Local Climate Zone Classification System Using Web GIS Approach

Abstract

Climate change along with population growth and urbanization trends across the coastal areas are pressuring our cities to become more and more resilient to the ''new normal''. Our resilience is a function of hazards’ severity and probability, people’s and assets’ exposure and societies' sensitivity, and our ability in responding not only to a warmer climate, but also and foremost to extreme climate events will determine our resilience, in the future. Midlatitude countries such as Portugal have been experiencing warmer summers, in which heatwaves are becoming more frequent and severe, exposing an increasingly ageing population to correlated health and energy poverty issues. Part of the challenge moving forward is ensuring our ability to monitor and predict extreme heatwaves in space and time, with enough level of detail as to allow to highlight hot spots within cities and larger metropolitan areas which should be prioritized. The temporal aspect, and the spatial accuracy are something that the national and international weather services already take well care of - but there is a gap in what concerns managing cities, especially in more complex and coastal environments, where temperatures asymmetries are greater due to the complex land-sea interactions. To overcome this issue, remote sensing now offers spatially complete time series of observational data from which to derive insights into what concerns to the way we occupy these territories, as well as the thermal footprint of that land use land cover. When processed using machine learning and artificial intelligence, time series of geospatial data obtained from satellite imagery hence allows us to downscale weather and climate up to the neighborhood scale. But to do that we need to ensure consistency, precision, and standardization of processes and benchmarking of climate-relevant land use land cover data. Local Climate Zones (LCZ) is now the gold standard scheme for Land Use/Land Cover (LULC) classification; nevertheless, while the LCZ’s overall criteria are well-established, the methods employed for classification are very diverse, often local-specific. Alternatively, some highly scalable satellite imagery-based classification approaches have been attempted, based on open-source data – but these lack sufficient local detail, and may lead to misleading urban climate modelling results. This thesis aims to build vii upon pre-existent attempts at transforming geospatial data products into accurate and subkilometric resolution LCZ classification maps, to optimize them in terms of precision and scalability using open-source tools and libraries and a WEB GIS interface. The GIS-based classification is implemented using Portuguese and Denmark cities for benchmark with previous works. The objective is to attain the best-performing algorithm without overfitting it to a specific location, while its usability to tackle urban climate shall be tested, particularly the predictability of the Urban Heat Island (UHI) effect using such LCZ as a predictor, but open for scientific community to use.