Document Type : Original Article

Authors

• Naser Ahmadi Sani ¹
• Sohrab Moradi ²

¹ Associate Professor, Faculty of Agricultural and Natural Resources, Mahabad Branch, Islamic Azad University, Mahabad, Iran.

² Assistant Prof., Department of Agricultural & Natural Resources Development, Faculty of Engineering, Payame Noor University, Tehran, I.R. Iran.

Abstract

Objective: Land use/cover has great importance for planning at different spatial scales in order to environmental sustainability. Land use/cover changes affects ecosystem services and products, socio-economic issues, climate change, natural resource and biodiversity. This study aimed to evaluate and compare different machine learning algorithms including classification and regression tree (CART), random forest (RF) and support vector machine (SVM) for land use/cover mapping in the south of Lake Urmia.
Methods: Sentinel-2A satellite data from 2023 were used within Google Earth Engine platform. Classification was performed using sample points with 70% for training and 30% for validation. The accuracy assessment was evaluated using the overall accuracy and kappa coefficient.
Results: Based on the land use / cover map, seven category were identified: water bodies, saline and rocky lands, irrigated farming, dry farming, built up areas, orchards, and ranges. The RF algorithm showed the highest overall accuracy (89%) while CART and SVM follow RF with 83% and 80%.
Conclusions: This study proved that RF is the best algorithm for optimal land use/cover classification, particularly in the study area. It also emphasizes the need to conduct similar studies with more advanced algorithms along with secondary data, especially in the Lake Urmia watershed, in order to achieve sustainable development.

Keywords

• Land use/cover
• Machine learning algorithms
• Sentinel 2 data

Main Subjects

• Remote sensing and Gis