Performance of supervised classification of Murundus Fields from the integration of radar and optical data

Abstract

Wetlands are located at the interface between terrestrial and aquatic environments, playing important ecosystem functions at both regional and global scales. The Murundus Fields are typical wetlands of the Cerrado biome, which face threats due to human activities and climate change, making their monitoring and conservation essential. This study aims to assess the accuracy of supervised classification of the Murundus Fields in the rio Claro basin, located in the Triângulo Mineiro region (Minas Gerais, Brazil), through the combination of optical and radar sensor data. For this purpose, radar images from the Sentinel-1 satellite and the Copernicus DEM digital elevation model, as well as optical images from the Sentinel-2A satellite and the NDWI spectral index, were used. The Google Earth Engine (GEE) platform was employed to create distinct models, varying the types of input data and the temporal scale. The results showed that, over the course of one year, the combined use of all datasets achieved the best classification accuracy, with 91% overall accuracy and a kappa index of 0.79. However, the optical sensor data had greater relative importance in the classification, resulting in high accuracy (89% overall accuracy and a kappa index of 0.74) even when radar-derived data were not used. Sentinel-1 imagery is more effective in detecting the Murundus Fields when the time interval covers only summer, due to its ability to detect in the presence of clouds.