Characteristics and Driving Factors of Forest Coverage in Western Fujian Based on Deep Learning

Forest is an essential component of terrestrial ecosystems. Investigating the spatiotemporal changes in regional forest cover, and the factors driving these changes, can provide critical guidance for the protection of forest resources and the sustainable development of the ecological environment. The research area in this study is Longyan City in Fujian Province. Using the GEE cloud platform, we obtained satellite remote sensing image data for Longyan City for the years 2003, 2013, and 2023. The Attention-UNet model, combined with the PALSAR FNF label layer, was used to segment Longyan City into forested and non-forested areas. We analyzed the spatiotemporal changes in forest cover over the past 20 years and used geographic weighted regression to explore the driving factors of changes in forest cover in Longyan City, resulting in the following conclusions. (1)The deep learning Attention-UNet model, combined with the PALSAR FNF label layer, effectively segmented forested and non-forested areas in Longyan City with accuracy, precision, recall rate, and F1 scores of 0.851, 0.854, 0.851, and 0.853, respectively. (2) Temporally, over the last 20 years, the forest cover area in Longyan City increased from 13 620.89 km² in 2003 to 15 675.44 km² in 2023, with an increase of 2 054.55 km². (3) Spatially, 27.5% of the total area had increased forest cover which was mainly concentrated in the central part of Changting County, the southern part of Yongding District, and the border areas between the eastern part of Wuping County and the western part of Shanghang County. Areas with decreased forest cover accounted for 6.3% of the total area and were primarily located in the southwestern and northeastern parts of Yongding District and the southwestern part of Xinluo District. Forest cover in Longyan City exhibited significantly positive spatial correlation and spatial heterogeneity. (4) GWR results show that from 2003 to 2023, nighttime light (NL), land surface temperature (LST), precipitation (PRE), and vapor pressure deficit (VPD) had negative impacts on forest cover in Longyan City, while soil moisture shifted from a negative influence in 2003 and 2013 to a positive influence in 2023. The regression coefficients of the five driving factors displayed distinct spatial differentiation characteristics. For example, the impact of NL on forest cover decreased over time, and areas with significantly negative impacts of LST on forest cover were mainly concentrated in the central and western regions of Longyan City. Over the past 20 years, soil moisture, NL, VPD, and PRE were the primary drivers of increased forest cover, while NL, PRE, LST, and soil moisture were the main drivers of decreased forest cover. This study provides a scientific basis for achieving forest resource protection and sustainable development.