Environmental Monitoring of Land Use/ Land Cover by Integrating Remote Sensing and Machine Learning Algorithms

• Published in: Journal of engineering and sustainable development (Online) Vol. 28; no. 4; pp. 455 - 466
• Main Authors: Aljanabi, Firas, Dedeoğlu, Mert, Şeker, Cevdet
• Format: Journal Article
• Language: Arabic; English
• Published: Mustansiriyah University/College of Engineering 01.07.2024
• Subjects: Geographic information system; Land use/ land cover; Machine learning; Random trees; Remote sensing; Support vector machine
• ISSN: 2520-0917; 2520-0925
• DOI: 10.31272/jeasd.28.4.4

Evaluation of the land use/ land cover (LULC) case over large regions is very important in a variety of domains, including natural resources such as soil, water, etc., and climate change risks and LULC change has emerged as a high anxiety for the environment. Therefore, we tested and compared the performance of three classification algorithms: Support Vector Machines (SVM), Random Trees (RT), and Maximum Likelihood (MaxL) to derive and extract LULC information for the district of Sarayönü/ Konya across five distinct classes: water, plantation, grassland, built-up, and bare land. Two remote sensing indices, the normalized difference vegetation index (NDVI) and the normalized difference water index (NDWI), were used as supplementary inputs for the classification of LULC. To evaluate the performance of the algorithms, a confusion matrix was employed. The average overall accuracy of support vector machines, random trees, and maximum likelihood algorithms was found 85.60%, 79.20%, and 74.80%, respectively, and 82.00%, 74.00%, and 68.50% for the Kappa coefficient. These results indicate that the support vector machines algorithm outperforms other algorithms in terms of accuracy. As a result of the research, it was determined that classification algorithms integrated with remote sensing in LULC change monitoring/determination could produce accurate classification maps that can be used as base data. This is due to the ability of machine learning algorithms to learn complex patterns, adapt to diverse data, and continuously improve, making them achieve higher accuracy compared to traditional classifiers. Therefore, their use was recommended for decision-makers.