<![CDATA[Desert locust infestations continue to threaten agricultural productivity, food security, and environmental sustainability across East Africa. Conventional locust surveillance approaches remain largely dependent on field-based monitoring systems that are reactive, labor-intensive, and constrained by limited spatial coverage. This study investigated the hyperspectral signatures and environmental variables associated with locust breeding and feeding zones in South Nyanza, Kenya, using an Artificial Intelligence (AI)-driven hyperspectral remote sensing framework. Environmental variables including vegetation indices, soil moisture, rainfall, temperature, vegetation stress indicators, and hyperspectral bands were analyzed using correlation analysis, binary logistic regression, Random Forest classification, and variable importance ranking. Correlation analysis revealed weak relationships between individual environmental indicators and historical locust occurrence, with all predictors demonstrating statistically insignificant relationships (p > 0.05). Logistic regression findings similarly indicated that no individual environmental variable independently predicted locust occurrence. However, Random Forest classification achieved a classification accuracy of 75.33% with an Out-of-Bag error rate of 24.67%, suggesting moderate predictive capability when multiple variables were integrated. Variable importance analysis identified rainfall, soil moisture, temperature, and vegetation-related indicators as dominant predictors. The findings suggest that locust breeding and feeding habitats are influenced by complex multidimensional interactions rather than isolated environmental indicators. The study demonstrates the practical potential of AI-driven hyperspectral frameworks for strengthening locust surveillance and early warning systems within emerging high-risk agricultural regions.]]>
