Optimising crowsourced weather data point density for urban climate studies

Over the last decade, the use of Citizen Weather Stations (CWS) has gradually gained acceptance within the urban climate community. Among the reasons for that uptake are the increasing spatial and temporal coverage provided by these stations, together with sophisticated quality control and gap filling techniques. Although their potential is undeniable, challenges remain regarding uneven spatial coverage and the presence of sensor deserts, particularly in vulnerable areas. In addition, the growing volume of CWS data requires screening out the most representative points to reduce density and facilitate data processing. In this study, we use CWS data from Madrid, to investigate how CWS density and their spatial distribution might affect the accuracy of urban climate severity maps. A spatial optimisation algorithm is used to identify the most representative CWS. This is tested with six clustering techniques, which results in 120 different optimisation scenarios. Urban climate severity maps are then produced for each scenario using Empirical Bayesian Kriging (EBK) regression. The accuracy of each scenario is evaluated against a baseline scenario using all the available CWS. Results show that selecting 30-50% of the total number of stations yields  relatively small deviations from the baseline while preserving the overall urban heat distribution pattern. Furthermore, for the same number of CWS, accuracy among clustering techniques varies by up to 18% when compared to the baseline. These findings underscore the dual benefits of our approach: first, it facilitates the identification of optimal locations for new sensors in areas with insufficient coverage, addressing sensor deserts. Second, it provides a framework for scenarios with high density CWS data, enabling efficient resource allocation when not all data is necessary to produce reliable results. This highlights the potential of spatial optimisation in enhancing urban climate monitoring and decision-making, particularly in addressing vulnerabilities associated with uneven sensor distributions.