Terrestrial ecosystems can exhibit various behaviors in response to climate change and human activities. Nonlinear and abrupt shifts in ecosystems are particularly important as they indicate substantial modifications in ecosystem structure and function, posing a threat to the provision of ecosystem services. Here we distinguish between linear, curvilinear, and abrupt shifts in ecosystem productivity from 2000 to 2020 in China's Loess Plateau. We utilize spatial Random Forest models to analyze the driving factors behind these change patterns. Our findings indicate that 84.1% of the ecosystems experienced a positive change in plant productivity, while a small proportion (2.5%) exhibited a negative change. Abrupt shifts are prevalent in both positive and negative changes in ecosystem productivity, with negative changes often manifesting as abrupt shifts (79.3%). Negative changes in plant productivity, particularly the negative abrupt shifts, are primarily associated with human activities characterized by increased nighttime light and urbanization. Land conversion to forest is linked to a curvilinear trajectory in plant productivity, characterized by nonlinear changes with acceleration. Higher water availability and a wetter environment are more likely to promote positive changes in plant productivity. Moderate warming trends contribute to abrupt positive changes in plant productivity, while high warming trends are associated with increased negative abrupt and curvilinear changes. We highlight the importance of accounting for diverse change behaviors in ecosystems for the development of targeted conservation and restoration measures.

Rapid urbanization and intensification of human activities increases the risk of disturbance of ecological systems via multiple sources, with consequences for regional ecological security and health. Landscape ecological risk assessment (LERA) is an effective way to identify and allocate risk to resources. We used the north and south Qinling Mountain area as a case study to analyze the spatial heterogeneity of landscape ecological risk using a potential- connectedness-resilience three-dimensional (PCR 3D) framework based on an integrated and dynamic risk assessment concept from adaptive cycle theory. We explored factors driving the risks with a spatial model GeoDetector. The results show that the comprehensive landscape ecological risk was north–south polarized and dominated by low and moderate risk levels (90.13% of total risk) across the whole study area. The high-risk area was centered on the Weihe plain north of the Qinling Mountains (NQL), while low-risk areas accounted for 86.87% of the total area and were prevalent across the south of the study area. The areas with high potential and connectedness risks were centered in the Xi’an–Xianyang urban agglomeration and those with high-resilience risk were in the upper reaches of the Hanjiang River. The vast majority of the area to the south of the Qinling Mountains (SQL) is at low risk. In terms of driving forces, population density and vegetation coverage (NDVI) are the primary factors affecting landscape ecological risk. Our findings suggest that anthropogenic activity is the primary cause of landscape ecological risks in the study area and regional socioeconomic exploitation and environmental conservation need to be rebalanced to achieve sustainability for the social ecosystem. The PCR 3D LERA framework employed in this study can be used to inform landscape ecological health and security and to optimize socioeconomic progress at regional scales.