The role of spatial representation in the developm

Authors: MariePierre Parenteau Michael Charles Sawada

Publish Date: 2010/10/08

Volume: 5, Issue: 3, Pages: 311-323

PDF Link

Abstract

A land use regression LUR model for the mapping of NO2 concentrations in Ottawa Canada was created based on data from 29 passive air quality samplers from the City of Ottawa’s National Capital Air Quality Mapping Project and two permanent stations Model sensitivity was assessed against three spatial representations of population population at the dissemination area level population at the dissemination block level and a dasymetrically derived population representation A spatial database with land use roads population zoning greenspaces and elevation was created Polycategorical zoning data were used in dasymetric mapping to spatially focus population data derived from the dissemination blocks to a subblock level for comparison purposes Dasymetric population mapping provided no significant LUR model improvement in explained variance when compared to block level population however both the former were significantly better than the dissemination area level population representations However where block level population is not available or too costly to acquire our method using polycategorical zoning data provides a viable alternative in LUR modelling endeavoursModelling chronic air pollution exposure to constituents like nitrogen dioxide NO2 at an intraurban scale is fundamental for health planning and intervention within cities The land use regression LUR model was first introduced in 1997 by a team of European researchers Briggs et al 1997 but it was not until 2005 that a first attempt at using this methodology in North America was published Gilbert et al 2005 Since then LUR models have been developed for only a limited number of large centres in Canada Gilbert et al 2005 Henderson et al 2007 Jerrett et al 2007 Marshall et al 2008 Su et al 2008 Wheeler et al 2008 Sahsuvaroglu et al 2009 Poplawski et al 2009 The modelling of air quality based on LUR requires accurate data on a number of human and environmental factors such as land use street networks location of greenspace and population distribution Each of these variables can and has been integrated within LUR models using a wide variety of spatial representations and spatial scales As the number of articles published that employ land use regression models has been increasing a research agenda that focuses on the role of spatial representation and scale in the LUR model performance is warrantedIn order to improve LUR model development and choice our principal objective is to examine the role of spatial representation of the LURindependent variables used to model atmospheric NO2 concentrations A second objective of this research aims at developing a reasonably accurate LUR model for Ottawa Canada which has not been attempted before Ottawa is often considered a unique city because of its small manufacturing base and large government and technology sector activities So developing a LUR for this city is challenging considering the size of Ottawa and the low industrial activity found within its boundaries Wentz et al 2002 Jerrett et al 2007 The use of a population independent predictor is the key element that many published European and North American LUR models have in common There are a numerous ways in which population has been represented in LUR modelling efforts and these commonly include the number of dwellings per unit area the population count per unit area and population density Henderson et al 2007 Ryan et al 2008 Beelen et al 2009 For the most part operationalising the population variable is achieved by using available census data at different geographic levels The use of multiple population representations found in the learned literature begs the question how robust are LUR results to the use of different population representations at different spatial scales Since the question of spatial representation is a fundamental consideration the modifiable areal unit problem Openshaw 1984 in population representation is a concern Andresen and Brantingham 2008 With regard to our present LUR undertaking we have tried to limit our scope by looking more specifically at the spatial representation of the population variable and its ensuing effects on LUR model output performance We thus address the specific question How robust are LUR models to different population representations as independent variablesTo our knowledge no research has yet studied the role of spatial representation in the development of a LUR model more specifically for the population variable With this research we propose to address this issue for the first time by developing regression models based on three different representations of population from the Canadian Census of Population the population count at the dissemination area DA level the population count at the dissemination block DISB level and the population count at a subdissemination block level using dasymetric mapping DASYMThe need for data integration arises when one wants to use data collected under a different spatial division eg noncensus tract or nondissemination area level but a finer or custom geographic boundary set than the one used by the census Fisher and Langford 1996 or when wanting to understand or intervene at a scale that is finer than that collected by the census This would be the case when the goal is to examine natural socioeconomic processes that are indifferent to the imposed nonphysical boundaries As such spatial units are often incompatible with respect to the required or intended needs of the researcher and so areal interpolation techniques are required Langford 2007 Solving this problem of incompatibility requires the assignment of one aggregated dataset to another incompatible dataset using various available spatial algorithms Sadahiro 1999 Mennis 2003 Reibel and Bufalino 2005 Reibel and Agrawal 2007 Langford 2007 The approaches developed to solve the problem of incompatible spatial units have the capability of generating a more precise map of population distribution or many other census derived variables Dasymetric mapping which can also be pycnophylactic Tobler 1979 in nature is the spatial interpolation method used in this research It is a method that is based on the integration of ancillary spatial data Ancillary datasets like roads greenspace water land use and cadastral data can help to define both where people could live as well as where they cannot live within a predefined area As such a dasymetric approach provides a method by which the original dataset representing for example population counts in census tracts can be disaggregated and redistributed to a finer spatial scale The use of this approach also corresponds to the last goal of this research which is to work toward the development of a standardised methodology for dasymetric mapping Langford and Higgs 2006We use dasymetric mapping in the context of LUR but it could also have numerous other applications where population data at fine spatial resolutions are required For example the availability of an accurate representation of the population distribution for governance the governance of oneself and of others is very important for the task of administrating services Crampton 2004 It can be argued that an accurate map of human population is essential to municipal planning even more so for public health planning and healthcare provisions Hay et al 2005 Global disaster management for the developing world has given rise to projects like LandScan Bhaduri et al 2007 and others that are producing dasymetric gridded global population estimates at fine spatial resolutions that compare well with known population distributions in the developed world Sutton et al 2003 Sabesan et al 2007 Patterson et al 2009 It is clear that accurate data on the spatial distribution of population is fundamental to a number of endeavours Liu et al 2008 but few studies have focused on the question of spatial representation in terms of population distribution and its impact on policies One of the few examples of research on the subject is the work of Langford and Higgs 2006 who investigated the influence of alternative spatial population representations to the measure of potential access to primary healthcare services The authors found that the modelling method for population impacted the results The authors concluded that the use of dasymetric mapping consistently provides lower estimates of accessibility to healthcare which in terms of policy and planning could have a significant impact Research in the field of environmental justice has also started addressing the question of using alternative population representations Most et al 2004 Brindley et al 2005 Mohai and Saha 2006 2007 Hence this research will also contribute to the advancement of these other fields of studiesIn Europe and North America LUR modelling efforts to map exposure to NO2 have performed generally well with R 2 values varying from approximately 05 to 09 In general Canadian research has yielded acceptable results with R 2 values between 054 and 077 Even though the predictor variables have been generally the same for most studies their specifications have been significantly different Jerrett et al 2007 Ryan and LeMasters 2007 Hence not only it is very important to understand the sensitivity of the models to spatial representation in order to obtain consistency in the results but it is also very important as these exposure models are in most cases one of the first steps in the study of the relationship between exposure and health Exposure models with an improved spatial resolution have been found to produce more robust associations Sahsuvaroglu et al 2009 when compared to health conditions We expect dasymetric mapping will aid in obtaining better LUR models Hence better LUR models will contribute to the literature on the relationship between exposure and health Still the main objective of this research is to give a first approximation to the role of spatial representation in the development of LUR models and contribute to the research agenda on the role of spatial representation more generallyLand use regression models were developed for application in European cities and research on the subject was first published by Briggs et al 1997 Those authors were interested in exposure models that would allow the study of the relationship between health and air pollution at a local scale Work by these authors also corresponds to the period when academia started to be interested in the spatial distribution of pollutants not only in the context of interurban studies but also for intraurban studies Jerrett et al 2005

Keywords: