 We propose a modified version of the UTAE model which uses a single multiband image composite as input instead of multiple individual images. This allows us to reduce the amount of noise present in the original images, resulting in better performance compared to other models. Additionally, we evaluate the effectiveness of different combinations of input data, such as Sentinel-2, Sentinel-1, spectral indices, and ALO's elevation data. Finally, we demonstrate that our model outperforms other popular approaches, such as random forest, UNET, and SEG-former, while requiring fewer training parameters. This article was authored by Anastasios Sapkinlis, Konstantinos Marthaglou, and Nikos Grimalidis.