Combining ENMAP Hyperspectral Imaging and Machine Learning for Land Use/Cover Classification

Abstract Imaging spectroscopy enables detailed recording of the Earth’s surface and ecosystems’ properties at a global level. Due to the specifications of the hyperspectral sensors, it becomes possible to utilize the hyperspectral data in a multitude of applications, among which the mapping, monitoring and understanding of land use and land cover (LULC). The Environmental Mapping and Analysis Program (EnMAP) is a German satellite mission that aims to monitor and characterize the Earth’s surface on a global scale providing high-quality data with the advantage of open access. This study explores the capabilities of the EnMAP satellite in LULC mapping. The goal of the study is to leverage the advantages and potential of state-of-the-art machine learning algorithms, in particular of Support Vector Machine (SVM) and Random Forest (RF), in combination with EnMAP’s hyperspectral data, for mapping LULC. The study area is characterized as a typical Mediterranean environment. The thematic LULC maps accuracy was assessed using statistical methods and further analysis was conducted through McNemar’s statistical test to evaluate the statistical significance of the differences in the results. All in all, the SVM algorithm proved to be more accurate than RF algorithm, with an overall accuracy of 91% compared to 86%, respectively. The McNemar’s test confirmed the higher accuracy of the SVM algorithm’s results, at least this was the case in our study. The study’s findings highlighted EnMAP’s hyperspectral data promising role of in the field of LULC classification. This study adds valuable information towards evaluating EnMAP satellite data in producing thematic maps in typical Mediterranean environments and allows for comparisons with similar data sets in other areas or with data from other sensors in similar environments.