Assessment of agricultural land use suitability of the Büyük Menderes river basin with different multi-criteria decision-making methods

This study aimed to determine the suitable agricultural areas of the Büyük Menderes River Basin, using both traditional and recent techniques. Traditional methods included: i) Multiple Criteria Decision Making (MCDM) and parametric classification techniques, together with recent techniques; ii) Machine Learning (ML) algorithms as part of non- parametric classification techniques. The main goal of the research was to compare the results from parametric techniques and the hybrid techniques part of non-parametric classification methods. Analytic Hierarchy Process (AHP), Fuzzy DEMATEL, and SIMOS techniques of MCDM methods together with Random Forest (RF), Logistic Regression (LR), and Artificial Neural Networks-Multilayer Perceptron (MLP) of ML were utilized to identify suitable agricultural areas in the Büyük Menderes River Basin. The results obtained from each technique and algorithm were divided into four classes (very suitable, moderately suitable, slightly suitable, not suitable) in the GIS environment and reclassified to determine the agricultural suitability maps. According to the parametric classification techniques, the agricultural suitability map showed that the "moderately suitable" areas have the highest coverage with a percentage of 58.39% in the Büyük Menderes River Basin. This was followed by the "less suitable" areas with a percentage of 40.64%. The lowest rate was observed as 0.33% for "highly suitable" areas. In the agricultural suitability map showed that non-parametric classification techniques have the largest coverage of "not suitable" areas with 59.7%, 67.4% and 77.0% for the RF algorithm, LR algorithm and MLP algorithm respectively. The "less suitable" areas accounted for 15.2% with the RF algorithm, while the "highly suitable" areas accounted for 20.2% with the LR algorithm and 18.69% with the MLP algorithm. The lowest percentage was observed in the "highly suitable" areas with 11.7% for RF algorithm, followed by the "moderately suitable" areas with 6.0% for the LR algorithm and 2.02% for the MLP algorithm.