The effect of auxiliary data (slope, aspect and elevation) on classification accuracy of Sentinel – 2A image using random forest classifier

Abstract

Land use/land cover (LULC) maps provide irreplaceable information about the earth’s surface, and investigation of their accuracy has always been an attractive research topic. Furthermore, it is known that accuracy of LULC maps could be improved using auxiliary data. One of the prevalent auxiliary data is the digital elevation model (DEM) because the surface landscape is affected directly by topography. Slope, aspect and elevation which are the main characteristics of the land surface are extracted from DEM. In this study, the effects of slope, aspect and elevation on classification accuracy were analyzed. For this purpose, Sentinel-2A satellite image together with the DEM data from the ALOS PALSAR satellite was assessed as auxiliary data for the classification process. Seven LULC classes covering the bulk of the study area were specified as urban, road, forest, water, bare and soil lands, cultivated and non-cultivated land in the classification process. To avoid possible bias among the determined classes, 700 pixels for training and 300 pixels for testing were chosen for each class. Classification results revealed that the highest accuracy (96.19%) were obtained when spectral bands were used together with elevation, slope, aspect data.