Radiation-correction of low-cost temperature stations in an urban context

Global warming will be one of the largest societal challenges to be addressed over the course of the 21 st century. In urban areas, different factors may be contributing to an amplification of the warming effect. Among others, especially high surface roughness inhibiting efficient ventilation, lower albedo due to a high fraction of dark surfaces, little vegetated areas and a high Bowen ratio due to non- permeable surfaces will promote the formation of an urban microclimate, the urban heat island (UHI). Quantifying the UHI effect should be the prioritized by city authorities, as this allows a focalization of heat mitigation strategies to improve the quality of life as well as the health of city dwellers. Heat mitigation measures may comprise de-pavement of surfaces, enlarging green spaces, correctly orientating new buildings within cold air density flows, and others. In collaboration with meteoblue AG, ZHAW has been operating a low-cost urban temperature network in Zurich and Basel, with a total of nearly 500 stations for a duration of 3 years. While quality assurance at night-time is relatively straightforward, this is less the case at daytime. The large heterogeneity of the station network in terms of urban climate zones as well as individual radiation errors of the stations made it indispensable to postprocess all the data. To this end, we developed a radiation correction algorithm based on station-based direct sunlight exposure as well as average incoming shortwave radiation and mean station error. Using reference stations in Zurich city to train the algorithm, we were able to reduce the root mean square error (RMSE) from 1.98 K to 1.47 K and the mean average error (MAE) from 0.96 K to 0.58 K in Basel. In Zurich, the training dataset allowed reducing the RMSE from 1.48 K to 0.92 K and the MAE from 0.92K to 0.29 K. As a tradeoff, while the algorithm may restrain the individual station dynamic to a certain degree, unrealistic temperatures (up to 6 K of radiation error) can be efficiently corrected. Once applied onto the entire dataset, the postprocessed data can then be used for other derived products (land use regression, model nudging, etc.)