Interactive effects of geo-ecological factors on soil respiration and its temperature sensitivity on the global scale

Abstract Soil respiration is one of the main processes in the global carbon cycle, which comprises both autotrophic (root) and heterotrophic (soil organic matter decomposition) components. However, large uncertainties remain in estimating its flux ( R s ) and its temperature sensitivity ( Q 10 ) globally, and interactive effects from soil properties are often overlooked in models. To better predict global R s and Q 10 , we applied four nonlinear models (Random forest (RF), AdaBoosted, Bagged Tree, and XGBoost models) to 3325 observations of R s and Q 10 from 1971 to 2023 with 18 predictors and used SHapley Additive exPlanations (SHAP) analysis to explore the interactive effects. The RF model provided the best performance, with the highest R 2 values and minimal root mean squared errors. The results indicated that the global average annual R s was 98.4 ± 19.4 Pg C. Excluding the effects of soil variables (soil texture (Stex), microbial composition, and solution chemistry) led to a decline in R 2 (from 0.79 to 0.7 for R s models and from 0.40 to 0.34 for Q 10 models) and an underestimation on R s . The results of rotated principal component analysis showed that these soil variables collectively explained 50% of the cumulative variance in R s variability. Soil variables significantly constrained Q 10 values, especially in nutrient-poor soils. SHAP analysis showed that the interactive effects of high soil sand proportion and high soil fertility may suppress R s . High soil evapotranspiration may also decrease R s at the extreme conditions of Stex. These findings highlight the underappreciated mechanistic influence of soil properties in Earth system models and underscore the need to incorporate multivariate interactions when modeling future soil respiration and its feedbacks to global change.