Integrating citizen science and machine learning to guide urban biodiversity planning: A case study of Belgium's top 10 urban regions

作者：Xiapeng Jiang, Liancheng Zhang, Hongwu Liang, Ying Qin, Kaidong Feng, Junjie Wang, Dries Bonte, Tim Van de Voorde · 发表于：Environmental Impact Assessment Review · 年份：2026 · DOI：10.1016/j.eiar.2026.108403 · 被引用次数：3 · 研究领域：Species Distribution and Climate Change、Smart Cities and Technologies、Wildlife-Road Interactions and Conservation

This study combines the Google Earth Engine (GEE) and machine learning algorithms (Random Forest and XgBoost) to analyze waterbird distribution patterns and the ecological constraints of the urbanization across Belgium's top 10 urban regions. By integrating INBO winter surveys (1991–2016) and eBird data (1973–2024), we developed a multi-scale model incorporating 25 natural and 2 anthropogenic factors. The results show that: (1) Under the “global modeling, local validation” framework, the Random Forest (RF) model shows superior generalization and stability compared to XgBoost (XgB). While XgB excels in computational speed, RF provides balanced performance across the full urban gradient, making it the optimal tool for developing national-scale conservation baselines. (2) Urbanization intensity acts as critical ecological filters reshaping habitat patterns. This drives a “core-periphery” distribution in metropolises like Brussels and also causes functional habitat collapse in industrial regions like Charleroi and Liège, whereas cities with “Blue-Green” networks (e.g., Antwerp) exhibit remarkable resilience. (3) Gap analysis reveals a “High suitability, low protection,” pattern. Cities like Kortrijk and Ghent face the most severe deficits, with protection gaps exceeding 99%, whereas Liège exhibits significant spatial conflict (>25%). Consequently, differentiated planning strategies are proposed: establishing flexible “urban micro-reserves” to fill legal vacuums in gap areas and d...