A Global Soil Dataset for Earth System Modeling (Version 2, GSDEv2)

Abstract. Soil properties are essential inputs to Earth system models. However, existing global soil datasets vary significantly in property coverage, spatial and vertical resolution, uncertainty characterization, and the integration of regional soil information. Here, we present the Global Soil Dataset for Earth System Modeling version 2 (GSDEv2), a model-oriented global dataset developed within a framework that synergizes observation-informed spatial prediction, soil-domain-specific modelling, and stacking-based incorporation of regional soil products with a globally consistent baseline. GSDEv2 provides predictions of 20 physical and chemical soil properties for six standard depth intervals on a nominal 90 m grid, together with spatially aggregated 1 km mean and conditional-quantile products. Compatible regional soil products for China, the United States, Africa, Canada, and Australia were incorporated as auxiliary predictors through stacking-based fusion. Global baseline performance was evaluated using a profile-level random hold-out design with 90 % of profiles used for training and 10 % for testing, characterizing interpolation performance within the sampled observational domain. The incremental value of regional fusion was evaluated for surface soil organic carbon in China using nested 60 km spatial-block validation. Among 54 property–depth combinations jointly supported by GSDEv2, GSDEv1, SoilGrids 2.0, and HWSD 2.0, GSDEv2 showed the highest mean modelling efficiency coefficient and the lowest mean root-mean-square error across the evaluated combinations. Median RMSE reductions relative to GSDEv1, SoilGrids 2.0, and HWSD 2.0 were 22.2 %, 5.8 %, and 24.6 %, respectively. In the spatially blocked China case study, regional stacking reduced RMSE by 31.8 % and the mean 90 % prediction-interval width by 38.7 % while maintaining approximately nominal coverage. In a downstream CoLM2024 sensitivity experiment, the GSDEv2-driven simulation reduced the annual-mean spatial RMSE of surface soil moisture from 0.109 to 0.089 m³ m⁻³ relative to the SoilGrids-driven experiment and increased the spatial correlation with adjusted ERA5-Land from 0.741 to 0.802. GSDEv2 thus provides a vertically harmonized and spatially detailed soil resource for regional-to-global Earth system applications. The GSDEv2 dataset is freely available at https://doi.org/10.11888/Terre.tpdc.303613 or https://www.scidb.cn/s/my2aea (Shi et al., 2026).