Deconstructing ecosystem disturbance: Driving mechanisms across frequency, intensity, and cumulative dimensions

Addressing the challenge that ecosystem vulnerability assessments often rely on static assessments and fail to distinguish between pulse and press disturbances, a multi-dimensional disturbance analysis framework based on Frequency-Intensity-Cumulative (FIC) effects was developed. By integrating XGBoost ensemble learning with SHAP interpretability models, we decouple the non-linear drivers of ecosystem disturbances in the Hohhot-Baotou-Ordos-Yulin (HBOY) urban agglomeration. Results indicate that human activities manifest as persistent spatial expansion and high-frequency chronic stress, contrasting with the pulse-like fluctuations of climatic factors and the cumulative recovery driven by ecological factors. Key quantitative thresholds define the boundaries of ecosystem stability: NDVI below 0.1 marks a critical survival baseline where vulnerability sharply increases, while cumulative grassland cover above 96.0% indicates enhanced structural stability. Crucially, when the frequency of nighttime light exceeds 6.1, the system reaches a fatigue limit, compressing recovery windows and exacerbating the risk of persistent degradation. Climate regulation exhibits a bistable pattern, where synergistic buffering effects emerge when cumulative temperature exceeds 3.4°C and precipitation intensity surpasses 34.4 mm. Furthermore, zonal attribution distinguishes differentiated control logics: vulnerable zones, driven by the coupling of high-frequency anthropogenic disturbance and intense climatic stress, require frequency controls and hard constraints. Conversely, improving zones, dominated by cumulative vegetation recovery, necessitate protective management. Overall, the FIC framework bridges statistical patterns with ecological mechanisms and planning rules, providing threshold-based scientific evidence for sustainable human-land governance.