Predicting Crop Losses using Machine Learning

Timely forecasting of crop production losses from droughts can help government agencies and NGOs to strategize when to deploy measures to mitigate food insecurity. Regional, large-scale reporting exist from various sources, yet they may not be adequate to detect the village-level losses due to the inherent spatiotemporal variability often at the substantial level. A new study co-authored by James M. Warner from IFPRI, Predicting high-magnitude, low-frequency crop losses using machine learning: an application to cereal crops in Ethiopia, proposed a data fusion method combining remotely sensed data with agricultural survey data using machine learning algorithms to improve the accuracy and resolution of crop loss predictions.


Timely and accurate agricultural impact assessments for droughts are critical for designing appropriate interventions and policy. These assessments are often ad hoc, late, or spatially imprecise, with reporting at the zonal or regional level. This is problematic as we find substantial variability in losses at the village-level, which is missing when reporting at the zonal level. In this paper, we propose a new data fusion method—combining remotely sensed data with agricultural survey data—that might address these limitations. We apply the method to Ethiopia, which is regularly hit by droughts and is a substantial recipient of ad hoc imported food aid. We then utilize remotely sensed data obtained near mid-season to predict substantial crop losses of greater than or equal to 25% due to drought at the village level for five primary cereal crops. We train machine learning models to predict the likelihood of losses and explore the most influential variables. On independent samples, the models identify substantial drought loss cases with up to 81% accuracy by mid- to late-September. We believe the proposed models could be used to help monitor and predict yields for disaster response teams and policy makers, particularly with further development of the models and integration of soon-to-be available high-resolution, remotely sensed data such as the Harmonized Landsat Sentinel (HLS) data set.

Mann, M.L., Warner, J.M. and Malik, A.S., 2018. Predicting high-magnitude, low-frequency crop losses using machine learning: An application to cereal crops in Ethiopia. Climatic Change, pp.1-17.