Developing a Remote Sensing-Based Early Warning System for Agricultural Food Security Crises in the Sahel Zone of North East Nigeria

Food security early warning systems in North East Nigeria operate under severe data constraints, compounded by conflict-induced field access limitations and the collapse of institutional data collection infrastructure in affected areas, creating a critical gap that remote sensing-based approaches are uniquely positioned to address. This dissertation develops an original remote sensing-based early warning system for agricultural food security crises in the Sahel zone of North East Nigeria, covering Borno, Yobe, and the northern districts of Adamawa State. The system architecture integrates MODIS and Sentinel-2 vegetation phenology indicators, TAMSAT and CHIRPS satellite rainfall estimates, MODIS land surface temperature anomalies, and soil moisture products from the ESA CCI Soil Moisture archive within a machine learning-based multi-indicator fusion framework. An original Conflict-Adjusted Crop Loss Index is developed to separately attribute vegetation anomalies to climatic drivers and conflict-induced cultivation abandonment, based on ACLED event density weighting of pixel-level productivity deviations. The early warning system is trained and calibrated against IPC food insecurity classification data from 2016 to 2023 using gradient boosting classification with leave-one-year-out cross validation. Classification performance yields a macro-average F1 score of 0.83 for predicting IPC Phase 3 and above conditions with a four-month lead time. Operational prototype deployment is conducted in collaboration with the Borno State Emergency Management Agency over a six-month evaluation period. The dissertation makes original contributions in conflict-adjusted remote sensing methodology and early warning system architecture for conflict-affected dryland environments. Keywords: early warning system, food security, remote sensing, North East Nigeria, Sahel.