A long-term (2000-2022), high-resolution (0.005°) aboveground biomass dataset of global grasslands

Grasslands are critical to the global carbon cycle and support livestock production, yet long-term, high-resolution global datasets of grassland aboveground biomass (AGB) remain scarce, since existing products are often spatially coarse, temporally discontinuous, or regionally limited. Here, we present a global grassland AGB dataset spanning 2000-2022 at 0.005° spatial resolution. The dataset was generated by employing an optimized random forest model, which was trained on 11,265 spatially aggregated samples with global distribution. The multi-source predictors included Normalized Difference Vegetation Index (NDVI), climate variables, atmospheric CO₂ concentration, and topographical factors. At the global scale, the model achieved an R² of 0.60 for both validation and independent testing, with RMSEs of 85.97 and 93.69 g m^-2, respectively. When evaluated across all samples, the model exhibited an R² of 0.93 and an RMSE of 51.85 g m^-2. Comparison with other datasets showed its comparable value distributions and interannual trajectories, and closer agreement with field observations. This dataset provides a foundation for advancing research on global grassland carbon dynamics and management practices.