Backscatter statistical analysis using multi-temporal sentinel-1 SAR data for land cover mapping in Europe

Abstract

The aim of this study is to demonstrate the potential of using Sentinel-1 SAR data for mapping land cover using CORINE Land Cover nomenclature and Google Earth Engine, based on the analysis of the temporal variation of the backscatter. To this aim dualand singlepolarization, interferometric Wide swath mode (IW) for year 2016 over five different test sites in Europe was used. Backscatter temporal variation for different statistic parameters and for different combination of VV and VH polarization was analyzed for a better understanding of its signature for different land cover classes and for obtaining a good classification of the data. In addition, the statistical class separability for different class pairs and the NDVI for each test area were analyzed. Support Vector Machine (SVM) method was used to classify the dataset. During the classification, different scenarios were used to find out which combination would give a good classification with a high accuracy. The different combinations of VV and VH polarizations were used in conjunction with other data, like DEM, aspect and local incidence angle to obtain a high accuracy. The resulting classified images have been assed using the overall accuracy and the Kappa coefficient. Results demonstrated that while using only the mean single polarization, the classification accuracy would be low, but by using a combination of dual polarization in conjunction with ancillary data, the accuracy would increase, namely the overall accuracy increases up to 90.82% and Kappa coefficient up to 0.85. Thus, the information on the temporal variation of the backscatter can be used successfully for mapping different land cover types.