Precise mapping of coastal wetlands using time-series remote sensing images and deep learning model

Abstract

Mapping coastal wetlands' spatial distribution and spatiotemporal dynamics is crucial for ecological conservation and restoration efforts. However, the high hydrological dynamics and steep environmental gradients pose challenges for precise mapping. This study developed a new method for mapping coastal wetlands using time-series remote sensing images and a deep learning model. Precise mapping and change analysis were conducted in the Liaohe Estuary Reserve in 2017 and 2022. The results demonstrated the superiority of Temporal Optimize Features (TOFs) in feature importance and classification accuracy. Incorporating TOFs into the ResNet model effectively combined temporal and spatial information, enhancing coastal wetland mapping accuracy. Comparative analysis revealed ecological restoration trends, emphasizing artificial restoration's predominant role in salt marsh vegetation rehabilitation. These findings provide essential technical support for coastal wetland ecosystem monitoring and contribute to the study of sustainability under global climate change.