Land Use Changes Modeling and Future Predictions Using CA-ANN Simulation in the Watershed of 25 (Shenroud, Siahkal)

In this study, we analyzed the spatial–temporal trends of land use dynamics from 2000 to 2021 using remote sensing data. The image classification was based on three main land use classes, i.e. forest, artificial areas (agriculture and built-up), and other natural areas (bare lands, grassland, plantation, shrubland, water bodies, and woodlands). Maps of land use changes in the area for 2000-2021 show that built-up areas have increased by 9.3%. In contrast, forest and other natural areas decreased by 7.1% and 2.2%, respectively. In addition, an integrated CA-ANN (Artificial Neural Networks- Cellular Automata) model was used to predict land use changes from 2021–2042. The percentage of correctness for the simulation was 91%, and the overall kappa value was 0.86. Similar to the classified maps in 2000-2021, the prediction maps from 2021–2042 illustrated increasing trends in built-up areas by 4.7% and decreasing trends in the forest by 4.26% and other natural areas by 0.4%. In this work, we implemented ordinary least squares (OLS) regression models to predict land cover changes in the study area as a function of explanatory variables [elevation, slope, and proximity variables - distance to the city center, roads, villages, and streams-]. The results of the OLS models showed a relatively good performance for predicting land use changes with an R-squared value greater than 0.5. These results provide important knowledge that can help develop future sustainable management and planning and help managers make informed decisions to improve environmental and ecological conditions.