Monitoring ecosystem services in the Guangdong-Hong Kong-Macao Greater Bay Area based on multi-temporal deep learning.
Journal:
The Science of the total environment
PMID:
35122846
Abstract
Assessment of ecosystem service supply and demand, as well as the budgets of ecosystem service supply and demand, is the basis of scientific urban planning. In the 20 years between the proposal and formation of the Guangdong-Hong Kong-Macao Greater Bay Area (GBA), the natural ecosystem has been degraded and the ecological balance has been destroyed. In this paper, in order to assess the changes in ecosystem services in the GBA, a deep learning method composed of deep change vector analysis and the ResUnet model is proposed to achieve land use/land cover (LULC) mapping for 2000 and 2020. An index-based non-monetary evaluation method is then employed to quantify the value of the ecosystem services, and the spatial and temporal characteristics of the ecosystem service changes are analyzed. The results reveal that: (1) the proposed deep learning approach that combines deep change vector analysis (CVA) and model fine-tuning is able to achieve rapid and efficient LULC mapping in a large-scale area with multi-temporal image sequences. The overall accuracy of LULC mapping is 86.06% for 2000 and 86.67% for 2020. (2) The impervious surface area of all the cities in the GBA has increased significantly between 2000 and 2020, with an overall increase of 11.95%. (3) The mismatch between supply and demand for ecosystem services in the GBA has intensified, especially for provisioning, regulation, and cultural services. (4) The spatial distribution of the ecosystem service budget changes in the GBA shows aggregation characteristics and spatially positive correlation. These findings will provide important insights for promoting the coordinated development of the regional ecosystems and social economy in the GBA.