Crowdsourced air temperatures contrast satellite measures of the urban heat island and its mechanisms

Abstract

The ubiquitous nature of satellite data has led to an explosion of studies on the surface urban heat island (SUHI). Relatively few have simultaneously used air temperature measurements to compare SUHI with the canopy UHI (CUHI), which is more relevant to public health. Using crowdsourced citizen weather stations (>50,000) and satellite data over Europe, we estimate the CUHI and SUHI intensity in 342 urban clusters during the 2019 heat wave. Satellites produce a sixfold overestimate of UHI relative to station measurements (mean SUHI 1.45°C; CUHI 0.26°C), with SUHI exceeding CUHI in 96% of cities during daytime and in 80% at night. Using empirical evidence, we confirm the control of aerodynamic roughness on UHI intensity, but find evaporative cooling to have a stronger overall impact during this time period. Our results support urban greening as an effective UHI mitigation strategy and caution against relying on satellite data for urban heat risk assessments.

Fig. 1. Distribution of city-rural temperature differentials for July 2019 in Europe.

(A) Private weather station (n = 59,810) air temperature and (B) satellite-derived LST are relativized to mean rural temperatures for 342 urban clusters. Temperature differentials give an indication of UHI intensity illustrated by the zoomed-extent windows for London, Paris, and Berlin.

Fig. 2. SUHI and CUHI intensities.

(A) Frequency distribution of all available hourly (local time) UHI intensities for European urban clusters (n = 342) during July 2019. (B) Hourly mean (point) and interquartile range (vertical line) for UHI intensities over each hour of the day.

Fig. 3. Determinants of SUHI and CUHI intensities.

(A) Predictor variable importance scores are plotted for RF models explaining the variance in local-scale (n = 5703 stations; 47,652 pixels) SUHI and CUHI over a subset of urban clusters (30) with building height data. Models are stratified into nighttime and daytime mean UHI for roughness and evapotranspiration. Mean importance values are plotted for the total predictor variable suite in (B) and (C).

Fig. 4. Partial dependence of UHI on roughness and evapotranspiration.

Lines represent the smoothed effect of roughness and evapotranspiration on local-scale (n = 5703 stations; 47,652 pixels) SUHI and CUHI intensity after controlling for the effect of all other explanatory variables. RF models were used for daytime and nighttime daily UHI separately.