GF-1-Assisted Pseudolabel Augmentation and Leakage-Free Stacking for Sentinel-2 Chlorophyll-a Retrieval in Narrow Urban Rivers

Narrow urban rivers remain challenging for Sentinel-2-based chlorophyll-a (Chl-a) retrieval because mixed pixels, adjacency effects, and bank shadows can contaminate spectral signals, while in situ observations are often limited. This study develops an integrated workflow that integrates a high-confidence urban-river water mask with explicit shadow suppression to constrain feature extraction and sample selection, cross-resolution pseudo-label transfer by mapping a GF-1 Chl-a field to the Sentinel-2 10 m grid and stratifying it to form pseudo-labeled samples for augmentation, and a leakage-controlled out-of-fold stacking regressor with an XGBoost meta-learner to improve stability under data scarcity. On the in situ test set (n = 6), the GF-1 retrieval achieves R² = 0.86 and RMSE = 3.32 μg/L. Using mapped pseudo-labels for Sentinel-2 training, the model attains R² = 0.84 and RMSE = 3.88 μg/L on a held-out pseudo-label set (n = 40), while the in situ-only baseline yields R² = 0.70 and RMSE = 4.47 μg/L on the separate in situ test set. These results indicate that cross-resolution pseudo-label augmentation and leakage-controlled ensembling can be effective under limited field supervision; the pseudo-label evaluation reflects internal consistency rather than independent ground truth.