Coastal Wetlands Classification and Carbon Storage Estimation: A Case Study of the Region Surrounding the South China Sea

Abstract

Coastal wetlands are typical carbon sinks and play a crucial role in achieving global carbon neutrality goals. The region surrounding the South China Sea (SCS) contains abundant coastal wetland resources and strong carbon sequestration capabilities, which can be effectively assessed by the well-known integrated valuation of ecosystem services and tradeoffs (InVEST) model. InVEST requires accurate spatial distribution information of wetlands as input data, which can be obtained by coastal wetlands classification methods. Among all classification methods, deep learning (DL) is the state-of-the-art. However, designing a DL method that is both easily trainable and suitable for large-scale coastal wetland classification remains a challenging issue. This article proposes a novel DL method and a new carbon correction strategy for large-scale coastal wetland classification and carbon storage assessment. First, the remote sensing (RS) data from the study area is acquired and preprocessed by the Google Earth Engine. Second, the Otsu algorithm and decision tree are used to extract the maximum wetland extent. Third, a multidirectional squeeze attention network (MDSAN) is proposed for large-scale coastal wetland classification. Finally, a new strategy is designed to correct measured carbon pool data using meteorological data. Experiments show that the proposed wetland classification method achieves an overall accuracy and Kappa coefficient of 0.9500 and 0.9411, respectively, demonstrating the effectiveness of MDSAN. Furthermore, the estimated carbon storage in the mangroves, tidal-flats, and swamps surrounding the SCS is approximately 1.2112$\boldsymbol{\times }10\boldsymbol{^{9}}$ t, 6.9138$\boldsymbol{\times }10\boldsymbol{^{7}}$ t, and 1.6980$\boldsymbol{\times }10\boldsymbol{^{8}}$ t, respectively, revealing the carbon distribution pattern in the region.