Using interpretable machine learning to reveal the impact of rapid development in the emerging urban area on carbon storage: Marginal and interaction effects
Year: 2025
Type: preprint
Abstract: With the rapid urbanization of emerging metropolitan areas, accurately assessing carbon storage changes driven by land use transitions is essential for understanding regional carbon cycling and identifying spatial patterns of carbon sources and sinks. This study employs a “past-to-future” analytical framework to explore the spatiotemporal evolution of carbon storage. By integrating interpretable machine learning (IML) methods, we systematically identify main drivers and quantify their marginal effects and interaction mechanisms. This approach overcomes the limitations of traditional models, which often assume linear relationships and overlook complex nonlinear interactions. Our key findings are as follows: (1) From 2000 to 2020, total carbon storage decreased by 6.25 Tg, with woodlands having the highest carbon density and built-up areas the lowest. (2) Under future scenarios SSP126 and SSP245, carbon storage is projected to increase by 52.20 Tg and 39.48 Tg, respectively, mainly due to woodland expansion in eastern hilly regions. In contrast, under SSP585, a decline of 23.15 Tg is expected, primarily caused by urban expansion replacing carbon-rich ecosystems. (3) Digital Elevation Model (DEM), slope (SLOPE), and population (POP) were identified as the most influential variables. DEM and SLOPE show a positive influence on carbon storage, while POP has a negative effect. Meanwhile, these main drivers also exhibit strong interaction effects and temporal variability. This study offers a solid scientific basis for regional land use planning and provides actionable insights for enhancing carbon storage in rapidly urbanizing regions.
Authors: Yugang Wang, Zilin Ye, Xun Gong, Zhixiao Zhang, Fanqiang Meng, Liuhui Zhao, Dongmei Wang, Yuanjun Tong, Zhengjun Gong
Institution: Southwest Jiaotong University
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4413859327