hyperparameter tuning

Optimal classification of satellite images enhances the extraction of spatial-temporal information for land use and land cover classification, enabling the examination of various socio-economic and environmental impacts. Initially, three supervised classification algorithms—K-Nearest Neighbors, Support Vector Machines, and Random Forests—were utilized with default parameters. The results indicated that the Random Forest algorithm outperformed the others in accuracy. Consequently, the Random Forest classification algorithm on the Google Earth Engine platform, using Landsat 8 images as input for 2013 to 2023, was employed. The impact of different image combinations based on several spectral indices on the final classification accuracy was assessed. The findings show that using Landsat products (TOA) and the GEE platform provides faster and more accurate classification, with optimal hyperparameter values for 2013, 2018, and 2023 being 280, 180, and 80, respectively. The findings indicate an increase in vegetation cover alongside gradual urban expansion, reflecting stable development policies. However, this growth is primarily attributed to the establishment of large leisure areas and tree planting on the city outskirts. Despite initiatives aimed at reducing air pollution and promoting community well-being, vegetation in central urban areas has either remained stagnant or declined in certain areas. Considering the implications of climate change and increased heavy rainfall, the lack of vegetation in central zones may lead to the formation of impermeable surfaces, exacerbating the risk of flash floods. This underscores the critical need for enhanced research focus and proactive measures in this field.