Analysis of past and future urban growth on a regional scale using remote sensing and machine learning

Abstract

This research investigates Land Use and Land Cover (LULC) changes in the Porto Alegre Metropolitan Region (RMPA). A 30-year historical analysis using Landsat satellite imagery was made and used to develop LULC scenarios for the next 20 years using a Multilayer Perceptrons (MLP) model through an Artificial Neural Network (ANN). These maps analyze the urban area’s expansion over the years and project their potential development in the future. This research considered several critical factors influencing urban growth, including shaded relief, slope, distances from main roadways, railway stations, urban centers, and the state capital, Porto Alegre. These spatial variables were incorporated into the model’s learning processes to generate future urbanization scenarios. The LULC historical maps precision showed excellent performance with a Kappa index greater than 88% for the studied years. The results indicate that the urbanization class witnessed an increase of 236.78 km2 between 1990 and 2020. Additionally, it was observed that the primary concentration of urbanized areas since 1990 has predominantly occurred around Porto Alegre and Canoas. Lastly, the future forecasts for LULC changes in 2030 and 2040 indicate that the urban area of the RMPA is projected to reach 1,137.48 km2 and 1,283.62 km2, respectively. In conclusion, based on the observed urban perimeter in 2020, future projections indicate that urban areas are expected to increase by more than 443.29 km2 by 2040. The combination of remote sensing data and Geographic Information System (GIS) enables the monitoring and modeling the metropolitan area expansion. The findings provide valuable insights for policymakers to develop more informed and conscientious urban plans, as well as enhance management techniques for urban development.