A hierarchical rule-based land use extraction system using geographic and remotely sensed data: A case study for residential uses

Abstract

Lack of detailed land use (LU) information and of efficient data gathering methods have made modeling of urban systems difficult. This study aims to develop a hierarchical rule-based LU extraction system using very high resolution (VHR) remotely sensed imagery and geographic vector data. Land cover information extracted from remote sensing and several types of geographic data from the study area, City of Fredericton, Canada, are fused into a comprehensive database, in order to develop a sophisticated LU Extraction Expert System (LUEES). This paper illustrates how the proposed LUEES though a case study for residential uses in the study area. Morphological (individual-based) analysis at the building-level is carried out through a step-wise binary logistic regression model, which differentiates residential and non-residential buildings and results in an overall accuracy of 93.1%. The results derived from morphological analysis are then subject to a post-correction process using a spatial arrangement analysis, in order to further mitigate the misclassification issues arising from the morphological analysis. In this regard, Gabriel Graph connectivity examines the spatial structure and arrangements of urban features concerning different LU types. It is found that the spatial arrangement analysis further enhances the residential LU classification accuracy, which gives rise to an overall accuracy of 97.4%. It is believed that, equipped with such a powerful LU data collection tool and resulting detailed/accurate LU data, urban planners/modelers should be able to more reliably and precisely represent/predict economic interactions, activity locations, space and housing developments, business expansion, and trip patterns.