A salt marsh vegetation extracting method with heterogenous-sensor remote sensing images

Salt marsh vegetation performs a vital role in wetland ecosystems. Accurate information on the distribution of salt marsh vegetation plays a crucial role in the sustainable management of wetland ecosystems. Extracting salt marsh vegetation with remote sensing is urgent and requisite for scientific control and management of wetland ecosystem. This study addressed to develop a method for extracting salt marsh vegetation based on multidimensional features from combined Landsat-8 and Sentinel-2 image stacks. Specifically, we explored the potential of using individual satellite and the integration of various sensors in extracting salt marsh vegetation. Multidimensional features obtained from the satellite images were used to evaluate the importance of different dimensional features. The results mainly indicated that multidimensional features synthetic image stacks achieved the best accuracy (overall accuracy (OA = 92.4%, kappa coefficient = 0.88). Methods with spatial feature incorporation show better performance in extracting sparse salt marsh vegetation, with the accuracy improved by 4.87% (OA = 87.23%, kappa coefficient = 0.78) comparing to the method with only spectral features (OA= 82.36%, kappa coefficient = 0.69). The 10m salt marsh vegetation map in 2015~2023 generated by the method enhanced both the accuracy and cost-effectiveness of salt marsh vegetation monitoring.