Classifying settlement types from multi-scale spatial patterns of building footprints

Warren C Jochem,Andrew J Tatem,Patricia Jones,Heather R Chamberlain,Oliver Pannell,Douglas R Leasure

doi:10.1177/2399808320921208

Warren C Jochem, Andrew J Tatem + Show 4 more

Open Access

https://doi.org/10.1177/2399808320921208

Copy DOI

Abstract

Urban settlements and urbanised populations continue to grow rapidly and much of this transition is occurring in less developed countries. Remote sensing techniques are now often applied to monitor urbanisation and changes in settlement patterns. In particular, increasing availability of very high resolution imagery (<1 m spatial resolution) and computing power is enabling complete sets of settlement data in the form of building footprints to be extracted from imagery. These settlement data provide information on the changes occurring in cities, particularly in countries which may lack other data on urbanisation. While spatially detailed, extracted building footprints typically lack other information that identify building types or can be used to differentiate intra-urban land uses or neighbourhood types. This work demonstrates an approach to classifying settlement types through multi-scale spatial patterns of urban morphology visible in building footprint data extracted from very high resolution imagery. The work uses a Gaussian mixture modelling approach to select and hierarchically merge components into clusters. The results are maps classifying settlement types on a high spatial resolution (100 m) grid. The approach is applied in Kaduna, Nigeria; Kinshasa, Democratic Republic of the Congo; and Maputo, Mozambique and demonstrates the potential of computational methods to take advantage of large spatial datasets and extract meaningful information to support monitoring of urban areas. The model-based approach produces a hierarchy of potential clustering solutions, and we suggest that this can be used in partnership with local knowledge of the context when creating settlement typologies.

Highlights

Urban settlements and urbanised populations have been growing at unprecedented rates around the world (Seto et al, 2010; UN Department of Economic and Social Affairs, 2019)
Very high resolution (VHR) imagery has been used to monitor changes and to classify land uses within urban areas (Graesser et al, 2012; Kuffer et al, 2014), but such imagery requires different analysis approaches using object-based classification and textural features instead of methods based on per-pixel spectral indices (Engstrom et al, 2015; Kit et al, 2012; Kuffer et al, 2016)
The reduced sets of layers were used for the model-based clustering, and Bayesian Information Criterion (BIC) values were used to compare models

Summary

Introduction

By 2050, the United Nations predicts 68% of the world’s population will live in cities and towns, and most of this change will be occurring in low- and middle-income countries (UN Habitat, 2016) This transition towards more urbanised living has potential to impact health, livelihoods, and family structures (Benza et al, 2017; WHO & UN-Habitat, 2010). At the same time, well-designed and managed urban areas have the potential to improve sustainability and more efficiently provide services, access to facilities, and resources to larger numbers of people (Seto et al, 2010) These tensions in the transition towards urbanising populations have gained attention from policy makers. Microsoft has used neural networks to produce complete building footprint datasets from imagery for the US and Canada (Bing Maps Team, 2018, 2019) These automated extraction approaches are building on efforts such as OpenStreetMap (http://www.osm.org) and crowd-sourced efforts to manually digitise structures from imagery. These efforts, and others, have produced a range of outputs mapping human settlements at different spatial resolutions (Roy Chowdhury et al, 2018)

Objectives

Methods

Results

Discussion

Conclusion