Response of SOC stocks in Northeast China to climate warming and precipitation changes

• Introducing a Bayesian XGBoost framework as a new paradigm for SOC studies. • High-resolution (90-m) SOC time-series mapping from 1985 to 2020. • Unveiled a temperature-triggered cascade and precipitation buffering mechanism. • Gradient experiments quantify non-linear climate change effects on SOC. • Ecosystem-dependent sensitivity and buffering capacity to climate are quantified. Changes in temperature and precipitation strongly affect soil organic carbon (SOC) stock dynamics, which in turn affects the climate feedback process, but the mechanism of the compound effects remains unclear. In this study, we integrated the spatio-temporal substitution method and meta-analytical framework, used a Bayesian optimized XGBoost model to predict the spatio-temporal dynamics of SOC with high accuracy, combined structural equation modeling and partial dependency plots to analyze the driving mechanisms, and simulated the future scenarios under the gradient changes of temperature (TAVG: 0–4 °C) and precipitation (PR: −50 % to + 50 %). The results show that: (1) the regional SOC stock fluctuated in a “decrease-increase–decrease” phase between 1985 and 2020, with a net loss of 5.38 Mg C/ ha, and the area of the carbon loss zone is nearly twice as large as the area of the positive change; (2) climatic factors contributed to 50 % of the SOC variability, with TAVG as the core negative driver and there is a critical threshold of −1 °C: below the threshold, low temperatures inhibit microbial activity and promote carbon accumulation, above the threshold, cascading degradation is triggered (warming → degradation of parent material → weakened carbon sequestration by vegetation → reduction of carbon inputs) to amplify the SOC loss, and PR mitigates the loss through multi-pathway synergistic effects; (3) warming dominates the systematic decline of SOC in future scenarios (the loss increases significantly for every 1 °C warming), and the synergistic effect of high temperature and drought increases the SOC loss (TAVG = 4 °C + PR = −50 % results in 26.89 % loss), and the compensatory effect of increased precipitation diminishes with warming (PR + 50 % at TAVG = 4 °C only increases SOC by 0.70 %); (4) ecosystem response shows a gradient: cropland is the most sensitive (28.83 % loss from warming and drying), forests are the most stable (reliant on litter cover buffer), and wetlands were highly dependent on water (PR + 50 % offset the negative effect of 4 °C warming). The study emphasizes the need to fully consider the combined effects of temperature and precipitation in assessing the impacts of future climate change on soil carbon pools, especially under extreme climate scenarios.