Intercomparison and combination of low-cost urban air temperature measurement approaches

Measurements of urban air temperatures (Tair) are vital to successful adaptation and mitigation policies to increasing urban heat stress. However, in-situ measurements in cities are often scarce and costly, and therefore low-cost approaches are increasingly used to study urban Tair. This allows for inexpensive, yet still highly spatially and temporally resolved observations of urban Tair. Despite their merits, a common issue of such low-cost approaches is lacking data quality and potential measurement errors. In this case study, we compare three low-cost measurement approaches regarding their ability to capture intra-urban variability of Tair over a period of 24 hours in Bern, Switzerland: a) A network of 79 low-cost measurement devices (LCD), b) bicycle mounted mobile measurements (BCY), and c) 581 Netatmo citizen weather stations (CWS). As the BCY sensor is actively ventilated, it is used as the reference for intercomparisons with LCD and CWS. Compared to the BCY, the median difference of Tair for LCD is found to be slightly negative over the entire study period (???0.08 K) as well as during night-time (???0.10 K), and positive during daytime (0.05 K). As the LCD are known to exhibit a positive bias during the daytime, the good agreement of BCY and LCD is speculated to indicate a positive daytime bias in BCY as well. The CWS show a positive median difference of 0.67 K over the entire study period, 0.98 K during night-time, and a negative difference of ???0.23 K during daytime. It is hypothesized that these biases result from incorrect siting of the CWS by their owners installing CWS to close to buildings or walls. At night, these emit thermal radiation which could lead to the positive bias whilst the negative bias during daytime might result from buildings shading the CWS. BCY and LCD both show a distinct pattern of nocturnal intra-urban Tair variability, which is less pronounced in the CWS measurements. Furthermore, the intercomparison of the three approaches across local climate zones reveals that CWS do not well represent forested areas. Whilst the bias sources of the individual approaches require in-depth investigation in future studies (e.g., external heat sources and measurement height for BCY, daytime shortwave radiation errors for LCD, and nocturnal thermal heating by nearby buildings for CWS), we conclude that combining the three measurement approaches can allow to reduce the shortcomings of each approach regarding spatial and temporal resolution or correct biases inherent to one approach.