High-resolution monitoring of intra-seasonal agricultural drought using sentinel-2 and machine learning across bimodal growing seasons in Kenya

Drought presents significant challenges to agriculture, threatening food security and livelihoods, across many regions. In Kenya, recurrent droughts across diverse agro-ecological zones emphasize the urgent need for reliable and scalable drought assessment methods. Although drought assessment with various datasets has been carried out for this region, many of them often use course or moderate resolution data. This study uses high-resolution Sentinel-2 observations and machine learning to monitor intra-seasonal crop conditions and assess drought impacts across bimodal growing seasons. Using pixel-based supervised random forest models trained with multiple vegetation indices as input, we classify croplands into drought-affected and unaffected areas. The phenology information is integrated into the framework to take into account two growing seasons. Our results from 2016 to 2022 highlight an increasing frequency in drought-affected croplands, with 2022 exhibiting widespread crop stress, especially during the long rains season. The results have strong agreement to governmental yield statistics and national and international reports as well as regional and global drought monitoring platforms. Despite uncertainties from cloud cover, land cover misclassification, and other stressors, the proposed framework provides high-resolution and operationally relevant agricultural drought monitoring approach with high resolution at national scale. Future enhancements, including radar-based data integration and real-time monitoring and prediction, can further strengthen early warning systems, supporting policymakers in developing targeted drought mitigation strategies. • Sentinel-2 and random forest enable field-scale agricultural drought impact monitoring in Kenya. • Phenology-aware modeling captures bimodal seasons and intra-seasonal crop stress. • Increasing drought impacts detected, with strong agreement to yield and drought reports.