Ecological Security Pattern Based on XGBoost-MCR Model: A Case Study of the Three Gorges Reservoir Region

Deliang Sun,Xiaoqing Wu,Haijia Wen,Xianglong Ma,Fengtai Zhang,Qin Ji,Jialan Zhang
DOI: https://doi.org/10.1016/j.jclepro.2024.143252
IF: 11.1
2024-01-01
Journal of Cleaner Production
Abstract:The ecological resistance surfaces (ERS) are a critical component in constructing the Ecological Security Pattern (ESP) and reflects the ecological security level of a region. However, research on optimizing the construction of the ERS is limited. This study focuses on selecting 11 ecological resistance factors from two categories: natural conditions and human activities. Using the XGBoost (eXtreme Gradient Boosting) - MCR (Minimum Cumulative Resistance) algorithm, a comprehensive framework to analyze the ESP in the Three Gorges Reservoir Region (TGRR) is constructed. The results reveal the following key findings: (1) The ecological source areas in the TGRR decrease from northeast to southwest, covering a total area of approximately 12,400 km2, accounting for 21.57% of the total area. (2) The ecological corridors amount to 45, exhibiting spatial distribution patterns that vary across different regions. In the northeast, ecological corridors demonstrate a branching and radial distribution pattern, while in the southwest, they primarily extend along the edges of the study area. Corridors in the central region extend from west to east. (3) The ecological security assessment based on the XGBoost algorithm demonstrates high reliability, as the predicted results align well with the actual conditions in the TGRR. (4) The comprehensive framework of the ESP, based on the XGBoost-MCR model, allows for automatic optimization of algorithm parameters using known environmental samples. The study contributes scientific references to ecological environmental protection in the TGRR, while also offering valuable insights for optimizing ERS construction, ESP development, and integrating machine learning techniques in ecological security research.
What problem does this paper attempt to address?