The urban configuration compensation framework; Connecting spectral datasets employing spatial pattern recognition in favour of detecting configuration styles
By Yneke van Iersel
This century could be named the “Urban Century”, where the urban population is expected to become 2.4 billion by 2050. In the coming years, the design of a city becomes more important, as more people are expected to be living in urban areas and challenges have arisen, including climate change, energy transition and biodiversity conservation. Urban configuration plays an important role in these challenges. Improving urban planning could help solving environmental challenges, however, a reliable and accurate method to gather land-use data is essential. In the research field of geo-information science (GIS), remotely sensed datasets are used widely to deduce landcover and land-use maps. The accessibility to remote sensing data used to be a limitation for public organisations, however, in the last decades more remote sensing data has become available without charges. Despite the potential of this reliable input data, aerial pictures hinder the analysis of areas where landcover classes overlap vertically, as merely the ‘top layer’ is shown. As space is limited in urban areas, stacking landcover classes vertically is implemented more often, which triggers a mismatch between input data and real-world characteristics. Therefore, this study proposes the urban configurations style framework. The framework has been constructed by executing four steps. First, the spatial characteristics of six main Dutch urban configuration styles were delineated by five landcover classes. Second, spatial indicators were selected to describe the morphology per configuration style. The sensitivity study has pointed out that the spatial scale to perform the spatial indicators should be 300 by 300 meters. Third, high-resolution spectral data was classified into landcover classes utilizing a Random Forest classifier and compared to a topographical dataset to test the hypothesis that spectral data can improve the urban planning analysis. The most striking result is the increase of vegetation coverage in the predicted landcover class compared to the topographical dataset, which seizes differences up to 10%. The fourth step includes the deduction of locations of configuration styles by implementing the spatial indicators. Per spatial unit, the output of the spatial indicators was compared to the morphology description per configuration style. The 68% confidence interval was concluded as favourable to define the morphology per style. The spatial pattern of the classified configuration styles was expected to follow the distribution of Amsterdam’s historical extension. Although spatial patterns are recognisable, the results show a scattered distribution. The results of this thesis show the potential of the urban configuration compensation framework, however, some limitations of the results should be noted. One limitation of this framework is the number of configuration styles taken into account. Another limitation is the selection of spatial indicators, which determines the accuracy of distinction of the configuration styles based on their urban morphology. Besides, the classification of the landcover tiles into configuration styles depended on the point crediting framework. The final classification could be improved by executing the point crediting system for multiple scenarios or integrating this classifying step into an automated algorithm. Despite these limitations, the final classification of urban areas into urban configuration styles enables urban planners to detect focus areas. The results show that the framework enables the linkage of classified urban areas to extra information. After all, the urban configuration compensation framework could be adjusted to be implemented in other urban areas than Amsterdam.