This thesis examines urban inequalities in the Canadian context and focuses on Ottawa, the capital city.
Firstly, income inequality in the eight largest Census Metropolitan Areas in Canada, between 1971 and 2016, is examined through an analysis of income polarization and its spatial patterns. The middle-income group has declined across CMAs, while the low-income group has usually expanded. Concurrently, increasing spatial fragmentation is identified in every CMA examined. Local spatial autocorrelation identified clustering of high-income Census Tracts (CTs) suggesting that these areas are more resilient to fragmentation. Therefore, patterns of urban inequality are ones of an unequivocally disappearing middle-income population with an increasingly spatially fragmented urban income-scape.
At local levels, inequality is manifest in processes such as gentrification. To increase the spatial and temporal ability to monitor and map gentrification in a large city, artificial intelligence and Google Street View imagery were used to identify visual improvements to properties that are indicative of gentrification between 2007 and 2016. A deep Siamese convolutional neural network (SCNN) and VGG19 backbone was trained to recognize visual gentrification-like changes of properties over time. This deep mapping model achieved a 95.6% level of accuracy in identifying the visual signs of property improvement using 86110 georeferenced photographs of individual properties in Ottawa, Canada. Given that the residential/commercial property itself is the atomic object of gentrification, properties identified as having undergone a gentrification like visual change were mapped as points to produce kernel density maps that reduce noise and identify regions of high visual property change intensity (hot spots). The intensity of visual property improvements exhibited strong concordance with the spatial pattern of building permits between 2011 and 2016. The results confirmed areas known to be undergoing gentrification and also presented areas where the process was not previously suspected of occurring.
Thirdly, a select set of census-based quantitative methods of modelling gentrification were compared between 2006 and 2016 at both the CT and the finer dissemination area unit of analysis. All models were tested against their ability to predict the density of GSV-points per unit residential area that were predicted by the Siamese deep learning model. For the CT level, two new regression models were created using all the census variables identified within the learned literature. An OLS multivariable regression model was created using backward stepwise regression, after which only age (youth), dwelling-value, income & occupation were retained. Residual spatial dependence in the OLS required a spatial linear model specification. A spatially lagged simultaneous autoregressive model (SAR Lag_y) explained 57% of the variance in GSV-point density in Ottawa. Out of all the models tested, the SAR Lag_y possessed the strongest spatial correlation with the original pattern of GSV point density as measured using Lee's L bivariate spatial autocorrelation statistic. A second model was produced using quasibinomial regression in order to predict the probability of a given CT being gentrified. That model achieved 91.7% accuracy. Out of the five reproduced models from the literature, one preformed close to as well as our new models in predicting GSV-point density per unit residential area. While there is some agreement between models that purport to measure gentrification, there are considerable differences between models, which suggests that census-based gentrification measure should be locally focused.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/45082 |
| Date | 21 June 2023 |
| Creators | Ilic, Lazar |
| Contributors | Sawada, Michael C. |
| Publisher | Université d'Ottawa / University of Ottawa |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | application/pdf |
| Rights | Attribution-ShareAlike 4.0 International, http://creativecommons.org/licenses/by-sa/4.0/ |
Page generated in 0.0024 seconds