Nonlinear climate–ecosystem functional interactions shape aboveground forest biomass dynamics in the Qilian Mountains: Insights from remote sensing and explainable machine learning

Forest aboveground biomass (AGB) is an essential indicator of carbon storage and productivity in forest ecosystems. Its temporal dynamics reflect how ecosystems respond and adapt to climate change. Using Tianzhu County in the eastern section of the Qilian Mountains as the study area, this study integrated multi-source remote sensing data and a suite of multi-dimensional ecological variables to estimate forest AGB through multiple machine learning algorithms. The Shapley Additive Explanations (SHAP) framework was employed to quantify the nonlinear and lagged effects of climatic drivers. The main findings are as follows: (1) A total of 42 variables were analyzed. After feature selection with the Least Absolute Shrinkage and Selection Operator (Lasso) and modeling with the Extreme Gradient Boosting (XGBoost) algorithm, the optimal model achieved a test-set R 2 =0.68 and RMSE ≈15.01 Mg/ ha, demonstrating robust predictive accuracy and good interpretability. (2) From 2009 to 2023, forest biomass exhibited an overall increasing trend, with a mean annual growth rate of 0.0246 Mg/ ha. Regions with significant increases accounted for 27.21%, indicating continuous ecological recovery and improvement of stand structure. (3) Linear correlations were generally weak: precipitation showed a slight positive relationship with AGB (r = 0.045), whereas the effect of temperature was near zero and directionally unstable. Fewer than 10% of pixels were significant, and all turned insignificant after false discovery rate (FDR) correction. (4) Nonlinear analysis revealed distinct threshold characteristics, with biomass peaking when annual precipitation ranged from 330 to 460 mm and temperature from –2 °C to 2 °C. SHAP results further verified this combined regulatory effect. (5) The interaction between temperature and precipitation dominated the spatial variation of biomass. Moderate precipitation under cool conditions promoted growth, whereas high temperatures or excessive rainfall induced ecological stress. Sensitivity analysis confirmed the robustness of the model. (6) Mean AGB among the eight climatic zones was relatively similar (∼80 Mg ha -1 ). The moderate-temperature and low-precipitation zone had slightly higher biomass, while low-temperature zones did not exhibit disadvantages, reflecting the cold adaptation of dominant species and the mitigating influence of human interventions. (7) Lag analysis showed that the cumulative effect of precipitation (R 2 = 0.794) was much stronger than that of temperature (R 2 = 0.028). Precipitation with 1–2 year lags promoted growth, whereas medium- and long-term lags showed alternating positive and negative effects. This study systematically characterized the climatic response mechanisms of forest AGB from spatial, nonlinear, and temporal perspectives, contributing valuable insights for regional carbon-stock monitoring, climate-adaptive forest management, and ecological restoration planning. • Aboveground biomass (AGB) is used as a structural plant functional trait. • Remote sensing and machine learning (XGBoost + Lasso) are combined to estimate biomass. • SHAP analysis quantifies nonlinear interactions between climate variables. • Temperature–precipitation interactions contribute significantly to AGB variation, with observable threshold patterns. • Spatial heterogeneity of AGB is shaped by mid-temperature and low-precipitation regimes. • Lagged effects highlight trait persistence and temporal heterogeneity in climate response.