Determination of existing solar farms and site selection for solar farms using Google Earth Engine platform

Renewable energy sources are important in terms of reducing the negative effects of climate change, sustainability and protection of environmental values. In this study, by focusing on solar energy among renewable energy sources, it is aimed to determine the current locations of solar fields by using remote sensing data and machine learning algorithms, and then to recommend suitable locations for new installation areas. There are mainly three stages in the study: (i) Identification of existing solar farms using Machine Learning Algorithms, (ii) Identification of suitable areas for solar farms using GIS based Multi-Criteria Decision Analysis (MCDA) Methods and (iii) Comparison and evaluation of existing solar farms and potential areas. In the first stage, the existing solar farm installed capacity of Antalya province, which was determined as study area, was classified using Machine Learning Algorithms of Random Forest (RF), Support Vector Machines (SVM) and Classification and Regression Trees (CART) via the cloud-based Google Earth Engine (GEE) Platform. The Land Surface Temperature, Digital Elevation Model (DEM) and indices related to vegetation, water, soil, humidity, urban, were added to the satellite images as additional bands and hyper parameter tuning were applied. The effects of the additional bands on the accuracy results were evaluated. The accuracies of the classification algorithms were compared by Error Matrix and McNemar's Test methods. According to the results obtained, an up-to-date solar farm map has been created for Antalya province. In the second stage, site selection was carried out using MCDA methods to identify suitable areas for solar farms. At this stage, Analytic Hierarchy Process (AHP) and Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) methods were used. Criteria related to climate, logistics and geography were determined based on the literature review and expert opinions. After the criteria maps were created and criteria weights were obtained, potentially suitable area maps were created using Weighted Overlay analysis for AHP and Inverse Distance Weighted (IDW) interpolation method for TOPSIS. In the last stage, the detected solar farm areas were compared with the determined suitable areas, and the results were evaluated in terms of planning. According to the results obtained, the SVM algorithm provided the most accurate result in determining existing solar farms with overall accuracy value of 89%. The overall accuracy value was computed as 86% for RF algorithm result and 72% for CART algorithm result. These values were obtained with the Error Matrix, and they were found to be consistent and meaningful when compared with the results of the pairwise comparison matrix using McNemar's test. According to the site selection results, 27.46% of the study area was found to be very suitable for solar farm site selection with AHP, while this rate was 34.93% in TOPSIS results. While the results obtained with AHP were more detailed, the results obtained with TOPSIS showed a more generalised distribution due to the effect of interpolation.