Assessing and Predicting Ecological Quality Using a Novel Indicator and Machine Learning Framework: A Case Study of Beijing

Ecological quality is a comprehensive concept that reflects the overall condition of an ecosystem. Understanding its dynamic evolution and future trajectories is essential for sustainable ecosystem management. However, existing studies often rely on single-dimensional perspectives and overlook ecosystem complexity, nonlinear relationships, and interactions among drivers. Therefore, this study developed a novel ecological quality indicator based on a “function–status–disturbance” framework integrating ecosystem services, the Remote Sensing Ecological Index, the Modified Water Area Remote Sensing Ecological Index, and nighttime light intensity. We quantified the spatiotemporal evolution of ecological quality in Beijing from 2000 to 2020. Correlation analysis was used to select ten significant environmental and socioeconomic drivers from eleven candidate variables. A Random Forest model incorporating the ten drivers predicted ecological quality under three representative SSP–RCP scenarios for 2035. Permutation importance based on the percentage increase in mean squared error and partial dependence analysis were used to quantify the relative importance, nonlinear responses, and threshold effects of drivers. The results showed that (1) ecological quality in Beijing exhibited a three-tier concentric gradient pattern, increasing from the urban center to the northwestern mountains. (2) Ecological quality declined overall from 2000 to 2020, particularly during the first decade. (3) The Random Forest model showed high predictive accuracy, with an R 2 of 0.979, an MAE of 0.029 and an RMSE of 0.041. Among the three scenarios in 2035, the ecological protection scenario exhibited the highest ecological quality, followed by the middle path and baseline scenarios, but all remained below the 2020 level. (4) Ecosystem responses varied substantially across scenarios. Natural ecosystems performed best under the ecological protection scenario, whereas cultivated land showed the opposite pattern, with the baseline scenario achieving the highest quality. (5) Annual precipitation and distance to regional centers were the most influential drivers, and most factors exhibited nonlinear and threshold effects. These findings provide scientific support for adaptive ecosystem management and sustainable urban planning.