Abstract
Urbanized areas demonstrate an increase in temperature compared to areas with greater quantities
of vegetation and a low density of building structures. This temperature difference is known
as the urban heat island effect (UHIE). Remotely sensed land surface temperature (LST) maps
allow for the study of surface urban heat islands (SUHIs).
However, LST image data often contains pixels with missing values due to cloud obstruction
and voids where the data has been discarded due to sensor failure and other causes of signal
contamination. Thus, urban climate studies often focus on a limited set of cloud-free LST scenes
which provide a snapshot view of intra-urban thermal variability. To perform a long term analysis
of LST relationships with other factors it is fundamental that the maps are complete and all
the missing data values are filled prior to further analysis.
Meteorological ground stations are not affected by the presence of cloud, but establishing high
spatial resolution observational networks for long-term air temperature monitoring is a challenge.
Crowd-sourced air temperature measurements from personal weather stations (PWSs)
present an opportunity to supplement the official weather station record. Nevertheless meteorological
stations still represent a sparse and irregularly distributed dataset, while LST images
represent a large spatial field of temporally synchronized observations over diverse terrain.
This thesis investigates a novel multivariate copula-based approach to produce complete cloudfree
LST maps. The models exploit the dependence relationships between LST, and air temperature,
and vegetation. Since erroneous or poor quality data may lead to inaccurate analyses,
the air temperature measurements from all monitoring stations were assessed and anomalous
observations removed.
The predicted LST maps are compared to a climatological model and their application in urban
climate studies evaluated with respect to micro-urban heat islands (MUHIs), where different
landscape properties are described by multi-temporal local climate zone (LCZ) thematic maps.
Overall, the objectives of this study are to 1) evaluate the quality of air temperature data for
inclusion in an LST–air temperature model, 2) generate a copula-based model to predict LST
intensities, and 3) assess the spatial variability of LST and MUHIs with respect to LCZs by comparing
distinct zones associated with densely populated settlements, urban centres and differing
levels of vegetation in urban areas.
Keywords: satellite remote sensing, land surface temperature, local climate zones, urban heat
island, copula models, missing data values.