A Long-Term, High-Resolution SIF Reconstruction for Poyang Lake

Using MODIS surface reflectance products, we first generated continuous daily 250 m water/land classifications through the Spatio-Temporal Neighborhood Similarity Gap-Filling (STNS-GF) method, and reconstructed separate NDVI time series for water and land pixels using the Harmonic Analysis of Time Series (HANTS) algorithm.

Daily water surface coverage (WSC) at 500 m resolution was subsequently derived from these classifications.An XGBoost machine learning model was trained to reconstruct SIF at 500 m spatial resolution from 2000 to 2023, integrating OCO-2/OCO-3 SIF retrievals with daily WSC, NDVI, and ERA5-Land meteorological variables. Cross-validation demonstrated robust performance (validation RMSE = 0.137 mW·m⁻²·nm⁻¹·sr⁻¹; MAE = 0.104 mW·m⁻²·nm⁻¹·sr⁻¹).  The dataset comprises three core variables:Water Surface Coverage (WSC): daily fractional water coverage at 500 m.NDVI: daily normalized difference vegetation index for water and land pixels, gap-filled and smoothed.SIF: daily solar-induced chlorophyll fluorescence （mW·m⁻²·nm⁻¹·sr⁻¹) at 500 m, representing terrestrial and aquatic photosynthetic activity.  These data support research on wetland carbon cycling, eco-hydrological modeling, drought impact assessment, and validation of satellite-based photosynthesis products, and provide a critical basis for adaptive management under increasing climate extremes.Temporal Coverage: 2000–02–24 to 2023–12–31Spatial Resolution: 500 m