 The paper presents an enhanced convolutional neural network based on UNET for simulating urbanization in Tehran and Karaj regions, Iran. The proposed method uses a pixel-wise semantic segmentation approach with spatial drivers affecting urbanization as data cubes. Independent variables such as altitude, slope, and distance from various land cover types were considered as covariates. The modified UNET outperforms the random forest algorithm in terms of area under the total operating characteristic and percent correct metrics, demonstrating its ability to accurately simulate built-up land expansion while considering neighborhood effects. Therefore, the modified UNET is recommended for simulating urbanization without relying on a cellular automata model. This article was authored by Hanes Shoji, Saeed Nadi, Hossein Shafizadeh Mabadam, and others.