Predicting land cover changes and carbon stock fluctuations in Fuzhou, China: A deep learning and InVEST approach

Abstract

Predicting the spatiotemporal dynamics of land cover and its carbon stock holds significant importance in guiding regional sustainable development, enhancing regional carbon stocks, and addressing global climate change. However, there is insufficient research on the quantitative relationships between various land cover types and carbon stock changes, as well as their future spatial predictions. Focusing on the core area of Fuzhou City, China, this study constructs a streamlined framework by coupling deep learning and the InVEST model to predict urban land cover and carbon stock changes in 2025 and 2035. The results show that: (1) The prediction model for land cover change has high applicability and can produce the simulated images with high accuracy. Impervious surface is expected to increase by 53 km2 in 2025 and 131 km2 in 2035 compared to 2020, resulting in considerable reductions in forest and cropland. (2) Carbon stocks of the study area are expected to decrease by 1.68 × 106t in 2035 compared to 2020 due to large amounts of high-carbon-density forests and croplands being converted into low-carbon-density impervious surfaces. (3) Multiple regression analysis reveals that forests have the largest impact on carbon stocks in the area, with a magnitude 5.25 times greater than impervious surfaces and 11.5 times greater than cropland, whereas impervious surfaces are the second most influential land cover type on carbon stock changes. Therefore, expanding forest areas becomes an essential initiative as forests could offset the carbon stock loss caused by impervious surface growth. This study provides scientific references for optimizing land-use planning and formulating policies for the development of low-carbon cities.

Keywords

• Carbon stocks
• Deep learning
• InVEST model
• Land cover
• Prediction
• Fuzhou city