RSIF: A 0.005° Global SIF Dataset Based on an End-to-End Convolutional Neural Network with Spatial Redistribution

To improve the spatial resolution and preserve the spatial fidelity of TROPOMI Solar-Induced chlorophyll Fluorescence (SIF), we developed a global 0.005° SIF product (RSIF) covering May 2018 to December 2020 using the proposed One-Step Learned Spatial Redistribution Convolutional Neural Network (OSRNet). OSRNet directly redistributes coarse-resolution TROPOMI SIF into fine grids using high-resolution auxiliary variables, including MODIS reflectance, ERA5 reanalysis, GEBCO DEM, and cos(SZA). Validation against both original TROPOMI SIF and long-term tower-based SIF from five flux sites shows that RSIF improves agreement with coarse-resolution inputs and better captures fine-scale spatial details compared to traditional downscaling methods. RSIF represents clear-sky SIF and if needed, it can be readily converted to all-sky SIF using established correction factors.

This record serves as the parent dataset for the RSIF product covering the years 2018, 2019, and 2020. The complete dataset is split into three annual subsets due to Zenodo’s file size limitations:

• 2018: doi:10.5281/zenodo.16787900

• 2019: doi:10.5281/zenodo.16788564

• 2020: doi:10.5281/zenodo.16788879

Users are encouraged to cite this parent record when referencing the complete dataset, and to cite individual year records when using specific annual subsets.