Preparation of land use maps using machine learning algorithms and generation of future (2030-2050) land use projections in the Aşağı Kelkit Basin

In a world where natural resources are limited and diminishing every day, using these resources correctly is crucial for a sustainable future. In this study, it is aimed to create land cover/use maps of the Aşağı Kelkit Basin, which has not undergone much deterioration in terms of its natural elements, and to create land use projections by revealing its temporal and spatial changes. For this purpose, satellite images of 1990(Landsat 5), 2010 (Landsat 5) and 2020 (Landsat 8) have beenused to find the trend of change in land use. The land cover in the study area has been classified into a total of 9 classes using the CORINE land cover/use classification: forested area, dry agricultural fields/fallow, irrigated agricultural fields, wetlands, meadows/pastures, settlement/road areas, degraded areas, bare lands, and shrublands. Utilizing satellite images, the most commonly used machine learning algorithms - Random Forest, K-Nearest Neighbours, and Support Vector Machine - have been used to create land cover/use maps for the years 1990, 2010, and 2020. To find out the robustness and accuracy of the machine learning algorithms, Kappa accuracy analysis has been used. Due to the low accuracy rate of machine learning algorithms in the land cover/use classification for the year 1990, the land cover/use map of 1990 was created with maximum likelihood classification, which is a controlled classification method. In the land cover/use maps for the year 2010, the machine learning algorithm with the highest accuracy rate was the random forest algorithm, for the land cover/use maps of the year 2020, the machine learning algorithm with the highest accuracy rate was the K-nearest neighbours algorithm. Using the spatial change results with the highest accuracy coefficients for the years 1990 and 2010, a projection for the year 2020 was created. This projection was then compared with the reference 2020 land cover/use map to determine the accuracy and reliability of the projection. With an accuracy rate of 0.81, a well-performing classification was used to create land use projections for the years 2030 and 2050. According to the results obtained, in the study area by the year 2030, forested areas, irrigated agricultural fields, degraded areas, and shrublands are expected to decrease, while dry agricultural fields, wetlands, meadows, settlements, and bare lands are projected to increase. In the year 2050, forested areas, irrigated agriculture, wetlands, and degraded areas are anticipated to decrease, while dry agriculture, meadows, settlements, bare lands, and shrublands are expected to increase.