Monitoring ecosystem service change in the City of Shenzhen by the use of high‐resolution remotely sensed imagery and deep learning

Information concerning land‐use change is imperative for improving conservation policies that promote sustainable land development. However, to date, most of the previous studies have largely focused on the use of coarse‐ or moderate‐resolution data, with which it may not be possible to identify the land‐use classes in urban environments. Due to the improved spatial details, high‐resolution (HR) remote sensing imagery provides us with an opportunity for the semantic interpretation of urban landscapes. Therefore, in this study, we took the City of Shenzhen (1997 km2) in China as an example to assess the detailed land‐use change and its effect on ecosystem services (ESs), based on HR satellite data from 2005 to 2017. In particular, deep learning was used to obtain accurate land‐use maps, because this technique is able to model the hierarchical representations of features and can thus effectively characterize urban scenes. The results revealed the following findings: (a) The overall accuracy of the proposed approach was 96.9% and 97.1% for 2005 and 2017, respectively, outperforming state‐of‐the‐art semantic classification models; (b) residential and commercial areas in Shenzhen increased dramatically over the study period by 10,416 and 9,168 ha, at the expense of ecological land; (c) supply capacity of the ecosystem decreased by 13.7%, but demand for ESs showed an increase of 23.5%. By courtesy of HR images, detailed land‐use changes and the associated ESs can be monitored, which facilitates the in‐depth understanding of urban environmental systems.