Optimising crowdsourced weather data point density for urban climate studies

Crowdsourced data from Citizen Weather Stations (CWS) offer high spatial and temporal resolution for urban climate studies, but their uneven distribution and potential redundancy raise questions about spatial representation and optimal network design. This study systematically evaluates how CWS density and spatial configuration influence the accuracy and the preservation of urban thermal structures in climate severity maps. Using Madrid as a case study, a quality-controlled dataset of 589 stations was subjected to multiple clustering-based selection strategies and random sampling across a wide range of densities. For each clustering technique and density, Cooling Degree Days were interpolated using Empirical Bayesian Kriging Regression, and the interpolated maps were assessed against a full-data reference map using Mean Absolute Error and Local Moran's I statistic. Results show that increasing station density significantly improves interpolation accuracy below approximately 1.2 stations/km 2 , but gains diminish beyond this threshold, indicating a practical density threshold. While accuracy converges across configurations at moderate to high densities, spatial pattern preservation remains sensitive to network geometry, particularly at the municipal scale. The analysis further examines two forms of thermal structures underrepresentation: spatial sensor deserts, caused by insufficient coverage, and functional sensor deserts, where stations fail to capture local thermal variability despite proximity. These findings demonstrate that density alone does not ensure representativeness. Instead, combining sufficient density with strategic sensor placement is essential to improve both accuracy and spatial coherence in urban climate assessments.