Construction and Validation of a Predictive Model for Desert Locust Breeding Site Selection in Nigeria’s Sahel Zone Using Environmental Remote Sensing and Machine Learning

Desert locust outbreaks are notoriously difficult to predict, and early detection of breeding sites before populations reach gregarious phase is the most effective intervention point. Machine learning models trained on satellite-derived environmental data offer a promising approach to predicting locust breeding site locations, but such models do not yet exist for Nigeria's Sahel zone. This study constructs and validates a machine learning model for predicting desert locust breeding site selection in Borno, Yobe, Katsina, Sokoto, and Kebbi states. A comprehensive retrospective dataset of locust egg-pod and hopper sightings from state agriculture reports, FAO records, and field surveys will be compiled as the response variable. Predictor variables derived from satellite data will include vegetation greenness, soil moisture, rainfall anomalies, surface temperature, and land cover class, all extracted at 500-meter resolution. Random forest, gradient boosting, and convolutional neural network models will be trained on two-thirds of the dataset and validated on the remaining third. Model performance will be compared to expert rule-based forecasting systems. Validated model outputs will be calibrated for operational integration into the FAO early warning locust forecasting system. This study makes a pioneering original contribution by developing the first Nigeria-specific machine learning tool for desert locust outbreak prediction. Keywords: desert locust, machine learning, remote sensing, breeding site prediction, Sahel Nigeria