Determination of land use/land cover changes using Google Earth Engine and machine learning: a case study of Beypazarı, Ankara

Unregulated urban expansion, population growth, and climate change make monitoring land use and land cover (LULC) modifications an absolute necessity. To meet the need for efficient environmental tracking, cloud-based platforms like Google Earth Engine (GEE) now provide rapid pipelines for regional mapping. This study analyzes spatiotemporal LULC transitions across a 409.19-hectare area in the Beypazarı district of Ankara, Türkiye, using Landsat 8 satellite imagery from 2016 and 2020. The classification workflow utilized a Random Forest (RF) machine learning algorithm configured with 100 decision trees. To improve class separability, the primary Landsat spectral bands (Blue, Green, Red, NIR, SWIR1, and SWIR2) were integrated with three spectral indices: NDVI, MNDWI, and NDBI. Model training and validation were performed using 407 reference points, randomly split into a 70% training subset and a 30% testing pool. The results show that agricultural land and forest are the dominant land categories in the study area. For the 2016 dataset, the model achieved an overall accuracy of 89.51% and a Kappa coefficient of 0.868. For the 2020 dataset, the classification metrics dropped slightly to an overall accuracy of 77.93% and a Kappa coefficient of 0.724. Ultimately, this study demonstrates that cloud-computed remote sensing workflows offer reliable, large-scale mapping capabilities essential for sustainable land management.