Phenology-aware collaborative retrieval of structural and physiological cotton traits from Sentinel-2 imagery using...

Cotton leaf area index (LAI) and SPAD values characterize canopy structure and relative leaf chlorophyll status, respectively; however, most existing remote-sensing studies retrieve these two traits independently and pay limited attention to phenology-driven changes in their shared and target-specific spectral information. Using synchronous LAI and SPAD measurements collected from 109 sampling sites over eight observation periods in 2021, together with Sentinel-2 imagery, this study developed a phenology-dependent framework for the collaborative retrieval of LAI and SPAD. The maximal information coefficient (MIC) was used to identify sensitive spectral features for each observation period, after which three feature-input schemes and six regression algorithms were evaluated sequentially, with model performance evaluated using nested five-fold cross-validation grouped by sampling-site ID. The optimal model was subsequently applied to county-scale mapping and externally validated at the point scale using independent samples collected during T3, T4, and T6 in 2025. The results showed that (1) LAI, SPAD, and the composition of their sensitive spectral features exhibited pronounced phenological dependence, with red-edge features repeatedly selected across multiple periods, whereas features from other categories displayed stronger period specificity. (2) The combination of the period-specific union of sensitive features (MIC_union) and multi-task elastic net (MTE) achieved the best overall performance, yielding R² values of 0.91 and 0.79 for LAI and SPAD, respectively, and a joint normalized error of 0.08 in the outer-loop cross-validation for 2021. In the 2025 validation, T3 retained the highest cross-year explanatory capacity, with R² values of 0.61 and 0.53 and RMSE values of 0.23 and 3.04 for LAI and SPAD, respectively, whereas cross-year transferability was comparatively weak at T6. (3) Based on the period-specific evaluation for 2021 and the external validation for 2025, T3 was identified as the primary reliable window for collaborative LAI-SPAD mapping, with T4 serving as a supplementary window. This framework provides methodological support for the simultaneous remote-sensing characterization of cotton canopy structure and chlorophyll status during key growth stages, as well as for optimizing the timing of image acquisition and field surveys.