Authors
Ph.D., Remote sensing, School Buildings Department, the General Directorate of Education, Babylon, Iraq
Master degree, Geology/ Hydrology, Technical Engineering College of Kirkuk, Northern Technical University, Baghdad Road, 36001 Kirkuk, Iraq
Master degree, Photogrammetry, Technical Engineering College of Kirkuk, Northern Technical University, Baghdad Road, 36001 Kirkuk, Iraq
Ph.D. Prof. degree, Photogrammetry, Technical Engineering College of Kirkuk, Northern Technical University, Baghdad Road, 36001 Kirkuk, Iraq
[email protected]
Abstract
The purpose of this study is to investigate the performance of Sentinel-1A satellite dual-polarization radar data for land use land cover classifier in Kirkuk Governorate, northern Iraq. The area encompasses urban, agriculture and semi-arid plains with complex topography, so that traditional optical remote sensing methods are largely paralyzed by frequent dust storms and cloudiness. Two machine learning algorithms, Density Tree K-Nearest Neighbors (DT-KNN) and Random Forest (RF) were applied to enhance classification accuracy. The results indicated that the RF classifier obtained an accuracy of 98% for classification of the urban class and 99% for the DT-KNN classifier. Largely superior to DT-KNN while performance a relatively weaker for urban classes classifier, the robustness of RF to classify various land cover classes, as rocks and vegetation show, can be noted. But all of these classifiers had a problem when it came to certain categories, for example, rock areas being classified as urban land. This research shows that SAR data combined with machine learning can well remediate the lack of accurate optical data for land use and land cover classification in the spatially homogeneous regions, which are conducive for new era of smart farming and urban planning, especially in these complex terrains. It also means that there is a need for research regarding complementary models, to increase the classification accuracy in such complex environments.