Identify detailed dynamic processes and driving forces of forest fragmentation based on explainable AI: Taking a global biodiversity priority area as a case

Global forests are increasingly fragmented by population growth and intensified human activities, threatening the stability and sustainability of forest ecosystems. Thus, a comprehensive quantification of forest fragmentation is crucial to guiding effective forest protection and reforestation policies. However, most studies mainly examine the overall dynamics of forest fragmentation, without delving into its detailed dynamic processes, resulting in an insufficiently thorough understanding of the specific details of forest fragmentation in different regions across the globe. Here, taking the Wuyishan Biodiversity Priority Area, a priority area for global biodiversity conservation as a case, we developed a forest fragmentation index (FFI) and a dynamic forest fragmentation index (ΔFFI) that integrate patch edge, isolation, and size effects; then, spatial modes in the forest fragmentation process were identified by overlaying the three components of FFI—edge density (ED), patch density (PD), and mean patch area (MPA); finally, the explainable AI (XAI) model was employed to analyze the driving mechanisms of forest fragmentation. The results showed that from 2000 to 2023, over 97.9% of forest landscapes in the study area experienced moderate fragmentation, and 56.9% exhibited a decline in fragmentation (ΔFFI < 0). Spatially, FFI exhibited pronounced clustering, with coldspots and hotspots occupying approximately 32% and 27% of the area. Moreover, more than half of the region was characterized by synergistic clusters (High-High and Low-Low patterns). In contrast, ΔFFI displayed a more scattered distribution without obvious clustering. The typical modes of FFI decline (ED down PD down MPA up , i.e., decreasing ED and PD with increasing MPA) and increase (ED up PD up MPA down , i.e., increasing ED and PD with decreasing MPA) accounted for approximately 14.8% and 32.2% of the forest landscapes, respectively. Among the remaining atypical modes, the ED down PD down MPA down (i.e., decreasing ED, PD, and MPA) mode was the most prevalent (35.1%). The XAI revealed that distance to forest edge (DTFE) and its changes (DTFEC) were the most significant factors influencing forest fragmentation and its dynamics. At the grid level, more than 90% of the FFI/ΔFFI was negatively affected by DTFE/DTFEC. Our findings enhance the understanding of both the static and dynamic characteristics of subtropical forest fragmentation, emphasizing the need to improve connectivity between forest patches. • A multidimensional forest fragmentation index based on ED, PD, and MPA is developed. • Detailed forest fragmentation dynamics are revealed by different fragmentation modes. • Driving forces of forest fragmentation dynamics are identified by an explainable AI. • ED up PD up MPA down and ED down PD down MPA down are the most common fragmentation modes. • Distance to forest edge and its change are key drivers of forest fragmentation.