Advanced

Effect of distance measures and feature representations on distance-based accessibility measures

Azzopardi, Jeremy LU (2018) In Master Thesis in Geographical Information Science GISM01 20181
Dept of Physical Geography and Ecosystem Science
Abstract
Distance-based accessibility measures are often built using vector representations of origin and destination features, and Euclidean or network-based distances. There are few comparisons of how the choice of feature representations and distance types affects results. Existing comparisons often use Spearman’s rank correlation coefficients. This study seeks to understand the effect of using different types of distance and feature representation on accessibility measures by comparing accessibility measures using the Bland-Altman plot, which measures the agreement between two variables.

Accessibility measures for recreational areas in Malta's Grand Harbour Area (GHA) were calculated. Two distance-based measures were compared: distance to... (More)
Distance-based accessibility measures are often built using vector representations of origin and destination features, and Euclidean or network-based distances. There are few comparisons of how the choice of feature representations and distance types affects results. Existing comparisons often use Spearman’s rank correlation coefficients. This study seeks to understand the effect of using different types of distance and feature representation on accessibility measures by comparing accessibility measures using the Bland-Altman plot, which measures the agreement between two variables.

Accessibility measures for recreational areas in Malta's Grand Harbour Area (GHA) were calculated. Two distance-based measures were compared: distance to nearest recreational area (DNRA) and Nearest Recreational Area ID (NRAID), measured from all residential blocks within the study area. Each of these two measures was calculated using two different vector representations for destinations (the recreational areas), access points and internal geometric centroids, and three different ways of measuring distance, Euclidean, network and full-network distance (network distance plus distance from feature representation to network). The combinations were compared for each measure. DNRA results were compared using the Bland-Altman plot and NRAID results were compared using percentage overlap.

Results and conclusions
Analysis showed that NRAID and DNRA results are especially affected by the selection of Euclidean distances versus network and full-network distances, and to a lesser extent the selection between centroids and access points, especially when using network or full-network distances. An important assumption in this study was that full-network distance and access points are the most realistic alternatives for their respective groups. They are also the most demanding in terms of data collection, pre-processing, computation and analysis.

One can conclude that the type of data and distance measure used and their possible effects on distance-based results must be taken into account. Some of the methods used to compare results were not fully applicable to the case and led to generic results. The conclusions drawn are therefore indicative and more work needs to be done to achieve more statistically valid results for comparison with other studies in the field. (Less)
Popular Abstract
Physical accessibility is an important consideration in public health and urban planning, and is often the subject of studies within the fields. A variety of methods is used to estimate physical accessibility to facilities such as open spaces. This study explores different ways of calculating a simple measure of accessibility. Accessibility to recreational areas (RA) on the island of Malta, specifically the Grand Harbour area, was used in this study.
The selected accessibility measure was the distance to the nearest recreational area from a residential block. This measure is solely distance-based. A concomitant element to this measure is the identity of the nearest RA. Both measure and concomitant element were calculated for each... (More)
Physical accessibility is an important consideration in public health and urban planning, and is often the subject of studies within the fields. A variety of methods is used to estimate physical accessibility to facilities such as open spaces. This study explores different ways of calculating a simple measure of accessibility. Accessibility to recreational areas (RA) on the island of Malta, specifically the Grand Harbour area, was used in this study.
The selected accessibility measure was the distance to the nearest recreational area from a residential block. This measure is solely distance-based. A concomitant element to this measure is the identity of the nearest RA. Both measure and concomitant element were calculated for each residential block within the study area using combinations of distances and ways of representing the RA. Straight-line distance (Euclidean distance) and two types of network distance (distance along a road network) were used. RA features were represented by their geometric centres (centroids) and their entrances (access points).
A comparison of the accessibility measure produced using combinations of distance measures and RA representations showed that selecting Euclidean distances over network distances gave an arithmetic mean difference of 177 and 271 metres, depending on the network distance type and feature representation. Agreement was high between network distances and low between Euclidean and network distances. The arithmetic mean differences from using access points versus centroids were 45 metres for Euclidean distances and 101 or 112 metres for network distances. Differences between Euclidean distances and network distances increased as measured distance increased.

Conclusions
Some of the methods used to compare results were not fully applicable to the case and led to generic results. If one assumed that network distances and recreational area entrance points are the most realistic, then using Euclidean distance is not very realistic and not to be recommended in most cases, although it is often very easy and practical to use. Differences between the two network distances used were very small and could be compensated. The selection of centroids versus access points also has a considerable effect on results. In conclusion, the results shown are indicative and more work needs to be done to achieve more statistically valid results which can be easily compared to results from other urban areas around the world. (Less)
Please use this url to cite or link to this publication:
author
Azzopardi, Jeremy LU
supervisor
organization
course
GISM01 20181
year
type
H2 - Master's Degree (Two Years)
subject
keywords
geography, GIS, Bland-Altman, Euclidean distance, geographic accessibility, network distance, recreational area
publication/series
Master Thesis in Geographical Information Science
report number
87
language
English
id
8946364
date added to LUP
2018-06-07 15:33:38
date last changed
2018-06-07 15:33:38
@misc{8946364,
  abstract     = {Distance-based accessibility measures are often built using vector representations of origin and destination features, and Euclidean or network-based distances. There are few comparisons of how the choice of feature representations and distance types affects results. Existing comparisons often use Spearman’s rank correlation coefficients. This study seeks to understand the effect of using different types of distance and feature representation on accessibility measures by comparing accessibility measures using the Bland-Altman plot, which measures the agreement between two variables.

Accessibility measures for recreational areas in Malta's Grand Harbour Area (GHA) were calculated. Two distance-based measures were compared: distance to nearest recreational area (DNRA) and Nearest Recreational Area ID (NRAID), measured from all residential blocks within the study area. Each of these two measures was calculated using two different vector representations for destinations (the recreational areas), access points and internal geometric centroids, and three different ways of measuring distance, Euclidean, network and full-network distance (network distance plus distance from feature representation to network). The combinations were compared for each measure. DNRA results were compared using the Bland-Altman plot and NRAID results were compared using percentage overlap. 

Results and conclusions
Analysis showed that NRAID and DNRA results are especially affected by the selection of Euclidean distances versus network and full-network distances, and to a lesser extent the selection between centroids and access points, especially when using network or full-network distances. An important assumption in this study was that full-network distance and access points are the most realistic alternatives for their respective groups. They are also the most demanding in terms of data collection, pre-processing, computation and analysis. 

One can conclude that the type of data and distance measure used and their possible effects on distance-based results must be taken into account. Some of the methods used to compare results were not fully applicable to the case and led to generic results. The conclusions drawn are therefore indicative and more work needs to be done to achieve more statistically valid results for comparison with other studies in the field.},
  author       = {Azzopardi, Jeremy},
  keyword      = {geography,GIS,Bland-Altman,Euclidean distance,geographic accessibility,network distance,recreational area},
  language     = {eng},
  note         = {Student Paper},
  series       = {Master Thesis in Geographical Information Science},
  title        = {Effect of distance measures and feature representations on distance-based accessibility measures},
  year         = {2018},
}