Physiological and UAV-based spectral responses of soybean genotypes under drought stress and their application in biomass prediction

• UAV-based imagery and physiology tracked drought effects in six soybean lines. • UAV-derived vegetation indices correlated with drought-induced physiological changes. • BU Soybean 2 showed highest drought tolerance; BU Soybean 1 was most sensitive. • Ridge model predicted biomass more accurately using spectral and physiological data. Soybean, a globally important crop, suffers significant yield losses under drought stress, especially during reproductive stages, limiting its expansion in regions like Bangladesh. This study evaluated drought tolerance in six soybean genotypes (BD 2330, BU Soybean 1–4, and G00056) by integrating morpho-physiological traits with UAV-based multispectral imaging. Physiological traits were measured from the uppermost fully expanded trifoliate leaves under standardized conditions. Spectral data were processed in Pix4D for radiometric calibration, orthomosaic generation, and vegetation index (VI) extraction. Under drought stress, genotypes exhibited average reductions of 12% in cell membrane stability and 13% in relative water content. Total chlorophyll and carotenoids declined by 32% and 35%, respectively, while proline increased by 43%, indicating active osmotic adjustment. Chlorophyll fluorescence ( Fv / Fm and Fv / Fo ) dropped by 14%, pod weight by 14–31%, and above-ground biomass by 11–28%. Drought stress also caused substantial variation in VIs across the growing season, revealing clear genotypic differences in canopy spectral responses. These physiological and biochemical traits were strongly correlated with UAV-derived VIs, confirming the effectiveness of remote sensing in detecting drought-induced stress. To translate these relationships into a practical application, multiple linear regression and ridge regression models were developed to predict above-ground biomass (AGB) using key physiological and spectral traits, achieving R² values of 0.76 and 0.92, respectively. Cluster analysis grouped genotypes based on stress responses, identifying BU Soybean 2 as the most drought-tolerant and BU Soybean 1 as the most sensitive. Overall, this study demonstrates that UAV-based canopy sensing, combined with physiological traits and interpretable machine learning models, offers a robust, high-throughput approach for screening drought-tolerant soybean genotypes. This integrated strategy supports breeding programs and precision agriculture aimed at enhancing soybean productivity under water-limited conditions.