Statistical models for urban growth forecasting: With application to the Baltimore–Washington area

Abstract

Monitoring and governing the development of cities are the major concerns of urban planners, since involve physical and social aspects, such as land use and population trends. Models for spatial growth have been developed both from the mathematical and empirical viewpoints, with the aim of forecasting and decision-making. Statistical models require regular space–time datasets that are provided by recent remote-sensing and geographic information systems (GIS). In this paper, we consider space–time autoregressive (STAR) models that can be applied to the timelapse video of land transformations available on Internet. The corresponding datasets are in the form of big 3D arrays and require fast algorithms of parameter estimation and forecasting. An extended application to a hybrid timelapse video over 200 years of urban growth of the Baltimore–Washington area is carried out. The video is built by combining remote sensing imagery, census data, historical cartography and data interpolation, and can be fitted and forecasted by adaptive STAR models, with robust and varying parameters.