Artificial intelligence-based identification of drought-resilient bean cultivars using multi-source data in the Delmarva region

• A multi-year AI framework identifies drought-resilient bean cultivars under variable climate. • Integrating agronomic and environmental data improves yield prediction across Delmarva seasons. • Random Forest and XGBoost capture nonlinear genotype–environment yield interactions. • Heat stress and precipitation emerge as dominant drivers of bean yield variability. • Framework supports climate-smart, cultivar-specific selection for resilient bean production. Beans play a vital role in food and nutritional security while enhancing soil fertility; however, their productivity in the US Delmarva region is frequently negatively affected by recurrent droughts and irregular rainfall patterns. This study aims to identify drought-resilient bean cultivars by integrating multi-year field data with artificial intelligence (AI). Drought resilience was evaluated across four bean genotypes and varieties comprising eleven cultivars using multi-year field data (2014–2020 and 2023) under natural weather variability. Agronomic traits related to growth and yield were evaluated alongside drought indices and seasonal temperature and precipitation. Multi-year agronomic and environmental data were integrated into an AI workflow using Random Forest (RF) and Extreme Gradient Boosting (XGBoost), with repeated k-fold cross-validation, to predict yield and identify drought-resilient bean cultivars. Yield varied widely across seasons (93.5–6279.9 kg ha −1 ), reflecting substantial interannual climate variability, with drought indices indicating pronounced stress in dry years and reduced sensitivity in wetter seasons. Correlation analysis indicated generally weak relationships; pods, seeds, and moderate biomass were more consistently related to yield, while the heat stress index (HSI) showed strong collinearity (r = 0.88). Feature-importance analyses highlighted genotype and climate drivers as dominant predictors of drought resilience. Year-wise model fits were excellent in 2017–2019 and 2023 (RF R 2 ≤ 0.98; XGBoost ≤ 0.96) but weaker in 2014–2015. These findings demonstrate that integrating multi-year field data with AI enables robust identification of drought-resilient bean cultivars and supports data-driven cultivar selection for climate-resilient bean production in the mid-Atlantic region.