Spatiotemporal Distribution and Uncertainty of Soil C:N Ratios in Global 0-30cm and 30-100cm Soil Layers

We collected 36997 soil C:N data points for the 0-30 cm layer and 31691 data points for the 30-100 cm layer, integrating a total of 24 environmental covariates from five categories closely related to soil C:N, including climate, vegetation, topography, soil properties, and human activities. This study constructed a two-stage modeling framework. The modeling strategy in the first stage showed that the direct method performed better than the indirect method. In the second stage, a spatial zonal strategy was introduced. By comparing ecological zonal and geographic zonal, the one-step ecological zonal strategy was finally determined as the optimal strategy. Based on the above strategy, a prediction model was constructed by using direct modeling with ecological zonal combined with the quantile random forest method. The independent validation dataset results demonstrated that the direct modeling strategy with ecological zonal achieved an R² of 0.44 for C:N030 and 0.30 for C:N30100. Based on this model, the spatial distribution and uncertainty of global soil C:N for the 0-30 cm and 30-100 cm layers were mapped at a 1 km resolution. Comparative analysis indicates that the 1 km global soil C:N dataset produced in this study exhibits higher accuracy than existing products.