Land use and land cover classification from satellite images based on ensemble machine learning and crowdsourcing data verification

Abstract

ABSTRACT Land use and land cover (LULC) is a crucial spatial data because it is an essential input for most analysis and prediction tasks. Presently, LULC is obtained easily and more quickly than was in the past due to the rapid progresses in remote sensing, geographic information systems, and machine learning technologies. For this reason, the satellite images data are processed to classify land use. This method has advantages over conventional field survey at the actual area in various aspects, such as time and cost-saving. However, processing of LULC data from the low-resolution satellite images has some limitations; its results are of lower accuracy than when doing it from high-resolution satellite images. Furthermore, the LULC acquired from the processing at some areas are not consistent with the actual situation of the same. Therefore, this research proposed a novel LULC classification using the satellite images from Landsat 8 and Sentinel-2, based on the machine learning and ensemble machine learning techniques. The accuracy assessment results showed that the proposed method classified LULC with high accuracy at 88%. In addition, it introduces the development of an LULC notification system using volunteer or crowdsource data. Accordingly, the obtained LULC data was notably improved, in line with the actual states at the studied areas.