Wheat-WSI: Development and estimation of a seedling-stage waterlogging stress index using multimodal image features

Waterlogging stress during the seedling stage significantly restricts wheat growth and biomass accumulation, posing a major constraint on yield formation. While multimodal sensing technologies are increasingly applied to crop stress monitoring, challenges remain in achieving timely and accurate estimation under short-term stress conditions. This study addresses the need for a rapid, non-destructive, and quantifiable method to monitor wheat waterlogging stress. A composite index, Wheat-WSI (Waterlogging Stress Index), was developed using principal component analysis (PCA) based on six key physiological and agronomic parameters. Two estimation models were constructed: Wheat-WSI (HSM), a hyperspectral single-modality model using four spectral indices; and Wheat-WSI (MFM), a multimodal fusion model incorporating eight features derived from hyperspectral, thermal infrared, and chlorophyll fluorescence images. Both models employed the random forest (RF) algorithm, with feature selection and performance evaluation conducted using SHAP values, Pearson correlation, feature importance scores, and incremental modeling curves. Wheat-WSI (MFM) showed consistent estimation performance across treatments, with optimal results under 9-day stress (R2 = 0.92, CCC = 0.96) and reduced accuracy under 3-day stress (R2 = 0.75, CCC = 0.82). Wheat-WSI (HSM), despite relying solely on spectral indices, achieved reasonable accuracy (R2 = 0.78, NRMSE = 0.16). Field validation demonstrated the robustness and application potential of Wheat-WSI (MFM) under prolonged stress, though its early detection capability under short-term conditions requires further enhancement. The proposed Wheat-WSI enables effective quantification of seedling-stage waterlogging stress and provides a robust modeling framework for monitoring crop stress responses.