Spatio-temporal evolution and prediction of ecological quality in ecological protection redline areas using a machine learning algorithm: a case study of the Poyang lake basin

This study aims to investigate the long-term and effective protection mechanisms for ecological protection redline areas, which are crucial for maintaining ecological security patterns, ensuring ecosystem functionality, and supporting sustainable socioeconomic development.

The research took the ecological protection redline areas in the Poyang lake basin as its study object. It constructed a dynamic ecological quality monitoring system based on the "matrix-stress-wildness" framework by integrating multi-source remote sensing data and ground monitoring data. The study introduced the XGBoost machine learning algorithm to adaptively adjust weight thresholds and establish a modified ecological quality index. Furthermore, it employed Theil-Sen trend analysis, Mann-Kendall test, and long short-term memory (LSTM) neural network to systematically reveal and predict the spatiotemporal evolution patterns and future development trends of ecological quality within the ecological protection redline areas of the Poyang lake basin.

The ecological quality within the ecological protection redline areas of the Poyang lake basin showed distinct phased characteristics temporally. During the rapid change period, spatial heterogeneity of ecological quality significantly intensified, with widening disparities among different land use types. In the stable adjustment period, the overall fluctuation range of ecological quality tended to moderate, with restorative improvements observed in some areas, reflecting the positive effects of ecological protection policies. Spatially, regions with lower ecological quality are mainly concentrated in the northwestern mountainous and hilly areas, the central basins characterized by concentrated agriculture and urbanization, and the northern lake areas affected by significant human activities. In contrast, areas with higher ecological quality are primarily distributed in the southern part of the basin, which features high natural vegetation coverage and minimal human disturbance. LSTM model predictions indicated that future ecological quality is primarily classified as "poor" and "medium," with "good" quality areas covering only 10 600 km² and concentrated in the Ganjiang river source reserve. Typical ecologically vulnerable areas continue to face multiple anthropogenic and natural pressures.

The monitoring and assessment system established in this study effectively identifies the spatiotemporal evolution characteristics of ecological quality, providing data support and theoretical references for precise management and policy formulation in ecological protection redline areas.