Agricultural adaptation to drought

Abstract

This research integrates geospatial and social datasets to understand the adaptive strategies employed by individuals, communities, and institutions during periods of drought. I apply dimension reduction techniques to remotely sensed data to identify agricultural communities in which cultivation occurred during an extreme drought. Two dominant adaptive strategies emerged from this research: crop diversification and a complex land reallocation process known locally as bethma. I supplemented this qualitative research with Bayesian analysis of project survey data to identify the multi-scalar factors driving participation in these adaptive behaviors. Results suggest that farmers with more assets (agrowell, land, higher socio-economic status) are more likely to engage in adaptive behaviors. To better understand the role of farmers’ risk perception in influencing adaptive behaviors, I constructed an agent-based model to assess the extent to which farmer decision heuristics affect participation in crop diversification and bethma. Simulation results suggest that though environmental changes may produce sudden disruptions in agricultural outcomes, the variations in these outcomes are strongly influenced by the mental models farmers use to make agricultural decisions. I also collaborated with a data scientist to apply machine learning techniques to large remotely sensed datasets to create an open-source prediction software to forecast vegetation health. This is particularly important in tropical countries where cloud cover significantly reduces data availability. The latest version of the software is global in coverage and performs extremely well across agroecological contexts, time, and levels of data availability. The results of my research in Sri Lanka will increase the capacity of decision makers to monitor agricultural systems, identify and promote successful adaptive strategies, and increase agricultural adaptation to changing climate.