SIFting Through Surveys: Leveraging Satellite Data for Improved Understanding of Crop Yields and Food Insecurity in Sub-Saharan Africa and Beyond

Abstract

Background: Sub-Saharan Africa (SSA) is one of few regions of the world that has recently experienced a significant reversal of the historical trend of declining food insecurity. The precise quantification of food insecurity is essential for comprehending the extent of the issue and for devising efficacious policies and programs to mitigate it. A vital component of food insecurity assessment is the accurate measurement of crop yields, which contributes to food availability estimations and potential food shortage predictions. However, gauging crop yields presents a multifaceted challenge, particularly in low-income nations constrained by resources and infrastructure. In this dissertation, the challenges and implications of accurately monitoring crop yields and food security in the context of smallholder farmers in SSA are investigated. Both remote sensing data and ground-level survey data are leveraged to see whether a combination of the two disparate data sources can explain and predict crop yields and food insecurity among smallholder farming households. I also critically examine the current “gold standard” of global yield reporting – that is, the Food and Agriculture Organization (FAO) reported yields – to see where and when this standard potentially fails. Overall, this research emphasizes the urgent importance of accurate crop yield estimates estimates for understanding and addressing the food security crisis today. Methods: This body of work explores the relationship between satellite-derived contiguous solar-induced fluorescence (CSIF) measurements and ground-level crop yield and food insecurity indicators, presenting a novel methodology tested across four studies, with Bayesian generalized multilevel models (BMMs) as the methodological foundation. CSIF is a measurement of the natural phenomenon that occurs when plants absorb sunlight during photosynthesis and re-emit a small amount of that energy in the form of fluorescence. This fluorescence measurement has been validated as a useful remote sensing tool to monitor and predict crop yields in various parts of the world, though not in remote rural areas where the only existing “groundtruth” (i.e. empirical evidence) consists of surveys not intended for rigorous crop yield analysis. In this particular and challenging context, I attempt to model the relationships between survey-derived crop yields and remote-sensed CSIF, as well as that of self-reported food inadequacy and CSIF, while accounting for time and georeferenced locations as fixed and random effects, respectively. The ground-level survey data for the first three chapters comes from the Living Standards Measurement Study - Integrated Surveys on Agriculture. In the last chapter, the ground-level data comes from the FAO. All models are designed for a specific outcome distribution; in this study, robust linear, logistic, and negative binomial families are fit to the data. All BMMs are fit using the Hamiltonian Monte Carlo (HMC) algorithm, which offers computational efficiency and validity in estimating posterior probabilities of parameter values. Resulting estimates are accompanied by 95% credible intervals, generated via four chains of 2,000 iterations each. Lastly, throughout the four studies, additional methods are used to fully explore the questions at hand: these include correlation analyses, simple linear regressions, and principal component analysis. Results: In Chapter 1, I successfully conduct a methodological proof-of-concept in relating maize yields to CSIF in the United States of America. However, when the methodology is then applied to Nigeria, I fail to find a non-null relationship between maize yields and CSIF, despite extensively controlling for other environmental factors. In Chapter 2, by pooling across Nigeria, Malawi, Tanzania, and Uganda, I find a statistically significant predictive effect of CSIF on maize, bean, and pigeonpea yields. However, mixed results at the national and sub-national level warrant further study before CSIF can be used as a reliable yield proxy at policy-relevant scales. In Chapter 3, I find that lagged CSIF partially explains food inadequacy in Uganda, though only if it is embedded in a model that includes prior lagged food inadequacy status. While CSIF alone cannot yet predict food inadequacy, I do find that lagged CSIF can predict up to 14% of variability in present consumption of certain foods. In Chapter 4, I show that countries with high-yield production, high-income, and low-corruption have better agreement between crop yields reported by the FAO and my CSIF proxy. I explore maize prices as a complement to (and potentially replacement for) crop yield estimates, and find surprising corroboration between local market retail prices and lagged CSIF in Zambia, providing encouraging evidence for highly localized price-CSIF alignment. Conclusion: Presently, CSIF alone is not yet a reliably consistent proxy for crop yields or food inadequacy across all of SSA; nevertheless, I have provided evidence for its potential to be so one day. I have proven the theoretical framework for this endeavor by showing that CSIF is strongly associated with maize and other major crop yields in the United States, indicating it is an appropriate candidate for crop yield proxy in that context. In SSA, my findings suggest that satellite-derived measurements can estimate yields and food inadequacy in very specific contexts, such as maize yields when pooled across the four SSA countries, or certain food item consumption in Uganda. At present, however, such efforts are limited by data quality and scale, both at the ground-level and from space. From this present work, I have also identified key research priorities going forward, namely the pursuit of more reliable survey data, understanding regional variations in the yield-CSIF relationship, exploring the link between food consumption and CSIF, and refining FAO-reported yield measurements. These research areas are not only critical in the context of worsening food insecurity, but they are also wholly achievable, especially given the wealth of freely available data sources that exist today – several of which are utilized in the present studies. Overall, this dissertation highlights the importance of understanding smallholder farming and its food security implications in SSA, and emphasizes the urgent need for continued research in this critical domain.