Scenario modeling of urban expansion and its impact on farmland using artificial neural networks: A case study of Chiro Town, Oromia, Ethiopia

This study employs a data-driven, Machine Learning approach to model and predict urban expansion and its impact on farmland in Chiro Town, Oromia, Ethiopia. Historical Land Use and Land Cover (LULC) maps for 1990, 2005, and 2020 were generated using Random Forest classification of Landsat imagery, validated through ground truth and accuracy metrics (overall accuracy: 87–99%; Kappa: 0.78–0.98). Key spatial drivers including NDVI, NDBI, slope, building density, and proximity to roads and rivers were integrated into an Artificial Neural Network (Multilayer Perceptron) to simulate future urban growth scenarios for 2035. Results indicate substantial urban expansion, with built-up areas increasing from 0.27 km 2 (4.57%) in 1990 to 1.32 km 2 (22.37%) in 2020, projected to reach 1.65 km 2 (27.97%) by 2035. Farmland has concurrently declined from 5.56 km 2 (94.08%) to 3.32 km 2 (56.32%), and is projected to fall below 50% of the study area by 2035, highlighting persistent conversion of agricultural land to urban uses. The model further identifies spatial patterns of urban growth concentrated along roads and flat terrains, with shrubland expansion and increasing bare land reflecting land degradation and farmland abandonment. By combining historical change detection, Machine Learning-based scenario modeling, and GIS-based spatial analysis, this study provides a robust methodological framework for predicting LULC dynamics and informs urban planning policies aimed at mitigating farmland loss and promoting sustainable land management in rapidly urbanizing contexts.