The first gap-free 20 m 5-day LAI/FAPAR products over China (2018–2023) from integrated Landsat-8/9 and Sentinel-2 Analysis Ready Data

Leaf Area Index (LAI) and Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) are essential land variables for environmental monitoring and climate modeling. High resolution (≤30 m) gap-free LAI/FAPAR products are in high demand, but frequent cloud contaminations in optical data cause substantial data gaps. To address the ill-posed nature of land surface variable inversion by leveraging time-series information instead of traditional pixel-based inversions, this study presents a temporal deep learning model that jointly estimates gap-free, 20 m/5-day LAI/FAPAR from integrated Landsat-8/9 and Sentinel-2 sequential observations, denoted as High-resolution Global LAnd Surface Satellite (Hi-GLASS) LS20 LAI/FAPAR products, part of the Hi-GLASS level 3 product suite. A hybrid Bidirectional LSTM with an attention mechanism that synergizes multiple satellite observations effectively under different cloud cover conditions was trained on representative samples derived from GLASS LAI/FAPAR and 30 m land cover data, accounting for site heterogeneity. The algorithm was directly validated against 4046 in-situ measurements from 29 validation sites, achieving an R² of 0.79 for LAI and 0.86 for FAPAR, Root Mean Square Error (RMSE) of 1.0 for LAI and 0.155 for FAPAR. Intercomparisons with existing high and coarse resolution products showed superior continuity and accuracy. To implement the model, we constructed Landsat and Sentinel-2 Analysis Ready Data (LSARD) and generated the first 20 m gap-free LAI/FAPAR product over China from 2018 to 2023 (www.glasss.hku.hk). We also provide a web tool on Google Colab that can calculate LAI/FAPAR for any region of interest. Unlike methods that rely solely on clear-sky pixels from a single sensor, our approach enables spatiotemporally continuous and physically consistent LAI/FAPAR estimates from multiple sensors.