + 160.1(GSAVI)
S = -61.1 + 288.5(GSAVI)
S = 31.3 – 128(GSAVI) + 176(MCARI1)
13(RVI)
Y = -5068.4 + 18407.3(GNDVI)
19556.8(MCARI1)
Table 5.3. Top performing single and multiple linear regression models as ranked by AICc. *Bonferroni
correction applied. S = SPAD; Y = yield.
97
The primary objective of this research was to quantify the effective spatial resolutions for
conducting precision agriculture on smallholder farms in Malawi. Based on the weak correlations
drawn from the satellite imagery, the abundance of cloud cover, and infrequent satellite revisit
rates, it is clear that sUAS are the optimal platforms for evaluating intra-farm variability. Figure
5.10 includes all single variable linear regressions (p < 0.05) and the top performing multiple linear
regressions. Illustrated in this figure is a substantial drop in the statistical relationship between
spectral index and on-farm measurements at the satellite scale. While the satellite spatial resolu-
tions ranged from 2.5-m to 20-m, Pléiades, Planet, and Sentinel-2 all fell below an R2 of 0.2. The
relationship at the sUAS scale varied substantially depending on the index and spatial resolution,
generally ranging between R2 values of 0.2 to 0.6, with some instances nearing zero, while MLR
models reached upwards of approximately 0.8. The second and fourth scatterplots show the re-
gression model R2 values at the drone scale (7–27-cm). This visualization demonstrates that the
relationship between crop status and spectral signals varies substantially even at fine scales.
Emergently clear is that the 14-cm and 27-cm spatial resolutions are more effective than
7-cm for relating spectral indices with proximal farm measurements. Interestingly, resampling
from 7-cm to 14-cm and 27-cm did not produce an increased correlation—there was minimal
distinguishable change. The tests here were performed on the imagery as collected; however, it
may be the case that soil and shadow extraction will improve the correlation at the 7-cm spatial
resolution. However, for a generalizable methodology, it may not be more effective than imaging
at a plant-scale spatial resolution. In this regard, one hypothesis is that a spatial resolution close to
the plant dimensions is the optimal spatial resolution for spectral imaging.
98
Figure 5.10. Comparing linear regression correlations of spectral indices and on-farm measurements across spatial
resolutions. (a) Nyambi top performing multiple linear regressions (depicted in purple), (b) Nyambi
max single variable linear regressions (depicted in blue), (c) Ntubwi top performing multiple linear
regressions (depicted in red), (d) Ntubwi max single variable linear regressions (depicted in green),
(e) Nyambi mean single variable linear regressions (depicted in orange), (f) Ntwubi max single
variable linear regressions, and (g) all single variable linear regressions (depicted in gray). Left
panel: SPAD; right panel: yield.
5.5
Conclusions
We evaluated a range of spatial resolutions of multispectral images over two trial farms in the
central region of Malawi—Nyambi and Ntubwi in the district of Machinga. Single and multiple
linear regressions were performed for all possible permutations of the indices selected. We found
that the correlation between on-farm measurements and remote sensing spectral signals decreases
substantially at 2.5-m, 3-m, and 20-m spatial resolutions. For precision agriculture, the problems
of clouds and satellite revisit rates are hurdles that have not yet been overcome with satellite im-
agery for the smallholder farming context. For larger fields, coarser spatial resolutions may be
adequate, but they were insufficient for the size of the field experimented in this study. Interest-
ingly, we found that the top performing yield models were associated with the 14-cm and 27-cm
spatial resolution imagery, rather than the 7-cm spatial resolution. We suspect that this is due to
soils and shadows in the fine resolution imagery, and possibly a fractal problem where more pixels
99
results in a greater number of pixels containing mixed soil and shadow. One hypothesis in this
regard is that initial image collection at the resolution of the plant dimensions might be the most
effective. In this case, 14–27-cm is closer to the maize plant size than 7-cm. We also found that
GNDVI and GSAVI were consistent indices for relating spectral images to proximal farm meas-
urements. These findings suggest that green-band indices are critical to test when monitoring crop
production via sUAS. Outcomes of this research are consistent with other studies that have found
that considering a broad range of indices may be more effective for evaluating plant status and
farm productivity than selecting a single index; however, the specific set of indices appropriate for
a given farm may vary.8
8 This paper is made possible by the support of the American People provided to the Feed the Future
Innovation Lab for Sustainable Intensification through the United States Agency for International Devel-
opment (USAID). The contents are the sole responsibility of the authors and do not necessarily reflect the
views of USAID or the United States Government. Program activities are funded by USAID under Co-
operative Agreement No. AID-OAA-L-14-00006. The authors wish to thank the American Society for
Photogrammetry and Remote Sensing and those affiliated with the Robert N. Colwell Memorial Fellow-
ship for their support. The authors also wish to thank Africa RISING for collaboration in on-farm research
activities, as well as the Global Change Learning Lab in Sub-Saharan Africa at Michigan State University
for their involvement. Additional gratitude is extended to Planet’s Education and Research Program for
supplying access to their imagery archive.
100
CONCLUSIONS
6.1 Overall Contributions
The scientific community is largely in agreement that global environmental changes are occurring
and that innovative crop system solutions are needed to increase food production and ensure a
sustainable future. To address this need, the research conducted in this dissertation paired geo-
spatial analytics with extensive interdisciplinary collaboration from plant and soil scientists. The
three primary scales under analysis here included the continental, country, and farm scales.
Nature-Based Agricultural Solutions: Scaling Perennial Grains Across Africa (Study 1)—At the con-
tinental scale, government satellite-based imaging platforms (e.g., NASA MODIS) were used to
evaluate agricultural productivity and elucidate the spatial organization of maize, pigeonpea, and
sorghum suitability across Africa. Study 1 introduced a methodology for characterizing levels of
crop suitability, showing unique agroecological zones that these crops occupy, and discussed the
need for up-to-date crop climate-suitability mapping. One key finding from this study was that
sorghum suitability occupies a unique agroecological niche in Africa. Due to its low rainfall re-
quirements, sorghum is biophysically suited to persist in the Sahel region, the Horn of Africa
(including portions of Kenya and Tanzania), northwestern Africa, and southern Africa, where
pigeonpea and maize conditions are suboptimal. Another important outcome is that the optimal
area for pigeonpea cultivation across Africa complements the niche for maize. Products from this
study included: (a) high spatial resolution maps showing the fundamental climate niche for maize,
pigeonpea, and sorghum across Africa at a 1-km spatial resolution; (b) high spatial resolution maps
of Africa showing the intersection of marginal maize and optimal pigeonpea or sorghum at a 1-
km spatial resolution.
101
A Multiscalar Approach to Mapping Marginal Agricultural Land: Smallholder Agriculture in Malawi
(Study 2)—At the country scale, national government satellites were used to evaluate agricultural
productivity and crop niche in Malawi to identify areas of marginal production and marginal maize
biophysical constraints. To enhance geographical communication and support policy decisions,
Study 2 contributed a heuristic for generating crop productivity map outputs that are relative to
government boundary and agroecological zone delineations. Further, this study offered a com-
mentary on reclaiming marginal lands for agricultural improvement by redesigning the ways these
areas are represented through pixel categorization and cartography. A key outcome of this research
was illuminating a clear spatial organization of the factors driving marginal maize production in
Malawi. For example, in the southern region of Malawi, rainfall variability is a primary factor lim-
iting agricultural production; in the central region of Malawi, temperature is a dominant factor
limiting agricultural production. Products from this study included: (a) agricultural land-use map
of Malawi that minimizes errors of commission; (b) heuristic for classifying agricultural produc-
tivity relative to spatial scales of observation; (c) map of marginal maize agricultural land in Malawi
depicting measured agricultural productivity and suboptimal biophysical conditions.
Scaling Agricultural Innovations: Pigeonpea in Malawi (Study 3)—To communicate areas of focus
for extension efforts regarding integration of pigeonpea into maize-based cropping systems, Study
3 identified and classified suitability with respect to soil as a primary driver of marginal production.
This study also demonstrated that pigeonpea has extensive scaling potential across Malawi and
that the crop is not limited geographically by biophysical constraints. Products from this study
included: (a) pigeonpea climate suitability map of Malawi; (b) maps comparing the spatial organi-
zation of pigeonpea production and yields for 2005 and 2012; (c) map of the intersection of
102
marginal maize agricultural land and optimal pigeonpea conditions, highlighting areas where pi-
geonpea integration would likely have beneficial outcomes.
Multi-Spatial Resolution Satellite and sUAS Imagery for Precision Agriculture on Smallholder Farms
in Malawi (Study 4)—At the farm scale, a small unmanned aerial system (sUAS) was used to quan-
tify the relationship between remotely-sensed spectral indices and proximal crop-leaf measure-
ments. In Study 4, a methodology was devised for testing the linear relationship between on-farm
measurements and spectral signals at multiple spatial resolutions. This study offers a framework
for evaluating combinations of spectral indices in relation to on-farm productivity. Of note was
that 14-cm and 27-cm spatial resolution images were more strongly correlated with proximal crop
measurements than the 7-cm spatial resolution. One hypothesis here is that initial imaging at the
plant scale is the most effective resolution because of how soil and crop shadows are integrated
in the spectral signal. Additionally, it was observed that green-band indices (i.e., GNDVI and
GSAVI) were consistently among the strongest correlative indices with both crop yield and crop
chlorophyll. Products from this study included: (a) chart depicting the cost/km2 for remote sens-
ing imagery, and a chart illustrating problems of satellite revisit rates and clouds for a smallholder
farm in Malawi; (b) maps comparing varying remote sensing imagery spatial resolutions at a small-
holder farm in Malawi, as well as maps of a collection of spectral indices; (c) multiple linear re-
gression models quantifying the correlation between remote sensing spectral signals across spatial
resolutions and proximal farm measurements of crop chlorophyll and crop yield.
103
6.2
Future Research
6.2.1 Web-Based GIS for Near Real-Time Crop Niche Mapping
Remote sensing-based characterizations of crop suitability are regularly harnessed to visualize the
spatial distribution of crop suitability and production potential. Conventionally, such mapping
efforts occur at singular time intervals, include variable collections of crops, and are often pre-
sented with ambiguous suitability classifications. Even the most widely used land and crop suita-
bility maps are subject to diminishing relevance over time, particularly in the face of global land
and climate change. At present, the most notable crop suitability maps are the Global Agro-Eco-
logical Zones (GAEZ) distributed by the Food and Agriculture Organization of the United Na-
tions (Fischer et al. 2012). While these products are immensely valuable, they are not produced at
regular or frequent intervals.
Acquisition and preprocessing of large-scale time-series satellite imagery is often time-con-
suming and an arduous management task, requiring expensive computational hardware, software
licenses, and storage structures capable of processing and moving large volumes of data. Recently,
however, cloud-based alternatives are offering increased efficiency for large-scale geospatial com-
puting. Web-based geographic information science (GIS) has the potential to transform the crop
niche mapping paradigm, bringing series of static maps to a dynamic form complete with interac-
tive geovisualizations and temporal continuity. Google Earth Engine (GEE) is one platform of-
fering planetary-scale geospatial data and analytical tools on the cloud (Gorelick et al. 2017). With
platforms such as GEE, crop suitability maps can and should be produced at the same rate climate
and land data are procured. Leveraging cloud-based technology can improve the accessibility of
global climate information for crop niche mapping, granting more geographic decision power to
104
agronomy academics, government stakeholders, and policymakers to effectively scale agricultural
improvement and ultimately illuminate what crops can grow where.
Interactive Mapping for Geographic Communication—A valuable byproduct of web-based GIS is
interactive mapping that can be queried for information at the pixel scale. Over the last several
decades, the dominant platform for academic geographers to communicate geospatial science has
been largely through journal publication. This has meant that a large proportion of map-based
geographic science has been delivered at the scale of a printable page. Even with fine spatial res-
olution imagery, scaling a figure down for print on standard letter-size paper will result in data loss
from resampling. With advancements in cloud-computing and server-side geoprocessing, data can
be retrieved and analyzed at unprecedented scales and speeds. An obvious advantage here is that
with web-based GIS delivery tools, stakeholders can pan, zoom, and query information as they
need. Given these considerations, all models developed in this dissertation will be replicated using
Google Earth Engine to offer temporally continuous and interactive geospatial crop information.
6.2.2 Crop Modeling and Return on Investment from Variable-Rate Technology
Remote sensing instruments and derivative models offer data that are synergistic with modern
cropping system models (e.g., temperature, precipitation, and solar radiation). With increasingly
fine-resolution data (both temporal and spatial), future precision agriculture may be conducted
across geographies at intra-farm scales of analysis, particularly in areas where such data have pre-
viously been unavailable. Global satellite coverage of agriculture exists in myriad forms as demon-
strated here; however, there are substantial barriers in retrieving climate and crop information at
local scales due to problems of clouds, imaging spatial resolutions, and satellite revisit rates (Mulla
2013). To this end, remote sensing scientists are pairing optical imagery with synthetic aperture
radar (SAR) (such as the sensors onboard Sentinel-1) to address problems of clouds and better
105
estimate agricultural productivity on the ground during rainy seasons (e.g., Campos-Taberner et
al. 2017). Downscaling global climate data to the farm scale is one avenue for achieving finer
spatial resolution climate information; however, there is some contention regarding the accuracy
of climate downscaling and large errors emerging due to crop model sensitivity (Atkinson 2013;
Lobell 2013b).
Despite present hurdles, the potential for precision agriculture to take place at the farm
scale using global satellite systems may become one of the frontiers of future agricultural improve-
ment science. In this case, variable-rate technology applications will benefit from the use of crop
modeling software to predict yield outcomes from soil amendments, particularly when local scale
climate information becomes available and crop system models such as APSIM (Agricultural Pro-
duction Systems sIMulator) integrate remote sensing-based metrics into their algorithms
(Holzworth 2014). This dissertation addressed multi-spatial and multi-temporal remote sensing
imagery, pixel and time-series agricultural productivity classifications, and variable rate technology.
Research evaluating crop yield predictions and the return on investment from variable-rate tech-
nology applications will prove valuable in the future (Figure 6.1).
106
Figure 6.1. Linking present and future research goals. Integrated remote sensing, variable-rate technol-
ogy, crop model yield predictions, and potential return on investment from intra-farm amend-
ments.
6.2.3 Crop Biodiversity, Nutrition, and Land Equivalence Ratios
Three of the United Nations Sustainable Development Goals are to increase agricultural produc-
tion, eliminate hunger, and end malnutrition (United Nations 2018). The potential for pigeonpea
to enhance maize productivity and rehabilitate soils on Malawi smallholder farms was discussed
at length in this dissertation, but also of importance is that introducing crop biodiversity supports
household nutrient diversity (Jones et al. 2014; Luckett et al. 2015). Considering intersecting areas
of maize and pigeonpea suitability, crop nutritional value (e.g., calories, protein, and vitamins), and
crop arrangements, it is possible to map the scaling potential of crop systems through the lens of
nutrition. Since yield is a primary metric of farm productivity, the land equivalent ratio (LER) can
be used to evaluate the tradeoffs between yield (and calories) and nutritional properties for differ-
ent spatial arrangements of maize and pigeonpea on a farm (Mead and Willey 1980; TerAvest
107
2019). Therefore, LER may be used as selection criteria for crop arrangement, then nutritional
indices calculated and mapped based on crop suitability. Comparing maps such as these with food
insecurity indicators and measures of agricultural productivity may provide substantial insight for
extension efforts working toward food security, nutrition, sustainable agriculture, and environ-
mental conservation.
6.2.4 Crop Differentiation and Further Evaluation of Spectral Indices Using sUAS
There are two studies planned to follow this dissertation that will further utilize imagery collected
via sUAS in Malawi. The first is a land-use classification analysis of the smallholder agriculture
environment. Since land-use is mixed at a local-scale, national government satellites (e.g., MODIS)
are largely insufficient for land classification at local scales where mixing is high and subtle thresh-
olds determine categorization (Li et al. 2017b). MODIS land-use/land-cover is 500-m spatial res-
olution and, as demonstrated in Study 4, this problem is not completely solved with finer spatial
resolution national government satellites; there were zero complete Sentinel-2 (20-m spatial reso-
lution) pixels within the trial farms. With broad categories of land-use classifications a challenge
(e.g., agriculture or shrubland), differentiating between crop types within agricultural land-use clas-
sifications is even more of a challenge. At 7–27-cm spatial resolution, distinguishing between
unique spectral signals from different crops is feasible (Park and Park 2015; Pajares 2015). Using
our field data collection on crop type, coupled with RGB image interpretation, both supervised
and unsupervised classification models will be tested. SVM (support vector machine) algorithms
and CART (classification and regression trees) will also be evaluated for classification accuracy.
Similarly to Study 4, a range of spatial resolutions will be tested (7-cm, 14-cm, and 27-cm) and a
comprehensive collection of spectral indices will be used to classify each pixel. The next study
planned will follow the heuristic described in Study 4, but will include a more exhaustive collection
108
of all spectral indices currently used in precision agriculture at one spatial resolution and with a
substantial sampling size for relating crop chlorophyll.
6.3
Closing Remarks
First, increasing agricultural production to meet population demand and ensure a sustainable fu-
ture is a multiscalar problem; agricultural innovation will not be successful unless the appropriate
scales of analysis are addressed. Decisions at the farm-scale cannot be made using a global-scale
lens; every farm within a 1-km spatial resolution pixel is simply not the same. Prescribing solutions
to an individual farm based on mixed-pixel data ignores the ecological fallacy problem. In some
cases, coarse spatial resolution imagery will suffice (e.g., precipitation or temperature), but an area
with multiple marginal farms surrounded by substantial tree cover could be classified as highly
productive and thus be completely missed by innovation targeting. Awareness of the modifiable
areal unit problem can help in this regard, normalizing by agroecological zone so that farms are
evaluated relative to similar environments.
Second, the future of crop suitability mapping should consider a framework that can de-
liver maps with greater frequency. As global changes continue to occur at a rapid pace, current
crop climate-suitability maps are needed as regularly as possible. With satellite climate data being
delivered to the public at increasing rates, crop climate-suitability models can be automated to
ingest these data and generate continuously updateable geospatial products.
Third, intra-farm precision agriculture in the smallholder context is in need of fine-resolu-
tion remote sensing imagery that can penetrate the cloud layer and provide data at high revisit
rates. Perhaps more beneficial, however, is continued sUAS development that makes unmanned
aerial vehicles more affordable and able to cover larger spatial extents. It must also be noted here
that the most effective spectral index for any given farm or crop may vary; it is suggested here that
109
crop chlorophyll and spectral index correlation tests be performed so that recommendations are
made using the most appropriate index. Furthermore, crop system yield models that integrate
remote sensing-based spectral indices as input data and more readily handle raster structure pro-
cessing will greatly advance intra-farm precision agriculture and crop yield mapping in the future.
In conclusion, this dissertation has shown that multiple scales of analysis are required for
integrating remote sensing with precision agriculture. Agricultural extension efforts using a geo-
graphic or remote sensing-based approach would benefit greatly from considering multiple scales
of analysis. Finally, it is critical that the resolution of the geographical conclusions drawn match
the resolution of the selected data, and that the scale of analysis matches the scale of the objective.
110
APPENDIX
111
Mixed-method Land Characterizations: Remote Sensing and Survey Data
While remote sensing data and geospatial models can be prepared entirely in a lab, unique insights
often emerge from qualitative data collection in the field. To this end, integrative quantitative-
qualitative models can offer a representation of reality that unifies geospatial data and human
experiences (Yeager and Steiger 2012; Cheong et al. 2012). Aside from information collected from
preplanned survey instruments, interfacing with people and pixels often unveils critical infor-
mation that may be absent in the literature or not readily apparent through data mining.
In July of 2015, a field survey of farming areas across Malawi was conducted. Thirty-three
extension planning areas (EPA) were visited and a total of 200 sites were surveyed. Interviews
were held with farmers and EPA officers to inquire about cropping systems and the primary fac-
tors limiting crop production. Crop presence and the biophysical/social factors limiting produc-
tion at each of the 200 survey sites are listed in Table A.1 (each survey site represents an area of
approximately one square kilometer). Soil quality (e.g., soil degradation, lack of fertilizer, or soil
type) was the dominant response (approximately 75 percent of respondents) for factors limiting
production; the frequency of this response was expected given the volume of literature addressing
this problem in Malawi. Throughout the interview process, there was an unexpectedly frequent
response that erratic rainfall patterns and changes in the start of the rainy season were dominant
factors, or one of the major factors, contributing to marginal production. Many EPA officers
reported that the total amount of rainfall each year was consistent; however, the seasonal patterns
and start date were changing and unpredictable. These experiences are consistent with published
research (Tadross et al. 2009; Vizy et al. 2015; Sutcliffe et al. 2016), but the research regarding this
topic in Malawi is scant. The model devised in Study 2 initially only considered rainfall amount in
characterizing areas of marginal precipitation. However, given the abundance of marginal rainfall
112
responses, we found it necessary to evaluate rainfall variability in addition to rainfall amount. Had
these interview data not been integrated, a major limiting factor of marginal agricultural produc-
tion could have been vastly underestimated or entirely missed. Though not a focus of this survey,
we also discovered a common association of high productivity farms and the presence of a legu-
minous tree known as Feidherbia albida, which may also require more future attention.
Longitude Latitude Crop presence
33.7858
-13.7283 maize, groundnuts, soya, beans,
-13.6828 maize, groundnuts, Irish potatoes,
-13.6648 maize, groundnuts, Irish potatoes,
tobacco
tomatoes
tomatoes
bacco
33.8504
33.8226
33.7951
33.62
33.5923
33.6017
33.6388
33.6758
33.5741
33.6668
33.6761
33.7036
33.9436
33.9436
33.9159
33.8698
34.1117
34.084
34.0841
34.0935
-13.7463 maize, groundnuts, banana; guava
and citrus across stream
-13.9909 maize, groundnuts soya, beans, to-
-14.0181 maize, groundnuts, soya, tobacco
-14.0543
-14.0723 maize, groundnuts, tobacco
-14.0722 maize, groundnuts, tobacco
tobacco
-14.1538 maize, groundnuts, soya
-14.1716 maize, groundnuts, beans, tobacco
-14.1807
-14.0721 maize, groundnuts
-13.8724 maize
tobacco
-13.8814 maize
-13.8815 maize, groundnuts
-13.945
-14.1971 maize, groundnuts, soya, beans
non-agriculture
-14.2063 maize, groundnuts, soya, beans
-14.2334 maize, groundnuts, soya, beans,
sweet/Irish potato
-14.2605 maize, groundnuts, soya, beans
Production limitations
soil degradation
soil degradation, temperature
(cold)
soil degradation
soil degradation
rainfall pattern, soil degrada-
tion
soil degradation
none
soil degradation
soil degradation, rainfall pat-
tern
rainfall pattern
soil degradation
soil degradation
none
soil degradation, monocrop-
ping, no fertilizer, manage-
ment
soil degradation, erosion
rainfall pattern
N/A
soil degradation, rainfall pat-
tern
soil degradation
rainfall pattern
rainfall pattern
Table A.1. Field survey data—crop systems and biophysical factors limiting production.
113
Table A.1 (cont’d)
-14.2336 maize, groundnuts, soya, beans,
cassava
-14.3842 maize, groundnuts, pigeonpea,
cowpea, cotton
-14.3752 maize, cowpea
-14.4114 maize, groundnuts, pigeonpea,
cowpea
non-agriculture
-14.3477
-14.3388 maize, tobacco, cotton
-14.8354 maize, groundnuts, soya, beans,
tobacco, pigeonpea, cowpea, sweet
potato, cabbage
-14.8261 maize, groundnuts, soya, tobacco,
sweet potato, millet
-14.7899 maize, groundnuts, soya, sweet
potato
-14.8174 maize, millet
-14.7726 maize, beans, sugar cane, cabbage,
lettuce, tomato
-14.7543 maize, groundnuts, soya, beans,
tobacco, millet, sorghum
-14.7537 maize, groundnuts, soya, tobacco,
cassava, sweet potato
-14.7089 maize, groundnuts, beans, to-
34.6619
-14.673
bacco, cowpea, bananas, papaya
maize, tobacco, sweet potato, ba-
nana, pumpkin
-14.6551 maize, groundnuts, soya, beans,
34.0378
34.5949
34.5948
34.5858
34.641
34.6316
34.691
34.7281
34.7371
34.6909
34.6348
34.6626
34.7368
34.6901
34.634
34.6524
34.5965
34.5779
34.6259
33.5164
soil degradation, erosion
rainfall pattern
rainfall pattern, soil degrada-
tion
rainfall pattern
N/A
rainfall pattern
rainfall pattern
soil degradation, rainfall pat-
tern, management
topography
management
temperature, erosion
management
soil degradation
soil degradation, erosion
management, soil degradation,
erosion
soil degradation, erosion
soil degradation
rainfall pattern
erosion
soil degradation, erosion, cold
temperature
soil degradation, rainfall pat-
tern, pests/diseases, tempera-
ture (cold/hot)
-14.6459 maize, soya, tobacco, millet, ba-
tobacco, millet
nana, pumpkin
-14.6101 maize, groundnuts, beans, to-
bacco, cowpea, millet, sweet po-
tato
-14.6103 maize, groundnuts, beans, to-
bacco, cowpea, millet, sweet po-
tato
-14.8178 maize, groundnuts, beans, cowpea,
Irish potato, millet
-13.1774 maize, groundnuts, soya, beans,
tobacco, sweet potato, cassava; to-
mato, mustard greens, onion
114
Table A.1 (cont’d)
33.498
33.4704
33.4427
33.4426
33.5708
33.5707
33.5523
33.5338
33.6168
33.6537
33.6813
33.5984
33.4607
33.4791
-13.1955 maize, groundnuts, soya, beans,
tobacco, sweet potato, cassava; ir-
rigation - tomato, mustard greens,
onion, bananas
-13.2317 maize, groundnuts, soya, beans,
tobacco, sweet potato
-13.2408 maize, groundnuts, soya, beans,
tobacco, sweet potato, sugar cane,
rice
-13.2047 maize, groundnuts, soya, beans,
33.452
-13.277
33.4335
-13.277
tobacco, sweet potato, sugar cane
maize, groundnuts, soya, beans,
tobacco, sweet potato, sugar cane,
Irish potato, mustard greens, rape,
banana
maize, groundnuts, soya, beans,
tobacco, tomato, banana
-12.7342 maize, groundnuts, soya, beans,
tobacco, sweet potato, cassava
maize, groundnuts, soya, beans,
tobacco, cowpea, sweet potato,
cassava
-12.698
-12.6981 maize, groundnuts, soya, beans,
tobacco, cowpea, sweet potato,
cassava
-12.707
-12.6891 maize, groundnuts, soya, beans,
tobacco, cowpea, sweet potato,
cassava
maize, groundnuts, soya, beans,
tobacco, cowpea, cassava
maize, groundnuts, soya, beans,
tobacco, cowpea, cassava
maize, groundnuts, soya, tobacco,
cowpea
-12.734
-12.734
-12.7432 maize, groundnuts, soya, beans,
tobacco, cowpea
-12.9786 maize, groundnuts, soya, beans,
cassava, sweet potato, cotton, pap-
rika
-12.9876 maize, groundnuts, soya, beans,
cassava, sweet potato, cotton, pap-
rika
115
soil degradation, rainfall pat-
tern, pests/diseases, tempera-
ture (cold/hot)
soil degradation
soil degradation, rainfall pat-
terns
soil degradation, rainfall pat-
tern, pests/disease
waterlogging, soil degradation
waterlogging, soil degradation
soil degradation, rainfall pat-
tern
rainfall pattern
rainfall pattern
rainfall pattern, soil degrada-
tion
soil type, rainfall pattern, soil
degradation
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
soil degradation, rainfall pat-
tern
rainfall pattern, soil degrada-
tion
-12.9605 maize, groundnuts, soya, beans,
-12.9605 maize, groundnuts, soya, beans,
tobacco
cassava
-12.8971 maize, groundnuts, soya, beans to-
bacco, cassava, sweet potato, cow-
pea
maize, groundnuts, soya, beans,
tobacco
-12.87
-13.0507 maize, bananas, other/unknown
-13.0326 maize, other/unknown
-15.4347 maize, groundnuts, pigeonpea,
sorghum, banana, mango, cassava
-15.4168 maize, pigeonpea, sorghum, soya,
sugar cane, cassava, banana
-15.4526 maize, pigeonpea, sorghum,
groundnuts, soya, mango, oranges,
tangerines, lemons, papaya, guava
-15.4599 maize, pigeonpea, rice, sorghum,
cotton
lablab
lablab
-15.4687 maize, pigeonpea, sorghum, rice,
-15.4778 maize, pigeonpea, sorghum, rice,
-15.4704 maize, pigeonpea, groundnuts,
cassava, lablab, mango, soya
-15.5104 maize, pigeonpea
Table A.1 (cont’d)
33.4422
33.4699
33.5066
33.4973
33.5807
33.5899
35.3758
35.357
35.3854
35.5438
35.5625
35.5626
35.4135
35.0319
35.0128
35.0035
34.9851
34.957
34.9659
35.013
-15.4834 maize, pigeonpea, cowpea
-15.4563 maize, pigeonpea, sorghum, sweet
potato, banana
-15.4655 maize, sorghum, pigeonpea, okra
-15.4837 maize, sugar cane, sweet potato,
rice, mango
-15.4659 maize, pigeonpea, groundnuts,
sorghum, cowpea, sweet potato
-15.4297 maize, sorghum, cotton
34.9753
-15.4296 maize, sorghum, cotton
116
soil degradation
soil degradation
rainfall pattern, soil degrada-
tion, erosion
rainfall pattern, soil degrada-
tion
none
none
soil degradation
soil fertility, rainfall pattern
soil fertility
none
waterlogging
soil fertility
soil fertility
soil degradation, rainfall pat-
tern, labor
soil fertility, rainfall pattern, la-
bor
soil fertility
soil degradation, rainfall pat-
tern
rainfall pattern
rainfall pattern, soil fertility
management, labor, rainfall
pattern
rainfall pattern, soil degrada-
tion, temperature
Table A.1 (cont’d)
35.3806
-15.8502 maize, soya, sorghum, groundnuts,
35.3896
-15.814
sunflower, pigeonpea
maize, soya, sorghum, groundnuts,
sunflower, pigeonpea
-15.7872 maize, pigeonpea, chickpea, sun-
flower, millet, sorghum
-15.7783 maize, pigeonpea, sorghum, sun-
-15.7426 maize, sorghum, tomato, pigeon-
-15.7604 maize, pigeonpea, eggplant,
-15.7693 maize, pigeonpea, sorghum, rice,
flower, eggplant
pea, sunflower
mango, pumpkin
tomato
-15.8633 maize, pigeonpea, pumpkin
34.961
-15.8905 maize bananas, sugar cane, beans
34.9425
34.9333
-15.9088 maize, beans, cassava, okra
-15.9179 maize, banana, cassava
34.952
-15.9268 maize, beans, rape
34.9507
-15.7912 maize, pigeonpea, mango
35.3613
35.3425
35.3048
35.333
35.3424
34.9794
34.9414
34.9321
34.9137
35.2891
35.2799
35.2051
35.1767
35.13
35.1487
35.2705
-16.014
-15.7913 maize, pigeonpea, cassava, sweet
-15.7914 maize, pigeonpea, cassava, sweet
-15.8187 maize, pigeonpea, cassava, man-
potato
potato
goes
-16.0048 maize, pigeonpea, cowpea, sor-
ghum, sweet potato, cassava
maize, pigeonpea, sugar cane, ba-
nana, mango
-16.0229 maize, pigeonpea, cassava, sor-
ghum, cowpea
-16.0147 maize, pigeonpea, banana, sor-
ghum, cowpea, cassava
-15.9788 maize, pigeonpea, cowpea, sor-
-15.9793
-15.9791
ghum, cabbage, banana
non-agriculture
non-agriculture
117
rainfall, soil degradation
rainfall, soil degradation
rainfall, soil degradation
rainfall, soil degradation
soil degradation, rainfall
rainfall, rainfall pattern, soil
degradation, management
rainfall pattern, rainfall
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil fertility
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, flooding, soil
degradation
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
rainfall pattern, soil degrada-
tion
land suitability, rainfall pat-
tern, soil degradation
N/A
N/A
Table A.1 (cont’d)
34.0482
-12.362
maize, cassava, rice
34.0667
-12.3801 maize, cassava, rice, sweet potato,
34.0852
-12.3981
banana
rice, sugar cane, cassava, banana,
some maize
-12.3981 maize, cassava, rice
-12.4342 maize, cassava, rice, beans, sweet
potato, tomato, banana
-12.4432 maize, cassava, rice, sweet potato
-12.4612 maize, bananas, some cassava
-12.4974
-12.5515
-12.5515
-12.5786
-12.5877
-13.7476
-13.7566 maize, tobacco, bananas
-13.6572 maize, groundnuts, tobacco
-13.6662 maize, tobacco, banana, mango,
non-agriculture
sugar cane
sugar cane
non-agriculture
non-agriculture
non-agriculture
groundnuts, soya, beans
-13.6662 maize, groundnuts, tobacco, soya,
beans
soya
-13.6933 maize, tobacco, papaya, mango,
-13.6933 maize, tobacco, sunflower,
groundnuts, beans, soya
-13.6843 maize, tobacco, sunflower,
groundnuts, soya
-13.7114 maize, tobacco, bananas, mango,
soya, beans, groundnuts, sun-
flower
-13.7205 maize, groundnuts, sunflower,
soya, tobacco, beans
-13.5125 maize, sweet potato, sugar cane
34.076
34.1037
34.0946
34.113
34.104
34.141
34.1502
34.1504
34.1504
33.0366
33.0366
33.0551
33.0459
33.0921
33.0829
33.1384
33.0274
32.9904
32.9719
32.9996
32.9996
33.0089
32.9719
-13.4854 maize, groundnuts, tobacco, man-
goes, soya
-13.4492 maize, groundnuts, tobacco,
mango, sunflower, soya
-13.3859 maize, cassava, soya
118
soil suitability, soil degrada-
tion, rainfall pattern
rainfall pattern
flooding, soil fertility
rainfall pattern, soil fertility
rainfall pattern, none
rainfall pattern
soil suitability, rainfall pattern
N/A
none
none
N/A
N/A
N/A
none
soil degradation, soil fertility
land suitability, soil degrada-
tion
soil degradation
soil degradation
land degradation
soil degradation, rainfall pat-
tern
soil degradation, rainfall pat-
tern
rainfall pattern
rainfall pattern, soil degrada-
tion
tree cover, rainfall, soil fertility
management
soil degradation, soil type
Table A.1 (cont’d)
32.9349
-13.612
maize, tobacco, sugar cane, beans,
soya, mango, banana, cassava,
sweet potato
-13.6391 maize, tobacco, groundnuts, soya
32.8979
35.4272
35.418
35.4088
35.4087
35.4091
35.3815
35.4003
35.4006
35.3663
35.3845
35.3751
34.2238
34.2056
34.1875
34.1326
34.1052
34.1875
34.0519
34.0337
34.0154
-14.1964 maize, groundnuts, beans, banana,
-14.2056 maize, groundnuts, beans, banana,
mango
mango
-14.2056 maize, groundnuts, beans, banana,
mango, cassava, pigeonpea
-14.1966 maize, groundnuts, beans, banana,
mango, cassava, pigeonpea
-14.2327 maize, cassava, banana, mango
-14.2511 maize, beans, mango
maize, banana, mango, beans
-14.278
-14.3051 maize, beans, banana, mango
-14.5675 maize, cotton, pigeonpea, tobacco,
banana
-14.5402 maize, pigeonpea, cotton
-14.5222 maize, rice, beans, shallots, to-
mato, banana, mango, potato
-10.3992 maize, cassava, rice, banana, sweet
potato, tomato, tobacco, mango,
sugar cane, groundnuts, vegetables
-10.4355 maize, cassava, groundnuts, sweet
potato, cotton, tobacco, mango,
banana
-10.4627 maize, cassava, groundnuts, sweet
potato, banana, mango, guava,
rice, tobacco, sugar cane, pigeon-
pea, soya, beans
-10.4448 maize, cassava, groundnuts
-10.4539 maize, cassava, groundnuts, pi-
geonpea, mango
rice, maize, banana, mango
-10.4808
-10.8882 maize, beans, mango, banana, cas-
sava, soya, tobacco
-10.9153 maize, beans, soya, mango, ba-
nana, cassava, tobacco
-10.9244 maize, beans, soya, mango, ba-
nana, cassava, tobacco
119
soil degradation, rainfall pat-
tern
soil degradation, rainfall pat-
tern
soil fertility, rainfall pattern
soil fertility, rainfall pattern
soil fertility
soil fertility
soil fertility
soil fertility
soil fertility
soil fertility
rainfall pattern, flood, soil fer-
tility
rainfall pattern, soil fertility
waterlogging
rainfall pattern
rainfall pattern, soil fertility
rainfall pattern, soil fertility
soil degradation
none
none
land suitability, soil degrada-
tion, rainfall pattern
washaway, land suitability, soil
degradation
soil degradation
Table A.1 (cont’d)
33.9971
33.988
33.9788
33.9606
33.9607
33.9515
33.9424
33.8973
33.8881
33.8972
33.9061
33.9519
33.9518
33.9242
33.4937
33.5028
33.4936
33.5302
33.5394
-10.9155 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts
-10.9245 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts
-10.9336 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts
-10.9608 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts, sweet potato
-10.9698 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts, sweet potato
soil fertility, rainfall pattern
soil fertility
soil fertility, washaway
soil fertility, soil degradation
soil fertility, soil degradation
soil degradation
-10.9699 maize, beans, soya, mango, ba-
nana, cassava, tobacco, ground-
nuts, sweet potato
maize, other/unknown
non-agriculture
-10.997
-11.1961
-11.1961 maize, banana, mango, vegetables management
-11.1599
none
rainfall patterns
sugar cane, banana, sweet potato,
vegetables
minimal, washaway
-11.0695 maize, banana, tomato, sugar cane,
vegetables
-11.0784 maize, soya, beans, tobacco
-11.0693
-11.0332 maize, soya, tobacco, beans, vege-
non-agriculture
tables, sugar cane, sweet potato
-10.8534 maize, groundnuts, cassava, sweet
potato, soya, beans, tobacco
-10.8262 maize, groundnuts, cassava, sweet
potato, soya, beans, tobacco
-10.8172 maize, groundnuts, cassava, sweet
potato, soya, beans, tobacco
-10.8171 maize, groundnuts, cassava, sweet
potato, soya, beans, tobacco, ba-
nana, mango
maize, groundnuts, cassava, sweet
potato, soya, beans, tobacco, ba-
nana, mango
-10.799
floods, rainfall pattern
rainfall pattern
N/A
floods, unimproved varieties
rainfall pattern
soil fertility
none, rainfall pattern
rainfall pattern
flood, washaway, waterlogging
120
Table A.1 (cont’d)
33.6227
33.6228
33.632
33.6595
33.6412
33.6412
33.3938
-11.2963 maize, soya, mango, sweet potato,
beans
-11.3234 maize, groundnuts, beans
-11.3324 maize, beans, groundnuts, soya
-11.3685 maize, beans, groundnuts, soya
-11.3776 maize, soya, groundnuts, cowpea,
beans, cassava, sweet potato
-11.3686 maize, soya, groundnuts, cowpea,
beans, cassava, sweet potato
-11.4323 maize, tobacco, groundnuts, soya
33.403
-11.4142 maize, tobacco, groundnuts, soya
33.3938
-11.369
maize, groundnuts, tobacco, soya
33.412
33.4307
33.4215
33.4031
33.4496
33.468
33.4863
33.468
33.4496
33.4404
33.468
33.4313
33.4221
33.4592
-11.3419 maize, groundnuts, tobacco, soya
-11.5318 maize, groundnuts
-11.5137 maize, tobacco, groundnuts
-11.4956 maize, groundnuts, soya, tobacco
-11.9206 maize, soya, groundnuts, cassava,
sweet potato, beans
-11.9206 maize, cassava, soya, sweet potato,
groundnuts, beans, cowpea
-11.9115 maize, cassava, sweet potato, soya,
groundnuts, beans, cowpea
-11.8934 maize, cassava, sweet potato, soya,
beans, banana, mango, cowpea
-11.9025 maize, cassava, sweet potato, soya,
beans, cowpea
beans, cowpea
-11.8935 maize, cassava, sweet potato, soya,
-11.9386 maize, cassava, sweet potato, soya,
beans, cabbage, tomato, cowpea
-11.9206 maize, groundnuts, soya, millet,
-11.9387 maize, groundnuts, soya, millet,
beans, cowpea
beans, cowpea
-12.1376 maize, beans, vegetables, banana,
guava, potato, sunflower, beans,
paprika, sugar cane
121
rainfall pattern
rainfall pattern
soil type (sandy)
soil fertility, soil type
soil fertility, rainfall
soil fertility
soil fertility, soil degradation,
management, income
soil fertility, soil degradation,
management, income
soil fertility, soil degradation,
management, income
rainfall patterns, soil fertility
rainfall pattern, soil fertility,
management
soil degradation
rainfall pattern, management,
soil degradation
soil degradation, soil type
soil fertility, soil type
soil type (sandy), soil fertility
soil fertility
soil type, soil fertility
soil fertility
rainfall amount
soil fertility, soil type
soil type, soil fertility
soil fertility, management (ac-
cess), rainfall amount
Table A.1 (cont’d)
33.3949
33.4132
33.4684
33.6622
33.6898
33.7082
33.6439
33.6663
-12.1467 maize, tobacco, cassava, soya,
groundnuts, beans
-12.1105 maize, soya, beans, paprika,
groundnuts, sugar cane, tomatoes,
tobacco
-12.1557 maize, bananas, mango, tobacco,
sugar cane, vegetables, soya,
groundnuts, sunflower, ground
bean, Irish potato, cassava, to-
mato, millet
-12.4718 maize, beans, soya, groundnuts,
cassava, millet, sugar cane, banana,
vegetables
-12.4717 maize, tobacco, beans, soya, sweet
potato, cassava, sunflower, Irish
potato, banana, sugar cane, vegeta-
ble
-12.4626 maize, beans, sweet potato, cas-
sava, tobacco, groundnuts, soya,
millet, banana, vegetables, sugar
cane
-12.4899 maize, tobacco, soya, beans, pi-
geonpea, cassava, sweet potato,
sugar cane, banana, Irish potato
-13.9818 maize, other/unknown
soil fertility
rainfall amount
soil fertility, rainfall patterns
rainfall pattern
rainfall pattern, soil degrada-
tion
rainfall pattern, land suitability
rainfall pattern, soil degrada-
tion
none
122
BIBLIOGRAPHY
123
BIBLIOGRAPHY
Akaike, H., 1974. A new look at the statistical model identification. In Selected Papers of Hirotugu
Akaike (pp. 215-222). Springer, New York, NY.
Apollo Mapping, 2018. Apollo Mapping: The Image Hunters. apollomapping.com/
Atkinson, P.M., 2013. Downscaling in remote sensing. International Journal of Applied Earth Observa-
tion and Geoinformation, 22, pp.106-114.
Aune, J.B., Coulibaly, A. and Giller, K.E., 2017. Precision farming for increased land and labour
productivity in semi-arid West Africa. A review. Agronomy for Sustainable Development, 37(3), p.16.
Banda, B., 2015. Malawi floods kill at least 48, damage crops. Reuters. http://www.reu-
ters.com/article/us-malawi-floodsidUSKBN0KN0NL20150114. Accessed 23 November 2016.
Barton, K., 2009. MuMIn: Multi-model inference. http://r-forge.r-project.org/projects/mumin/.
Bauduceau, N., Berry, P., Cecchi, C., Elmqvist, T., Fernandez, M., Hartig, T., Krull, W., Mayerho-
fer, E., Sandra, N., Noring, L. and Raskin-Delisle, K., 2015. Towards an EU Research and Inno-
vation Policy Agenda for Nature-based Solutions & Re-naturing Cities: Final Report of the Hori-
zon 2020 Expert Group on 'Nature-based Solutions and Re-naturing Cities'.
Bellon, M.R., Hodson, D., Bergvinson, D., Beck, D., Martinez-Romero, E. and Montoya, Y., 2005.
Targeting agricultural research to benefit poor farmers: Relating poverty mapping to maize envi-
ronments in Mexico. Food Policy, 30(5-6), pp.476-492.
Bezner-Kerr, R., Snapp, S., Chirwa, M., Shumba, L. and Msachi, R., 2007. Participatory research
on legume diversification with Malawian smallholder farmers for improved human nutrition and
soil fertility. Experimental Agriculture, 43(4), pp.437-453.
Bezner-Kerr, R., 2014. Lost and found crops: agrobiodiversity, indigenous knowledge, and a fem-
inist political ecology of sorghum and finger millet in northern Malawi. Annals of the Association of
American Geographers, 104(3), pp.577-593.
Blackie, M. and Dixon, J., 2016. Maize mixed farming systems: an engine for rural growth. In
Farming Systems and Food Security in Africa: Priorities for Science and Policy under Global Change. Routledge,
New York, NY.
Blaikie, P. and Brookfield, H., 1987. Land Degradation and Society. New York: Methuen.
Bonferroni, C.E., 1935. Il calcolo delle assicurazioni su gruppi di teste. Studi in onore del profes-
sore salvatore ortu carboni, pp.13-60.
124
Nations
(FAO)—FAONewsroom.
Borras, S.M., Franco, J.C., Carranza, D. and Alano, M.L., 2011. The fundamentally flawed ‘mar-
ginal lands’ narrative: insights from the Philippines. In International Conference on Global Land Grab-
bing (pp. 6-8).
Buchanan, J., 1885. The Shirè highlands (east Central Africa) as colony and mission. W. Blackwood and
Sons.
Buerkle, T., 2005. Malawi facing serious food crisis. Food and Agriculture Organization of the
United
http://www.fao.org/News-
room/en/news/2005/107298/index.html. Accessed 23 November 2016.
Bullock, D.G. and Anderson, D.S., 1998. Evaluation of the Minolta SPADâ€502 chlorophyll meter
for nitrogen management in corn. Journal of Plant Nutrition, 21(4), pp.741-755.
Byerlee, D. and Heisey, P.W., 1996. Past and potential impacts of maize research in sub-Saharan
Africa: a critical assessment. Food Policy, 21(3), pp.255-277.
Cai, X., Zhang, X. and Wang, D., 2010. Land availability for biofuel production. Environmental
Science & Technology, 45(1), pp.334-339.
Campos-Taberner, M., GarcÃa-Haro, F., Camps-Valls, G., Grau-Muedra, G., Nutini, F., Busetto,
L., Katsantonis, D., Stavrakoudis, D., Minakou, C., Gatti, L. and Barbieri, M., 2017. Exploitation
of SAR and optical Sentinel data to detect rice crop and estimate seasonal dynamics of leaf area
index. Remote Sensing, 9(3), p.248.
Candiago, S., Remondino, F., De Giglio, M., Dubbini, M. and Gattelli, M., 2015. Evaluating mul-
tispectral images and vegetation indices for precision farming applications from UAV images. Re-
mote Sensing, 7(4), pp.4026-4047.
Carberry, P.S., Ranganathan, R., Reddy, L.J., Chauhan, Y.S. and Robertson, M.J., 2001. Predicting
growth and development of pigeonpea: flowering response to photoperiod. Field Crops Re-
search, 69(2), pp.151-162.
Cassman, K.G., Dobermann, A. and Walters, D.T., 2002. Agroecosystems, nitrogen-use effi-
ciency, and nitrogen management. AMBIO: A Journal of the Human Environment, 31(2), pp.132-141.
Ceccato, P., Flasse, S., Tarantola, S., Jacquemoud, S. and Grégoire, J.M., 2001. Detecting vegeta-
tion leaf water content using reflectance in the optical domain. Remote Sensing of Environment, 77(1),
pp.22-33.
Challinor, A., Wheeler, T., Garforth, C., Craufurd, P. and Kassam, A., 2007. Assessing the vulner-
ability of food crop systems in Africa to climate change. Climatic Change, 83(3), pp.381-399.
Cheong, S.M., Brown, D.G., Kok, K. and Lopez-Carr, D., 2012. Mixed methods in land change
research: towards integration. Transactions of the Institute of British Geographers, 37(1), pp.8-12.
125
Chibwana, C., Jumbe, C.B. and Shively, G., 2013. Agricultural subsidies and forest clearing in
Malawi. Environmental Conservation, 40(1), pp.60-70.
Chinsinga, B. and O’Brien, A., 2008. Planting Ideas: How agricultural subsidies are working in Malawi.
Africa Research Institute, London.
Chinsinga, B. and Poulton, C., 2014. Beyond technocratic debates: The significance and transience
of political incentives in the Malawi Farm Input Subsidy Programme (FISP). Development Policy
Review, 32(s2), pp.123-s150.
Chipanshi, A.C., Chanda, R. and Totolo, O., 2003. Vulnerability assessment of the maize and
sorghum crops to climate change in Botswana. Climatic Change, 61(3), pp.339-360.
Cox, T.S., Glover, J.D., Van Tassel, D.L., Cox, C.M. and DeHaan, L.R., 2006. Prospects for de-
veloping perennial grain crops. BioScience, 56(8), pp.649-659.
Datt, B., 1999. Remote sensing of water content in Eucalyptus leaves. Australian Journal of
Botany, 47(6), pp.909-923.
DeHaan, L.R., Cox, T.S., Van Tassal, D.L., Glover, J.D., 2007. Perennial grains. In Farming with
nature: the science and practice of ecoagriculture (pp.61-82). Island Press, Washington DC.
Denning, G., Kabambe, P., Sanchez, P., Malik, A., Flor, R., Harawa, R., Nkhoma, P., Zamba, C.,
Banda, C., Magombo, C. and Keating, M., 2009. Input subsidies to improve smallholder maize
productivity in Malawi: Toward an African green revolution. PLoS Biology, 7(1), p.e1000023.
Devereux, S., 2002. State of disaster: Causes, consequences & policy lessons from Malawi. ActionAid Ma-
lawi, Lilongwe, Malawi.
Devereux, S., 2007. The impact of droughts and floods on food security and policy options to
alleviate negative effects. Agricultural Economics, 37, pp.47-58.
DeVisser, M.H. and Messina, J.P., 2013. Exploration of sensor comparability: a case study of
composite MODIS Aqua and Terra data. Remote Sensing Letters, 4(6), pp.599-608.
DeVries, J. and Toenniessen, G.H., 2001. Securing the harvest: biotechnology, breeding, and seed systems for
African crops. CABI.
Dicko, M.H., Gruppen, H., Traoré, A.S., Voragen, A.G. and Van Berkel, W.J., 2006. Sorghum
grain as human food in Africa: relevance of content of starch and amylase activities. African Journal
of Biotechnology, 5(5), pp.384-395.
Dionne, K. Y., Kramon, E., and Roberts, T., 2013. Aid effectiveness and allocation: Evidence
from Malawi. In Conference on Foreign Aid, Princeton University, Princeton, NJ.
Doggett, H., 1988. Sorghum. Harlow, Essex, England New York: Longman Scientific & Technical.
126
DuPlessis, J., 2008. Sorghum Production. Department of Agriculture, Pretoria, South Africa.
Duveiller, G., Lopez-Lozano, R. and Cescatti, A., 2015. Exploiting the multi-angularity of the
MODIS temporal signal to identify spatially homogeneous vegetation cover: A demonstration for
agricultural monitoring applications. Remote Sensing of Environment, 166, pp.61-77.
Eaton, R., Katupitiya, J., Siew, K.W. and Dang, K.S., 2008, April. Precision guidance of agricultural
tractors for autonomous farming. In 2008 2nd Annual IEEE Systems Conference (pp. 1-8). IEEE.
Ecker, O. and Qaim, M., 2011. Analyzing nutritional impacts of policies: an empirical study for
Malawi. World Development, 39(3), pp.412-428.
European Commission. Bauduceau, N., Berry, P., Cecchi, C., Elmqvist, T., Fernandez, M., Hartig,
T., Krull, W., Mayerhofer, E., Sandra, N., Noring, L. and Raskin-Delisle, K., 2015. Towards an
EU Research and Innovation Policy Agenda for Nature-based Solutions & Re-naturing Cities:
Final Report of the Horizon 2020 Expert Group on 'Nature-based Solutions and Re-naturing
Cities'.
European Space Agency (ESA), 2005. Global Land Cover (GLC) 2000. http://forobs.jrc.ec.eu-
ropa.eu/products/glc2000/products.php. Accessed 23 November 2016.
European Space Agency (ESA), 2008. GlobCover 2005–06. http://due.esrin.esa.int/page_glob-
cover.php. Accessed 23 November 2016.
European Space Agency (ESA), 2016. GlobCover 2009. http://due.esrin.esa.int/page_glob-
cover.php. Accessed 23 November 2016.
FAOSTAT, 2016. Food and Agriculture Organization of the United Nations (FAO) production,
area harvested, and yield of maize in Malawi. http://faostat.fao.org/. Accessed 23 November
2016.
Feder, G., Just, R.E. and Zilberman, D., 1985. Adoption of agricultural innovations in developing
countries: A survey. Economic Development and Cultural Change, 33(2), pp.255-298.
Feed the Future, 2015. Feed the Future Food Security Innovation Center: Sustainable intensifica-
tion. https://feedthefuture.gov/resource/feed-future-food-securityinnovation-center-sustaina-
ble-intensification. Accessed 18 July 2017.
Fischer, G., Nachtergaele, F., Prieler, S., Van Velthuizen, H.T., Verelst, L. and Wiberg, D., 2008.
Global agro-ecological zones assessment for agriculture (GAEZ 2008). IIASA, Laxenburg, Austria
and FAO, Rome, Italy, 10.
Fisher, M. and Snapp, S.S., 2014. Can adoption of modern maize help smallholder farmers manage
drought risk? Evidence from southern Malawi. Experimental Agriculture, 50, pp.533-48.
127
FAO
Grassland
Species
Information
change of Malawi
Food and Agriculture Organization of the United nations (FAO), n.d.-a Sorghum bicolor (L.)
Moench.
Profiles.
http://www.fao.org/ag/agp/agpc/doc/gbase/data/PF000319.HTM. Accessed 23 Nov. 2016.
Food and Agriculture Organization of the United Nations (FAO), n.d.-b Zea mays L. FAO grass-
land species profiles. http://www.fao.org/ag/agp/agpc/doc/gbase/data/PF000342.HTM. Ac-
cessed 23 November 2016.
Food and Agriculture Organization of the United Nations (FAO), 2004. The ethics of sustainable
agricultural intensification. ftp://ftp.fao.org/docrep/fao/007/j0902e/j0902e00.pdf. Accessed 3
November 2016.
Food and Agriculture Organization of the United Nations (FAO), 2013. Land cover and land
cover
(1990–2010). Rome: FAO. http://www.fao.org/geonet-
work/srv/en/main.home. Accessed 23 November 2016.
Food and Agriculture Organization of the United Nations (FAO), 2016a. FAOSTAT database:
Input. Land. http://faostat.fao.org/beta/en/#data/RL. Accessed 3 November 2016.
Food and Agriculture Organization of the United Nations (FAO), 2016b. GIEWS country
briefs—Malawi. Global
(GIEWS).
http://www.fao.org/giews/countrybrief/country.jsp?code=MWI. Accessed 23 November 2016.
Fritz, S., See, L., McCallum, I., You, L., Bun, A., Moltchanova, E., Duerauer, M., Albrecht, F.,
Schill, C., Perger, C. and Havlik, P., Mosnier, A., Thornton, P., Wood-Sichra, U., Herrero, M.,
Becker-Reshef, I., Justice, C., Hansen, M., Gong, P., Aziz, S. A., Cipriani, A., Cumani, R., Cecchi,
G., Conchedda, G., Ferreira, S., Gomez, A., Haffani, M., Kayitakire, F., Malanding, J., Mueller, R.,
Newby, T., Nonguierma, A., Olusegun, A., Ortner, S., Rajak, D.R., Rocha, J., Schepaschenko, D.,
Schepaschenko, M., Terekhov, A., Tiangwa, A., Vancutsem, C., Vintrou, E., Wenbin, W., van der
Velde, M., Dunwoody, A., Kraxner, F., Oberstenier, M., 2015. Mapping global cropland and field
size. Global Change Biology, 21(5), pp.1980-1992.
Frize, J., 2002. Food security situation analysis: Malawi. Oxford, UK: Oxfam GB.
Fujita, K., Ofosu-Budu, K.G. and Ogata, S., 1992. Biological nitrogen fixation in mixed legume-
cereal cropping systems. Plant and Soil, 141(1-2), pp.155-175.
Gao, B., 1996. NDWI—A normalized difference water index for remote sensing of vegetation
liquid water from space. Remote Sensing of Environment 58(3), pp.257-266.
Giller, K.E., Rowe, E.C., de Ridder, N. and van Keulen, H., 2006. Resource use dynamics and
interactions in the tropics: Scaling up in space and time. Agricultural Systems, 88(1), pp.8-27.
Gitelson, A.A. and Merzlyak, M.N., 1998. Remote sensing of chlorophyll concentration in higher
plant leaves. Advances in Space Research, 22(5), pp.689-692.
and Early Warning
System
128
Glasson, G.E., 2010. Developing a sustainable agricultural curriculum in Malawi: Reconciling a
colonial legacy with indigenous knowledge and practices. In Cultural Studies and Environmental-
ism (pp. 151-164). Springer, Dordrecht.
Glover, J.D., Reganold, J.P., Bell, L.W., Borevitz, J., Brummer, E.C., Buckler, E.S., Cox, C.M.,
Cox, T.S., Crews, T.E., Culman, S.W., DeHaan, L.R., Eriksson, D., Gill, B.S., Holland, J., Hu, F.,
Hulke, B.S., Ibrahim, A.M.H., Jackson, W., Jones, S.S., Murray, S.C., Paterson, A.H., Ploschuk,
E., Sacks, E.J., Snapp, S., Tao, D., Van Tassel, D.L., Wade, L.J., Wyse, D.L., Xu, Y., 2010. In-
creased Food and ecosystem security via perennial grains. Science 328(5986), pp.1638-1639.
Glover, J.D., Reganold, J.P. and Cox, C.M., 2012. Agriculture: Plant perennials to save Africa's
soils. Nature, 489(7416), p.359.
Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J.,
Robinson, S., Thomas, S.M. and Toulmin, C., 2010. Food security: the challenge of feeding 9
billion people. Science, 327(5967), pp.812-818.
Gopalakrishnan, G., Cristina Negri, M. and Snyder, S.W., 2011. A novel framework to classify
marginal land for sustainable biomass feedstock production. Journal of Environmental Quality, 40(5),
pp.1593-1600.
Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D. and Moore, R., 2017. Google
Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202,
pp.18-27.
Govereh, J. and Jayne, T.S., 2003. Cash cropping and food crop productivity: synergies or tradeâ€
offs? Agricultural Economics, 28(1), pp.39-50.
Government of Malawi, 1998. State of the environment report for Malawi. Lilongwe, Malawi:
Environmental Affairs Department.
Haack, B., Mahabir, R. and Kerkering, J., 2015. Remote sensing-derived national land cover land
use maps: a comparison for Malawi. Geocarto International, 30(3), pp.270-292.
Haboudane, D., Miller, J.R., Pattey, E., Zarco-Tejada, P.J. and Strachan, I.B., 2004. Hyperspectral
vegetation indices and novel algorithms for predicting green LAI of crop canopies: Modeling and
validation in the context of precision agriculture. Remote Sensing of Environment, 90(3), pp.337-352.
Harlan, J.R. and Stemler, A., 1976. The races of sorghum in Africa. In Origins of African Plant Do-
mestication (pp.465-478). De Gruyter.
Hatfield, J.L. and Prueger, J.H., 2010. Value of using different vegetative indices to quantify agri-
cultural crop characteristics at different growth stages under varying management practices. Remote
Sensing, 2(2), pp.562-578.
129
Haussmann, B.I., Fred Rattunde, H., Weltzienâ€Rattunde, E., Traoré, P.S.C., Vom Brocke, K. and
Parzies, H.K., 2012. Breeding strategies for adaptation of pearl millet and sorghum to climate
variability and change in West Africa. Journal of Agronomy and Crop Science, 198(5), pp.327-339.
Haworth, B., Biggs, E., Duncan, J., Wales, N., Boruff, B. and Bruce, E., 2018. Geographic infor-
mation and communication technologies for supporting smallholder agriculture and climate resil-
ience. Climate, 6(4), p.97.
Hazell, P. and Wood, S., 2007. Drivers of change in global agriculture. Philosophical Transactions of
the Royal Society B: Biological Sciences, 363(1491), pp.495-515.
Heinsch, F.A., Reeves, M., Votava, P., Kang, S., Milesi, C., Zhao, M., Glassy, J., Jolly, W.M., Loeh-
man, R., Bowker, C.F. and Kimball, J.S., 2003. GPP and NPP (MOD17A2/A3) products NASA
MODIS land algorithm. MOD17 User's Guide, pp.1-57.
Heisey, P.W. and Edmeades, G.O., 1999. CIMMYT 1997/98 world maize facts and trends; maize pro-
duction in drought-stressed environments: technical options and research resource allocation (No. 557-2016-
38824). CIMMYT.
Hellin, J., Shiferaw, B., Cairns, J.E., Matthew Reynolds, M.P., Ortiz-Monasterio, I., Banziger, M.,
Sonder, K. and La Rovere, R., 2012. Climate change and food security in the developing world:
Potential of maize and wheat research to expand options for adaptation and mitigation. Journal of
Development and Agricultural Economics, 4(12), pp.311-321.
Holden, S. and Lunduka, R., 2010. Too Poor to be Efficient? Impacts of the Targeted Fertilizer Subsidy
Programme in Malawi on Farm Pilot Level Input Use, Crop Choice and Land Productivity. Norwegian Uni-
versity of Life Sciences (UMB).
Holden, S.T. and Lunduka, R.W., 2013, March. Who benefit from Malawi's targeted farm input
subsidy program? In Forum for Development Studies (Vol. 40, No. 1, pp. 1-25). Routledge.
Holzworth, D.P., Huth, N.I., deVoil, P.G., Zurcher, E.J., Herrmann, N.I., McLean, G., Chenu,
K., van Oosterom, E.J., Snow, V., Murphy, C. and Moore, A.D., 2014. APSIM–evolution towards
a new generation of agricultural systems simulation. Environmental Modelling & Software, 62, pp.327-
350.
Houérou, L., n.d. Cajanus cajan (L.) Millsp. Grassland Species Profiles. Food and Agriculture Or-
ganization
(FAO).
http://www.fao.org/ag/agp/agpc/doc/gbase/data/PF000319.HTM. Accessed 7 Nov. 2016.
House, W. and Zimalirana, G., 1992. Rapid population growth and its implication for Malawi. Ma-
lawi Medical Journal, 8(2), pp.46-54.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment, 25(3),
pp.295-309.
Nations
of
the
United
130
Hunt Jr, E.R. and Daughtry, C.S., 2018. What good are unmanned aircraft systems for agricultural
remote sensing and precision agriculture? International Journal of Remote Sensing, 39(15-16), pp.5345-
5376.
IIASA/FAO. Fischer, G., Nachtergaele, F., Prieler, S., Van Velthuizen, H.T., Verelst, L. and Wi-
berg, D., 2008. Global agro-ecological zones assessment for agriculture (GAEZ 2008). IIASA,
Laxenburg, Austria and FAO, Rome, Italy, 10.
Impact Assessment Group, 2000. Impact methods to predict and assess contributions of technol-
ogy. http://cnrit.tamu.edu/IMPACT/. Accessed 3 November 2016.
Infonet Biovision,
http://www.infonet-bio-
Infonet Bio
vision.
2016. Maize.
vision.org/PlantHealth/Crops/Maize〉. Accessed 23 November 2016.
International Crops Research Institute for the Semi-Arid Tropics (ICRISAT), No Date (n.d.). Pi-
geonpea. http://exploreit.icrisat.org/page/pigeonpea/687. Accessed 5 September 2016.
International Food Policy and Research Institute (IFPRI), 2002. World land use–land cover.
Rome: Food and Agriculture Organization of the United Nations. http://www.fao.org/geonet-
work/srv/en/main.home. Accessed 23 November 2016.
International Fund for Agricultural Development (IFAD), 2001. Rural poverty report 2001. Rome:
IFAD.
Jackson, W., 2002. Natural systems agriculture: a truly radical alternative. Agriculture, Ecosystems &
Environment, 88(2), pp.111-117.
James, L., 2010. Theory and identification of marginal land and factors determining land use change (No. 1097-
2016-88829). Master’s thesis, Michigan State University, East Lansing, MI.
Jayanthi, H., Husak, G.J., Funk, C., Magadzire, T., Chavula, A. and Verdin, J.P., 2013. Modeling
rain-fed maize vulnerability to droughts using the standardized precipitation index from satellite
estimated rainfall—Southern Malawi case study. International Journal of Disaster Risk Reduction, 4,
pp.71-81.
Jenks, G.F., 1967. The data model concept in statistical mapping. International Yearbook of Cartog-
raphy, 7, pp.186-190.
Jones, A.D., Shrinivas, A. and Bezner-Kerr, R., 2014. Farm production diversity is associated with
greater household dietary diversity in Malawi: findings from nationally representative data. Food
Policy, 46, pp.1-12.
Josephson, A.L., Ricker-Gilbert, J. and Florax, R.J., 2014. How does population density influence
agricultural intensification and productivity? Evidence from Ethiopia. Food Policy, 48, pp.142-152.
131
Kalipeni, E., 1996. Demographic response to environmental pressure in Malawi. Population and
Environment, 17(4), pp.285-308.
Kandaya, H., 2016. Emergency appeal operation update Malawi: Food insecurity. Geneva: International
Federation of Red Cross and Red Crescent Societies.
Kane, D.A., Rogé, P. and Snapp, S.S., 2016. A systematic review of perennial staple crops literature
using topic modeling and bibliometric analysis. PloS One, 11(5), p.e0155788.
Kang, S., Post, W., Wang, D., Nichols, J., Bandaru, V. and West, T., 2013. Hierarchical marginal
land assessment for land use planning. Land Use Policy, 30(1), pp.106-113.
Kassie, G.T., Erenstein, O., Mwangi, W.M., La Rovere, P.S., Setimela, P.S. and Langyintuo, A.S.,
2012. Characterization of maize production in southern Africa: Synthesis of CIMMYT/DTMA
household level farming system surveys in Angola, Malawi, Mozambique, Zambia and Zimbabwe.
Working Paper 4, International Maize and Wheat Improvement Center, El Batán, Mexico.
Khumalo, S., Chirwa, P.W., Moyo, B.H. and Syampungani, S., 2012. The status of agrobiodiversity
management and conservation in major agroecosystems of Southern Africa. Agriculture, Ecosystems
& Environment, 157, pp.17-23.
Kimani, P.M., 2001. Pigeonpea breeding: Objectives, experiences, and strategies for Eastern Af-
rica. In Status and potential of pigeonpea in Eastern and Southern Africa: proceedings of a regional workshop.
Nairobi, Kenya. B-5030 Gembloux, Belgium: Gembloux Agricultural University; and Patancheru
502 324, Andhra Pradesh, India: International Crops Research Institute for the Semi-Arid Tropics
(ICRISAT).
Kong, A.Y. and Six, J., 2010. Tracing root vs. residue carbon into soils from conventional and
alternative cropping systems. Soil Science Society of America Journal, 74(4), pp.1201-1210.
Kremen, C., 2015. Reframing the landâ€sparing/landâ€sharing debate for biodiversity conserva-
tion. Annals of the New York Academy of Sciences, 1355(1), pp.52-76.
Kumar Rao, J.V. and Dart, P.J., 1987. Nodulation, nitrogen fixation and nitrogen uptake in pi-
geonpea (Cajanus cajan (L.) Millsp) of different maturity groups. Plant and Soil, 99(2), pp.255-266.
Kwesiga, F., Akinnifesi, F.K., Mafongoya, P.L., McDermott, M.H. and Agumya, A., 2003. Agro-
forestry research and development in southern Africa during the 1990s: review and challenges
ahead. Agroforestry Systems, 59(3), pp.173-186.
Lantican, M.A., Pingali, P.L. and Rajaram, S., 2003. Is research on marginal lands catching up?
The case of unfavourable wheat growing environments⋆. Agricultural Economics, 29(3), pp.353-361.
132
Li, F., Miao, Y., Feng, G., Yuan, F., Yue, S., Gao, X., Liu, Y., Liu, B., Ustin, S.L. and Chen, X.,
2014. Improving estimation of summer maize nitrogen status with red edge-based spectral vege-
tation indices. Field Crops Research, 157, pp.111-123.
Li, G., Messina, J.P., Peter, B.G. and Snapp, S.S., 2017a. Mapping land suitability for agriculture
in Malawi. Land Degradation & Development, 28(7), pp.2001-2016.
Li, X., Messina, J.P., Moore, N.J., Fan, P. and Shortridge, A.M., 2017b. MODIS land cover uncer-
tainty in regional climate simulations. Climate Dynamics, 49(11-12), pp.4047-4059.
Linn, J.F. 2012. Scaling up in agriculture, rural development, and nutrition (Vol. 19). International Food
Policy Research Institute (IFPRI).
Lobell, D.B., 2013a. The use of satellite data for crop yield gap analysis. Field Crops Research, 143,
pp.56-64.
Lobell, D.B., 2013b. Errors in climate datasets and their effects on statistical crop models. Agri-
cultural and Forest Meteorology, 170, pp.58-66.
Luckett, B.G., DeClerck, F.A., Fanzo, J., Mundorf, A.R. and Rose, D., 2015. Application of the
nutrition functional diversity indicator to assess food system contributions to dietary diversity and
sustainable diets of Malawian households. Public Health Nutrition, 18(13), pp.2479-2487.
Maes, J. and Jacobs, S., 2017. Natureâ€based solutions for Europe's sustainable development. Con-
servation Letters, 10(1), pp.121-124.
Malawi Vulnerability Assessment Committee (VAC), 2005. Malawi baseline livelihood profiles: Version
1. http://pdf.usaid.gov/pdf_docs/Pnadg538.pdf. Accessed 3 November 2016.
Manyonda, K., 1998. Official opening of the soil fertility network workshop: CIMMYT—Soil
fertility research for maize-based farming systems in Malawi and Zimbabwe. In Proceedings of the
Soil Fert Net Results and Planning Workshop. Mutare: Soil Fert Net and CIMMYT-Zimbabwe.
Maxwell, D., 1999. The political economy of urban food security in Sub-Saharan Africa. World
Development, 27(11), pp.1939-1953.
Mead, R. and Willey, R.W., 1980. The concept of a ‘land equivalent ratio’ and advantages in yields
from intercropping. Experimental Agriculture, 16(3), pp.217-228.
Mekbib, F., 2009. Farmers' breeding of sorghum in the centre of diversity, Ethiopia: I. Socio-
ecotype differentiation, varietal mixture and selection efficiency. Maydica, 54(1), pp.25-37.
Messina, J., Adhikari, U., Carroll, J., Chikowo, R., DeVisser, M., Dodge, L., Fan, P., Langley, S.,
Lin, S., Me-Nsope, N. and Moore, N., 2014. Population growth, climate change and pressure on
the land: eastern and southern Africa. Michigan State University, East Lansing.
133
Messina, J.P., Peter, B.G. and Snapp, S.S., 2017. Re-evaluating the Malawian farm input subsidy
programme. Nature Plants, 3(4), p.17013.
Mhango, W.G., Snapp, S.S. and Phiri, G.Y., 2013. Opportunities and constraints to legume diver-
sification for sustainable maize production on smallholder farms in Malawi. Renewable Agriculture
and Food Systems, 28(3), pp.234-244.
Millar, J. and Connell, J., 2010. Strategies for scaling out impacts from agricultural systems change:
the case of forages and livestock production in Laos. Agriculture and Human Values, 27(2), pp.213-
225.
Mishra, A., Hansen, J.W., Dingkuhn, M., Baron, C., Traoré, S.B., Ndiaye, O. and Ward, M.N.,
2008. Sorghum yield prediction from seasonal rainfall forecasts in Burkina Faso. Agricultural and
Forest Meteorology, 148(11), pp.1798-1814.
Morton, J.F., 2007. The impact of climate change on smallholder and subsistence agriculture. Pro-
ceedings of the National Academy of Sciences, 104(50), pp.19680-19685.
Mulla, D.J., 2013. Twenty five years of remote sensing in precision agriculture: Key advances and
remaining knowledge gaps. Biosystems Engineering, 114(4), pp.358-371.
Mungai, L.M., Snapp, S., Messina, J.P., Chikowo, R., Smith, A., Anders, E., Richardson, R.B. and
Li, G., 2016. Smallholder farms and the potential for sustainable intensification. Frontiers in Plant
Science, 7, p.1720.
Muthoni, F.K., Sseguya, H., Bekunda, M. and Hoeschle-Zeledon, I., 2016. Identifying recommen-
dation domains for scaling improved crop varieties in Tanzania. In Eighth annual ICT4D Conference;
from innovation to impact: using technology to achieve the sustainable development goals. Nairobi, Kenya.
Muyanga, M. and Jayne, T.S., 2014. Effects of rising rural population density on smallholder agri-
culture in Kenya. Food Policy, 48, pp.98-113.
Myaka, F.M., Sakala, W.D., Adu-Gyamfi, J.J., Kamalongo, D., Ngwira, A., Odgaard, R., Nielsen,
N.E. and Høgh-Jensen, H., 2006. Yields and accumulations of N and P in farmer-managed inter-
crops of maize–pigeonpea in semi-arid Africa. Plant and Soil, 285(1-2), pp.207-220.
Nalepa, R.A. and Bauer, D.M., 2012. Marginal lands: the role of remote sensing in constructing
landscapes for agrofuel development. Journal of Peasant Studies, 39(2), pp.403-422.
Nalepa, R.A., Gianotti, A.G.S. and Bauer, D.M., 2017. Marginal land and the global land rush: A
spatial exploration of contested lands and state-directed development in contemporary Ethio-
pia. Geoforum, 82, pp.237-251.
134
NASA/JAXA Tropical Rainfall Measuring Mission Project (TRMM), 2016. TRMM Product 3b43.
Goddard Space Flight Center Distributed Active Archive Center
(GSFC DAAC).
http://disc.sci.gsfc.nasa.gov/precipitation/documentation/TRMM_RE-
ADME/TRMM_3B43_readme.shtml. Accessed 23 November 2016.
NASA LP DAAC, 2015a. MODIS Land Surface Temperature (MYD11B3) Version 006. NASA
EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science (EROS) Cen-
ter, Sioux Falls, South Dakota. https://lpdaac.usgs.gov/dataset_discovery/modis/modis_prod-
ucts_table/myd11b3_v006. Accessed 23 November 2016.
NASA LP DAAC, 2015b. MODIS Land Surface Temperature (MOD11A2) Version 005. NASA
EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science (EROS) Cen-
ter, Sioux Falls, South Dakota. https://lpdaac.usgs.gov/dataset_discovery/modis/modis_prod-
ucts_table/mod11a2. Accessed 23 November 2016.
NASA LP DAAC, 2015c. MODIS Net Primary Productivity (MOD17A3) Version 055.NASA
EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science (EROS) Cen-
ter, Sioux Falls, South Dakota. https://lpdaac.usgs.gov/dataset_discovery/modis/modis_prod-
ucts_table/mod17a3. Accessed 23 November 2016.
NASA LP DAAC, 2015d. MODIS Land Cover Type (MCD12Q1) Version 051. NASA EOSDIS
Land Processes DAAC, USGS Earth Resources Observation and Science (EROS) Center, Sioux
Falls, South Dakota. https://lpdaac.usgs.gov/dataset_discovery/modis/modis_products_ta-
ble/mcd12q1. Accessed 23 November 2016.
National Statistical Office of Malawi (NSO), 2005. Malawi Integrated Household Survey (IHS2).
Zomba, Malawi: Government of Malawi.
Nelson, M., Henzell, T., and Malik, S., 1999. Consultative Group on International Agricultural Research
(CGIAR) research priorities for marginal lands. Rome: Food and Agriculture Organization of the
United Nations. http://www.fao.org/Wairdocs/TAC/X5784E/x5784e00.htm. Accessed 23 No-
vember 2016.
Nelson, G.C., Rosegrant, M.W., Koo, J., Robertson, R., Sulser, T., Zhu, T., Ringler, C., Msangi,
S., Palazzo, A., Batka, M. and Magalhaes, M., 2009. Climate change: Impact on agriculture and costs of
adaptation (Vol. 21). International Food Policy Research Institute (IFPRI).
Nesshöver, C., Assmuth, T., Irvine, K.N., Rusch, G.M., Waylen, K.A., Delbaere, B., Haase, D.,
Jones-Walters, L., Keune, H., Kovacs, E. and Krauze, K., 2017. The science, policy and practice
of nature-based solutions: An interdisciplinary perspective. Science of the Total Environment, 579,
pp.1215-1227.
Neumann, R.P., 2009. Political ecology: theorizing scale. Progress in Human Geography, 33(3),
pp.398-406.
135
Neumann, K., Verburg, P.H., Stehfest, E. and Müller, C., 2010. The yield gap of global grain
production: A spatial analysis. Agricultural Systems, 103(5), pp.316-326.
Ngwira, A.R., Aune, J.B. and Mkwinda, S., 2012. On-farm evaluation of yield and economic ben-
efit of short term maize legume intercropping systems under conservation agriculture in Ma-
lawi. Field Crops Research, 132, pp.149-157.
Nolte, K., 2014. Large-scale agricultural investments under poor land governance in Zambia. Land
Use Policy, 38, pp.698-706.
Ojiewo, C., Keatinge, D.J., Hughes, J., Tenkouano, A., Nair, R., Varshney, R., Siambi, M., Monyo,
E., Gangaâ€Rao, N.V.P.R. and Silim, S., 2015. The Role of Vegetables and Legumes in Assuring
Food, Nutrition, and Income Security for Vulnerable Groups in Subâ€Saharan Africa. World Medical
& Health Policy, 7(3), pp.187-210.
Okigbo, B.N. and Greenland, D.J., 1976. Intercropping systems in tropical Africa. In Symposium
on Multiple Cropping, Knoxville, Tennessee.
Okiror, M.A., 1986. Breeding for resistance to Fusarium wilt of pigeon pea (Cajanus Cajan/-L./MILLSP)
in Kenya (Doctoral dissertation).
Omanga, P.A., Summerfield, R.J. and Qi, A., 1996. Flowering in pigeonpea (Cajanus cajan) in
Kenya: responses of medium-and late-maturing genotypes to location and date of sowing. Exper-
imental Agriculture, 32(2), pp.111-128.
Openshaw, S., 1983. The modifiable areal unit problem. In Concepts and techniques in modern geography
No. 38 (pp.1-41). Study Group in Quantitative Methods of the Institute of British Geographers.
Norwich, UK: Geo Books.
Ortega, D.L., Waldman, K.B., Richardson, R.B., Clay, D.C. and Snapp, S., 2016. Sustainable in-
tensification and farmer preferences for crop system attributes: evidence from Malawi’s central
and southern regions. World Development, 87, pp.139-151.
Pachico, D., 2004. Scaling up and out: achieving widespread impact through agricultural research (Vol. 3).
CIAT.
Pajares, G., 2015. Overview and current status of remote sensing applications based on unmanned
aerial vehicles (UAVs). Photogrammetric Engineering & Remote Sensing, 81(4), pp.281-330.
Panda, S.S., Ames, D.P. and Panigrahi, S., 2010. Application of vegetation indices for agricultural
crop yield prediction using neural network techniques. Remote Sensing, 2(3), pp.673-696.
Park, J.K. and Park, J.H., 2015. Crops classification using imagery of unmanned aerial vehicle
(UAV). Journal of the Korean Society of Agricultural Engineers, 57(6), pp.91-97.
136
Parr, J.F., Stewart, B.A., Hornick, S.B. and Singh, R.P., 1990. Improving the sustainability of dry-
land farming systems: a global perspective. In Advances in Soil Science (pp. 1-8). Springer, New York,
NY.
Parrot Drones SA, 2017. Parrot Sequoia Technical Specifications.
Paterson, A.H., Bowers, J.E., Bruggmann, R., Dubchak, I., Grimwood, J., Gundlach, H., Haberer,
G., Hellsten, U., Mitros, T., Poliakov, A., Schmutz, J., Spannagl, M., Tang, H., Wang, X., Wicker,
T., Bharti, A.K., Chapman, J., Alex Feltus, F., Gowik, U., Grigoriev, I.V., Lyons, E., Maher, C.A.,
Martis, M., Narechania, A., Otillar, R.P., Penning, Bryan W., Salamov, A.A., Wang, Y., Zhang, L.,
Carpita, N.C., Freeling, M., Gingle, A.R., Hash, C.T., Keller, B., Klein, P., Kresovich, S., McCann,
M.C., Ming, R., Peterson, D.G., Mehboob-ur-Rahman, Ware, D., Westhoff, P., Mayer, K.F.X.,
Messing, J., Rokhsar, D.S., 2009. The Sorghum bicolor genome and the diversification of
grasses. Nature, 457(7229), p.551.
Paterson, A.H., Cox, T.S., Wegian, K., Navarro, M., 2014. Viewpoint: multiple-harvest sorghums
toward improved food security. Perennial Crops for Food Security. In Proceedings of the FAO Expert
Workshop. FAO, Rome.
Pauw, K., Thurlow, J., Bachu, M. and Van Seventer, D.E., 2011. The economic costs of extreme
weather events: a hydrometeorological CGE analysis for Malawi. Environment and Development Eco-
nomics, 16(2), pp.177-198.
Peter, B.G., Mungai, L.M., Messina, J.P. and Snapp, S.S., 2017a. Nature-based agricultural solu-
tions: Scaling perennial grains across Africa. Environmental Research, 159, pp.283-290.
Peter, B.G., Messina, J.P., Frake, A.N. and Snapp, S.S., 2017b. Scaling agricultural innovations:
pigeonpea in Malawi. The Professional Geographer, 70(2), pp.239-250.
Peter, B.G., Messina, J.P. and Snapp, S.S., 2018. A Multiscalar Approach to Mapping Marginal
Agricultural Land: Smallholder Agriculture in Malawi. Annals of the American Association of Geogra-
phers, 108(4), pp.989-1005.
Petersen, B. and Snapp, S., 2015. What is sustainable intensification? Views from experts. Land
Use Policy, 46, pp.1-10.
Phiri, F. and Macharia. J., 2016. Malawi declares national disaster after drought as food shortage
looms. Reuters. http://www.reuters.com/article/us-africa-drought-malawi-idUSKCN0XB1H3.
Accessed 23 November 2016.
Pimentel, D., Jackson, W., Bender, M. and Pickett, W., 1986. Perennial grains–an ecology of new
crops. Interdisciplinary Science Reviews, 11(1), pp.42-49.
Pineux, N., Lisein, J., Swerts, G., Bielders, C.L., Lejeune, P., Colinet, G. and Degré, A., 2017. Can
DEM time series produced by UAV be used to quantify diffuse erosion in an agricultural water-
shed? Geomorphology, 280, pp.122-136.
137
Pingali, P.L., 2001. CIMMYT 1999/2000 World maize facts and trends. Meeting world maize
needs: Technological opportunities and priorities for the public sector. International Maize and
Wheat Improvement Center, El Batán, Mexico.
Pittman, K., Hansen, M.C., Becker-Reshef, I., Potapov, P.V. and Justice, C.O., 2010. Estimating
global cropland extent with multi-year MODIS data. Remote Sensing, 2(7), pp.1844-1863.
Pix4D SA, 2017. Pix4Dmapper 4.1 User Manual. Pix4D SA: Lausanne, Switzerland.
Planet Team, 2017. Planet Application Program Interface: In Space for Life on Earth. San Fran-
cisco, CA. https://api.planet.com
Postel, S.L., 2000. Entering an era of water scarcity: the challenges ahead. Ecological Applica-
tions, 10(4), pp.941-948.
Puget, P. and Drinkwater, L.E., 2001. Short-term dynamics of root-and shoot-derived carbon
from a leguminous green manure. Soil Science Society of America Journal, 65(3), pp.771-779.
Reeves, M.C., Zhao, M. and Running, S.W., 2005. Usefulness and limits of MODIS GPP for
estimating wheat yield. International Journal of Remote Sensing, 26(7), pp.1403-1421.
Reilly, J.M. and Schimmelpfennig, D., 1999. Agricultural impact assessment, vulnerability, and the
scope for adaptation. Climatic Change, 43(4), pp.745-788.
Ricker-Gilbert, J., Jumbe, C. and Chamberlin, J., 2014. How does population density influence
agricultural intensification and productivity? Evidence from Malawi. Food Policy, 48, pp.114-128.
Robertson, G.P., Gross, K.L., Hamilton, S.K., Landis, D.A., Schmidt, T.M., Snapp, S.S. and Swin-
ton, S.M., 2014. Farming for ecosystem services: An ecological approach to production agricul-
ture. BioScience, 64(5), pp.404-415.
Rogé, P., Snapp, S., Kakwera, M.N., Mungai, L., Jambo, I. and Peter, B., 2016. Ratooning and
perennial staple crops in Malawi. A review. Agronomy for Sustainable Development, 36(3), p.50.
Rufino, M.C., Thornton, P.K., Mutie, I., Jones, P.G., Van Wijk, M.T. and Herrero, M., 2013.
Transitions in agro-pastoralist systems of East Africa: impacts on food security and poverty. Ag-
riculture, Ecosystems & Environment, 179, pp.215-230.
Sachs, J., 2005. Investing in development: A practical plan to achieve the Millennium Development Goals. New
York: United Nations.
Sakala, W.D., Cadisch, G. and Giller, K.E., 2000. Interactions between residues of maize and
pigeonpea and mineral N fertilizers during decomposition and N mineralization. Soil Biology and
Biochemistry, 32(5), pp.679-688.
Sanchez, P.A., 2002. Soil fertility and hunger in Africa. Science, 295(5562), pp.2019-2020.
138
Sanchez, P.A., 2015. En route to plentiful food production in Africa. Nature Plants, 1(1), p.14014.
Sánchez, B., Rasmussen, A. and Porter, J.R., 2014. Temperatures and the growth and development
of maize and rice: a review. Global Change Biology, 20(2), pp.408-417.
Sardana, V., Sharma, P., Sheoran, P., 2010. Growth and production of pulses. soils, plant growth
and crop production Vol.III. Encyclopedia of Life Support Systems (EOLSS).
Sawyer, J.E., 1994. Concepts of variable rate technology with considerations for fertilizer applica-
tion. Journal of Production Agriculture, 7(2), pp.195-201.
Schneider, U.A., HavlÃk, P., Schmid, E., Valin, H., Mosnier, A., Obersteiner, M., Böttcher, H.,
Skalský, R., BalkoviÄ, J., Sauer, T. and Fritz, S., 2011. Impacts of population growth, economic
development, and technical change on global food production and consumption. Agricultural Sys-
tems, 104(2), pp.204-215.
Schulz, S., Honlonkou, A.N., Carsky, R.J., Manyong, V.M. and Oyewole, B.D., 2003. Alternatives
to mucuna for soil fertility management in southern Benin: Farmer perception and use of tradi-
tional and exotic grain legumes. Experimental Agriculture, 39(3), pp.267-278.
Sexton, W.T., Dull, C.W. and Szaro, R.C., 1998. Implementing ecosystem management: a frame-
work for remotely sensed information at multiple scales. Landscape and Urban Planning, 40(1-3),
pp.173-184.
Shanahan, J.F., Schepers, J.S., Francis, D.D., Varvel, G.E., Wilhelm, W.W., Tringe, J.M.,
Schlemmer, M.R. and Major, D.J., 2001. Use of remote-sensing imagery to estimate corn grain
yield. Agronomy Journal, 93(3), pp.583-589.
Silim, S.N. and Omanga, P.A., 2001. The response of short-duration pigeonpea lines to variation
in temperature under field conditions in Kenya. Field Crops Research, 72(2), pp.97-108.
Simtowe, F., Asfaw, S., Shiferaw, B., Siambi, M., Monyo, E., Muricho, G., Abate, T., Silim, S.,
Ganga Rao, N.V.P.R. and Madzonga, O., 2010. Socioeconomic Assessment of Pigeonpea and
Groundnut Production Conditions–Farmer Technology Choice, Market Linkages, Institutions
and Poverty in Rural Malawi. International Crops Research Institute for the Sem-Arid Tropics
(ICRISAT) Research Report No. 6. Hyderabad, India.
Sirrine, D., Shennan, C., Snapp, S., Kanyama-Phiri, G., Kamanga, B. and Sirrine, J.R., 2010. Im-
proving recommendations resulting from on-farm research: agroforestry, risk, profitability and
vulnerability in southern Malawi. International Journal of Agricultural Sustainability, 8(4), pp.290-304.
Snapp, S.S. and Silim, S.N., 2002. Farmer preferences and legume intensification for low nutrient
environments. Plant and Soil 245(1), pp.181-192.
Snapp, S. and Heong, K.L., 2003. Scaling up and out. In Managing natural resources for sustainable
livelihoods: Uniting science and participation (pp.67-87).
139
Snapp, S.S., Jones, R.B., Minja, E.M., Rusike, J. and Silim, S.N., 2003. Pigeon pea for Africa: A
versatile vegetable—And more. HortScience, 38(6), pp.1073-1079.
Snapp, S.S., Blackie, M.J., Gilbert, R.A., Bezner-Kerr, R. and Kanyama-Phiri, G.Y., 2010. Biodi-
versity can support a greener revolution in Africa. Proceedings of the National Academy of Sci-
ences, 107(48), pp.20840-20845.
Snapp, S., Jayne, T.S., Mhango, W., Benson, T., and Ricker-Gilbert, J., 2014. Maize yield response
to nitrogen in Malawi’s smallholder production systems. Malawi Strategy Support Program, Work-
ing Paper 9. Washington, DC: ICRISAT.
Snapp, S.S., Cox, C.M. and Peter, B.G., 2019a. Multipurpose legumes for smallholders in sub-
Saharan Africa: Identification of promising ‘scale out’ options. Global Food Security, 23, pp.22-32.
Snapp, S., Roge, P., Okori, P., Chikowo, R., Peter, B. and Messina, J., 2019b. Perennial grains for
Africa: Possibility or pipedream? Experimental Agriculture, 55(2), pp.251-272.
Spectrum Technologies, Inc., 2019. SPAD 502 Plus Chlorophyll Meter. https://www.spec-
meters.com/nutrient-management/chlorophyll-meters/chlorophyll/spad502p/
Sripada, R.P., Heiniger, R.W., White, J.G. and Meijer, A.D., 2006. Aerial color infrared photog-
raphy for determining early in-season nitrogen requirements in corn. Agronomy Journal, 98(4),
pp.968-977.
Stark, B., McGee, M. and Chen, Y., 2015, June. Short wave infrared (SWIR) imaging systems using
small Unmanned Aerial Systems (sUAS). In 2015 International Conference on Unmanned Aircraft Sys-
tems (ICUAS) (pp. 495-501). IEEE.
Sutcliffe, C., Dougill, A.J. and Quinn, C.H., 2016. Evidence and perceptions of rainfall change in
Malawi: Do maize cultivar choices enhance climate change adaptation in sub-Saharan Africa? Re-
gional Environmental Change, 16(4), pp.1215-1224.
Syroka, J. and Nucifora, A., 2010. National drought insurance for Malawi. Policy Research Work-
ing Paper 5169, The World Bank, Washington, DC.
Tadross, M., Suarez, P., Lotsch, A., Hachigonta, S., Mdoka, M., Unganai, L., Lucio, F., Kamdonyo,
D. and Muchinda, M., 2009. Growing-season rainfall and scenarios of future change in southeast
Africa: implications for cultivating maize. Climate Research, 40(2-3), pp.147-161.
TerAvest, D., Wandschneider, P.R., Thierfelder, C. and Reganold, J.P., 2019. Diversifying conser-
vation agriculture and conventional tillage cropping systems to improve the wellbeing of small-
holder farmers in Malawi. Agricultural Systems, 171, pp.23-35.
Thierfelder, C., Chisui, J.L., Gama, M., Cheesman, S., Jere, Z.D., Bunderson, W.T., Eash, N.S.
and Rusinamhodzi, L., 2013. Maize-based conservation agriculture systems in Malawi: long-term
trends in productivity. Field Crops Research, 142, pp.47-57.
140
agriculture:
and
food
2016.
Sustainable
Agriculture
and Evidence Exchange
Tucker, C.J., 1979. Red and photographic infrared linear combinations for monitoring vegeta-
tion. Remote Sensing of Environment, 8(2), pp.127-150.
United Nations, 2018 About the Sustainable Development Goals. https://www.un.org/sustaina-
bledevelopment/sustainable-development-goals/
United Nations World Food Programme (WFP), 2016. What are the current issues in Malawi.
World Food Program. https://www.wfp.org/countries/malawi. Accessed 23 November 2016.
USAID, 2014. Scaling up the adoption and use of agricultural technologies. Paper presented at
the Global Learning
(GLEE). Bangkok, Thailand.
https://agrilinks.org/sites/default/files/resource/files/ScalingUpAgTechGLEEReportThai-
land2014_0.pdf. Accessed 3 November 2016.
security.
USAID,
https://www.usaid.gov/what-we-do/agricultureand-food-security/investing-sustainable-agricul-
ture. Accessed 3 November 2016.
Valbuena, D., Tui, S.H.K., Erenstein, O., Teufel, N., Duncan, A., Abdoulaye, T., Swain, B., Me-
konnen, K., Germaine, I. and Gérard, B., 2015. Identifying determinants, pressures and trade-offs
of crop residue use in mixed smallholder farms in Sub-Saharan Africa and South Asia. Agricultural
Systems, 134, pp.107-118.
Valenzuela, H. and Smith, J., 2002. Pigeonpea. Sustainable Agriculture Green Manure Crops. Co-
operative Extension Service, College of Tropical Agriculture and Human Resources
(CTAHR). Manoa, Honolulu, Hawaii: University of Hawaii.
Vanlauwe, B., Kihara, J., Chivenge, P., Pypers, P., Coe, R. and Six, J., 2011. Agronomic use effi-
ciency of N fertilizer in maize-based systems in sub-Saharan Africa within the context of integrated
soil fertility management. Plant and Soil, 339(1-2), pp.35-50.
Vincini, M., Frazzi, E.R.M.E.S. and D’Alessio, P.A.O.L.O., 2008. A broad-band leaf chlorophyll
vegetation index at the canopy scale. Precision Agriculture, 9(5), pp.303-319.
Vizy, E.K., Cook, K.H., Chimphamba, J. and McCusker, B., 2015. Projected changes in Malawi’s
growing season. Climate Dynamics, 45(5-6), pp.1673-1698.
Waldman, K.B., Ortega, D.L., Richardson, R.B., Clay, D.C. and Snapp, S., 2016. Preferences for
legume attributes in maize-legume cropping systems in Malawi. Food Security, 8(6), pp.1087-1099.
Waldman, K.B., Ortega, D.L., Richardson, R.B. and Snapp, S.S., 2017. Estimating demand for
perennial pigeon pea in Malawi using choice experiments. Ecological Economics, 131, pp.222-230.
141
Walker, T., Silim, S., Cunguara, B., Donovan, C., Rao, P.P. and Amane, M., 2015. Pigeonpea in
Mozambique: An emerging success story of crop expansion in smallholder agriculture (No. 1098-
2016-88859). Working Paper 78E. Maputo, Mozambique: Ministry of Agriculture and Food Se-
curity, Directorate of Economics.
Wall, L., Larocque, D. and Léger, P.M., 2008. The early explanatory power of NDVI in crop yield
modelling. International Journal of Remote Sensing, 29(8), pp.2211-2225.
Wheeler, T. and Von Braun, J., 2013. Climate change impacts on global food security. Sci-
ence, 341(6145), pp.508-513.
Wood, L. and Moriniere, L., 2013. Malawi climate change vulnerability assessment annex D. Crop
Phenology. USAID–African and Latin American Resilience to Climate Change (ARCC).
Wood, S., Sebastian, K., Nachtergaele, F., Nielsen, D. and Dai, A., 1999. Spatial aspects of the design
and targeting of agricultural development strategies (No. 581-2016-39428).
World Bank, 2016. Malawi. http://data.worldbank.org/country/malawi?viewDchart. Accessed
23 November 2016.
Yeager, C.D. and Steiger, T., 2013. Applied geography in a digital age: The case for mixed meth-
ods. Applied Geography, 39, pp.1-4.
Zhang, H. and Oweis, T., 1999. Water–yield relations and optimal irrigation scheduling of wheat
in the Mediterranean region. Agricultural Water Management, 38(3), pp.195-211.
Zhang, X., Shi, L., Jia, X., Seielstad, G. and Helgason, C., 2010. Zone mapping application for
precision-farming: a decision support tool for variable rate application. Precision Agriculture, 11(2),
pp.103-114.
Zhang, C. and Kovacs, J.M., 2012. The application of small unmanned aerial systems for precision
agriculture: a review. Precision Agriculture, 13(6), pp.693-712.
Zhu, Y., Wang, W. and Yao, X., 2012. Estimating leaf nitrogen concentration (LNC) of cereal
crops with hyperspectral data. Hyperspectral Remote Sensing of Vegetation (pp.187-206). CRC Press,
Boca Raton, FL.
142