Climate driven trends in London's urban heat island intensity reconstructed over 70 years using a generalized additive model

Long-term urban heat island (UHI) observations are uncommon and where available, are generally unable to distinguish changing climate drivers from urban expansion; neither driver is treated independently. We overcome this limitation using a generalized additive model to learn the variability in UHI intensity (UHII) at a central London weather station (St James's Park) over a 10-year observation period (2010-2019). We then use the model to reconstruct 70 years (1950-2019) of monthly night-time UHII variability using ERA5 reanalysis data both as a reference in UHII calculation and for the predictors. We find considerable variability both seasonally and annually within the UHII time series (monthly mean maximum UHIIs are 1.4-2.9 degrees C). Applying extreme value analysis to the time series we show that monthly mean maximum UHIIs are likely to exceed 2.75 degrees C once every 11 years. Considering that most studies observe or model UHIIs for less than a year, they will likely misrepresent this UHII variability. Nevertheless, despite moving to a warmer background climate, London's UHII has not significantly changed across the period of analysis (1950-2019). The data-driven methods we create in this study are easily transferable to other cities.