Application of artificial intelligence in prediction of future land use / land cover for cities in transition: a comparative analysis

Future simulations of land use and land cover (LULC) are required to effectively manage patterns of interactions between land use classes due to ambitious sustainability goals and integrated urban planning policies. This study investigates the performance and suitability of two simulation models, artificial neural networks (ANN) and Logistic Regression (LR), in a selected region of a Metropolitan area- Istanbul, where the dynamic LULC change due to new transportation facilities is challenging to monitor and to predict. The historical LULC data was derived from optical satellite images with the use of a hybrid classification approach. Explanatory variables were used in an ANN - multilayer perceptron (MLP) data-driven model to generate possible transition maps, whereas for LR, the transition from barren land to urban regions and the transition from barren land to industrial areas were modelled. Both methods achieved acceptable simulation performance around 75% kappa and 0.85 ROC metrics, and they both suggest that urban areas and industrial - commercial units show an increasing trend at the expense of barren lands. At this point, the LR method, which uses the transition pattern from barren lands to urban areas provided an advantage by reflecting the historical situation where the urban increment rate was still high when compared to industrial area increment that came to a standstill between 2014 and 2020. The projected results of 2026 endorse the increasing trend in urban settlements around the main arteries for the study area.