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 classification 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 intensification, perennial grains and long-

duration varieties offer a nature-based solution for improving farm productivity and smallholder 

livelihoods in suboptimal agricultural areas. Specific 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 intensification (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 intensified 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 benefit 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 five 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 benefit from adoption of technologies that support 

sustainable principles and the complex system processes of nature. At the same time, the “net 

benefit 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 

intensification 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, field/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 

fixation 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 field 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-

versification 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 benefit 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 classifications. 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 benefit 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 classification 
thresholds. Classifications 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  classification  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 reclassified 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 identified here using suboptimal temperature and precipitation ranges 

for maize (Table 2.1, Figure 2.1), in combination with an interannual variability model that iden-

tifies 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 classifications, 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 benefit 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 

identified 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 significantly 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 fine 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 Benefit 

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 benefit 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 classifications 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 (Rufino et al. 2013). Many African households are familiar with cultivation needs 

of perennial grains through generational knowledge (Rufino 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 benefits (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 benefits 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  fill  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 benefit 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 diversification offers unique environmental benefits, particularly in locations 

where soil fertility decline is limiting crop yields. The ability of pigeonpea to biologically fix 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 deficit 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 findings 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 intensification concepts have broad potential. In rural Africa, where 

land availability and fertilizer access are often limited, smallholder farming systems may benefit 

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 benefit 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 environments” 

(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 life”; 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 fine 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 diversification 

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 benefit 

from pigeonpea integration, with 9 percent receiving predictable and substantial benefits. These 

findings 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 definitions 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 specific, 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 profitability, 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-

tensification and extensification. In intensified 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  intensification  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 intensified agricultural productivity (USAID 

2016). 

Extensification 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 

extensification 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 extensification 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 extensification have been largely reduced to protected or marginal areas, nei-

ther of which are viable or sustainable for long-term production demand. Thus, intensification 

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 fix nitrogen and enhance 

phosphorus solubilization for soil rehabilitation (Snapp et al. 2010; Ngwira, Aune, and Mkwinda 

2012; Mhango, Snapp, and Phiri 2013). The benefits 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 profitability 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 fine spatial resolutions. For intervention initiatives to succeed, they need 

to be delivered in areas where benefits 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, fine 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-

sification 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 

fine spatial resolution, and on a pixel basis. Highlighted here are areas and sites where pigeonpea 

integration will provide predictable and substantial benefits in a classification 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 beneficial 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 benefit 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 reclassification of the marginal maize map tailored to soil marginality 

reveals areas where pigeonpea deployment would prove most effective. This reclassification 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 defined 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 reclassified 

to a “better bet” option for maximum benefit 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 benefit 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 reclassifications include locations that are suitable for pigeonpea, but benefits 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 classified as highly unpredictable. 

The following are the resulting categories: (1) Better Bet—maximum benefit from pigeonpea in-

tegration; (2) substantial benefit from pigeonpea integration; (3) suitable for pigeonpea— likely 

benefits from extension or market development; (4) suitable for pigeonpea—benefits somewhat 

unpredictable; and (5) suitable for pigeonpea—benefits 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 field survey in which we spoke with extension officers 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 field 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 confirm 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 findings suggest maximum benefit 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 benefit 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 benefit 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 benefits 

are less predictable. Twenty-one percent is suitable for pigeonpea but soil is not a driver of mar-

ginal maize; therefore benefits 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 benefits from pigeonpea adoption, with varying degrees of predictability and ef-

fectiveness, at least 9 percent of which should receive highly predictable and beneficial 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 profitable returns. To build soil organic matter requires sufficient or-

ganic  matter  inputs,  delivered  efficiently  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-

sified livestock systems that transfer manure to crop fields 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 diversification 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 profitable 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 findings 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 confirmation 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 benefit 

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 figures 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 undiversified, 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 fixation, and improved maize yields across 

many specific locations in Malawi. The maps and methodology herein are resources that might be 

used to intelligently target locations in need of intervention, where benefits will meet expectations 

and minimize risk. The fine 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 specific 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 field, 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 SPAD‐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., García-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 trade‐
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., Weltzien‐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  land‐sparing/land‐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. Nature‐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., Ganga‐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 Sub‐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., Havlík, P., Schmid, E., Valin, H., Mosnier, A., Obersteiner, M., Böttcher, H., 
Skalský, R., Balkovič, 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