Investigating the Effectiveness of Supervised and Unsupervised Classification for Landsat Images Utilizing Classification Accuracy Assessment.

Abstract

This study filled in gaps in 2012 Landsat-7 satellite images using the "Nearest Similar Pixel Processor" algorithm.  Changes in LU/LC coverage in Babil province were classified and tracked during the period from 1999 to 2023. Landsat-7 and Landsat-8 satellite images were relied upon, focusing on specific spectral bands that provide a spatial resolution of 30 meters. Two classification methods were utilized: Maximum Likelihood for supervised classification and ISO Data for unsupervised classification. A metric of precision was employed to assess the effectiveness of individual classification techniques. This metric is determined by the ratio of correctly classified data points to the overall dataset size. The research conducted a comparative analysis of land use/land cover classification methodologies utilizing Landsat imagery: maximum likelihood classification and unsupervised ISO Cluster classification. The findings indicated that the MLC approach outperformed the ISO Cluster method in terms of accuracy. Four distinct land use coverage categories have been identified: urban, bare soil, water bodies, and vegetated lands. The study revealed significant changes in all land use coverage categories within Babil province from 1999 to 2023. In particular, the urban land area showed a significant increase, while the arid land area showed a decrease. The area of ​​water bodies varied during the study period, most likely due to differences in water discharge from the Euphrates River, which is affected by agreements between Iraq and Turkey and changes in rainfall. The distribution of vegetation covers also showed annual differences due to the relevant authorities' marketing plan for growing agricultural crops.

Keywords

• land use/land cover
• classification algorithms
• supervised classification
• unsupervised
• accuracy assessment