Quantifying the spatio-temporal temperature dynamics of Greater London using thermal Earth observation

Abstract

Urban areas are highly sensitive to extreme events such as heatwaves. In order to understand how cities will respond to thermal stress it is critical to quantify not only their temporal temperature dynamics but also their spatial temperature variability. However, many cities lack weather station networks with a sufficient spatial distribution to characterise spatio-temporal intraurban temperature dynamics. One means by which spatially complete measurements of urban temperature may be derived is to employ satellite thermal Earth observed data. While some success has been achieved in understanding the temperature characteristics of cities using such data, relatively little work has been undertaken on establishing the use of long time-series Earth observed data as a supplement or alternative to screen-level air temperatures frequently utilised in urban climatological studies. In this thesis a software framework, centred around the use of a spatial database, is developed which can be used to gain an improved understanding of how satellite thermal Earth observed data can be used in the long timeseries analysis of urban temperature dynamics. The utility of the system is demonstrated by processing a 23 year time series (1985-2008) of 1,141 Advanced Very High Resolution Radiometer (AVHRR) images and hourly United Kingdom (UK) Met Office weather station measurements for the Greater London area. London was selected as the region of interest as it is the UK’s only megacity, and has been shown to exhibit both a significant urban heat island and a severe increase in population mortality during previous heatwave events. The software framework was employed to conduct two inter-related sets of analysis. First, the relationship over time between AVHRR estimated surface temperature (EST) and screen-level air temperature records is investigated and quantified. The resulting relationships are then used to produce an empirical model that can predict spatially complete summer-season air temperi atures for London. Cross-validation testing of the model at selected London weather stations showed model root mean square error (RMSE) ranging from 2.70 to 2.94°C and absolute errors in air temperature estimation of 0.45 to 1.67°C. A key finding of the thesis is that the minimal variation in prediction error between the different stations indicate a level of spatial robustness in the model across the urban surface, that is within the limits of the AVHRR EST precision. In addition, the model was used to estimate spatially averaged air temperatures over the Greater London area for selected summers, and showed a maximum error in air temperature prediction of 1.44°C. Furthermore, the prediction error for the heatwave summer of 2003 was 0.51°C, suggesting that such a model can successfully be used to estimate air temperatures for extreme heatwave summers. Such predictions are directly relevant to future assessments of urban population exposure to heatwaves, and it is envisaged that they could be used in conjunction with a population vulnerability index to create a spatially complete heatwave risk map for London. This work is then extended to investigate the utility of satellite estimated surface temperature measurements to characterise temporally and spatially intra-urban heatwave dynamics using the commonly employed urban heat island intensity metric (UHII). Analysis of the AVHRR EST found that the data are highly sensitive to local meteorological conditions, and that temporal aggregation at the monthly scale is required to provide robust data-sets for inter-year analysis of summer temperatures and generation of the UHII metric. Statistical testing of EST and air-temperature derived UHII for the heatwave summer of 2003 against other non-heatwave summers showed no significant increase in intensity at the 95% confidence level. This raises questions as to the applicability of the UHII metric to capture increases in urban temperatures during a heatwave event.