Integration of open-source solutions with deep learning for estimating crop production in data-scarce smallholder farming areas

In smallholder farming areas where crop production is the mainstay of livelihood, early identification and monitoring of crop production status will be very useful for assessing the forthcoming food availability, food security and food market stability of the region at a preharvest season. In the presence of climate related shocks, this information would be even necessary for estimation of damage and further insurance pay-outs. Collection of this information through manual approaches is time consuming, influenced by human and technical bias and mostly impractical because of inaccessible terrain, resources and time. Presence of wide array of earth observing satellite has provided possibility of monitoring and mapping of objects and phenomena everywhere in the world. Though this is the general trend, mapping of crops and crop production using conventional approaches is challenging which is constrained by inherent characteristics of smallholder farming areas like seasonality of crops, fragmented small fields and dominance in complex topography. Therefore, in our proposed project we have planned to integrate optical and radar satellite imagery with artificial intelligence (deep learning models) with powerful statistical tools to map crop types and crop production in smallholder farming areas, in selected Ethiopian landscape using open source innovative solutions.