Random forest classification of salt marsh vegetation habitats using quad-polarimetric airborne SAR, elevation and optical RS data

This research investigated the use of multi-source remote sensing data to map natural coastal salt marsh vegetation habitats. Coastal zones are very dynamic and provide a number of critical ecosystem services, particularly in relation to flood mitigation but they have been found to be difficult to monitor using remotely sensed data. This research analysed combinations of S-band and X-band quad-polarimetric airborne SAR, elevation data and optical remotely sensed imagery. In total 30 variables were analysed. The SAR inputs included backscatter intensity channels and Cloude-Pottier, Freeman-Durden and Van Zyl decomposition SAR descriptors. Classification was carried out using Random Forest classifiers at two thematic resolutions which generated a general mapping of salt marsh vegetation and a high-resolution mapping of thematically detailed salt marsh vegetation habitats. The results indicate that Random Forest models are able to handle multi-source datasets and generate high classification accuracies. Models based on either SAR or optical RS variables alone were found to be less accurate than models that combining variables from multiple sources. The results show that X-band SAR data provided the best information to map vegetation extent and analysis showed that S-band SAR data was better able to differentiate between different vegetation habitats. The methods and analyses suggested in this paper extend previous research into remote monitoring of costal zones and illustrate the opportunities for mapping natural coastal areas afforded through combinations of radar and optical remote sensing data.