Geospatial artificial intelligence for climate resilient agricultural landscape management in Nigeria: a machine learning approach for sustainable land use optimisation

Abstract

Nigerian agricultural landscapes sustain over 70 percent of the rural population yet face mounting pressures from climate variability, degradation processes, and inadequate spatial decision support. Traditional landscape management frameworks cannot capture the nonlinear dynamics operating across the six Nigerian agro-ecological zones at scales relevant to smallholder farming systems. This
research presents GEOCLIMA, a geospatial artificial intelligence framework integrating advanced machine learning with multisource satellite remote sensing to deliver climate resilient agricultural landscape management. A Convolutional Neural Network and Long Short Term Memory hybrid model achieved 93.4 percent overall accuracy with kappa coefficient 0.91 for land use change detection, surpassing standalone CNN performance at 91.7 percent and LSTM results at 87.3 percent. Multi Agent Deep Deterministic Policy Gradient reinforcement learning produced 23.4 percent
improvement in irrigation water use efficiency validated across three pilot sites spanning Guinea, Sudan, and Sahel savanna zones. The climate smart landscape classification module implemented as stacked ensemble combining Random Forest, Support Vector Machine, and Gradient Boosting classifiers identified 127,340 hectares exhibiting very high agricultural vulnerability requiring immediate intervention from total study coverage encompassing Nigerian primary agricultural zones. Field validation conducted across three pilot locations at Lokoja, Kaduna, and Maiduguri during 2023 to 2024 growing seasons yielded crop yield prediction with coefficient of determination 0.84 and root mean square error 287 kilograms per hectare. These findings establish that geospatial artificial intelligence can generate operationally meaningful intelligence for climate resilient agricultural
landscape governance while providing replicable architecture for smallholder decision support across West African agricultural systems.