A global gross primary productivity of sunlit and shaded canopies dataset from 2002 to 2020 via embedding random forest into two-leaf light use efficiency model

Gross primary productivity (GPP) is crucial for understanding the carbon cycle and maintaining ecosystem balance under climate change. We attempt to generate a long-term global dataset for GPP of sunlit (GPP(su)) and shaded leaves (GPP(sh)) by a hybrid model combining the random forest (RF) submodule with the two-leaf light use efficiency (TL-LUE) model. First, the TL-LUE model was optimized by considering the seasonal differences in the clumping index on a global scale (TL-CLUE). Then, we used the RF technique to integrate various environmental stress factors, including meteorological factors, hydrological variables, soil properties, and elevation, which originate from the NASA MERRA-2 dataset, ISRIC soil Grids, and USGS data center. Furthermore, the RF submodule was embedded into the TL-CLUE model to construct the hybrid model (TL-CRF), which was trained and evaluated based on global eddy covariance (EC) site data from the AmeriFlux and FLUXNET2015 datasets. We produced a global GPP, GPP(su), and GPP(sh) dataset with a spatial resolution of 0.05 x 0.05 degrees over 2002-2020 by the TL-CRF model driven by the LP DACC leaf area index and land cover, NASA MERRA-2 incoming shortwave solar radiation, and the above environmental variables. This GPP product provides a data basis for improving our understanding of the dynamics of global vegetation productivity and its interactions with the changes in environmental conditions.