Spatially Optimised Sensor Networks for Efficient Urban Temperature Monitoring and Prediction

This research presents an integrated approach that combines a high-resolution Weather Research and Forecasting (WRF)-based experimental environment, a terrain- and land-cover-informed sensor placement algorithm, and a hybrid Random Forest–Gaussian Process model for mapping urban temperatures. It assesses the impact of the spatial arrangement of sensor stations on the reconstruction of temperature fields across complex urban terrains. The results indicate that static surface characteristics such as elevation and land use explain most of the daytime thermal variation and a significant portion at night. However, strategically placed, non-uniform stations are essential for capturing the remaining fine-scale gradients. An optimised network of around 200 sensors achieved mapping accuracy comparable to a uniformly distributed network exceeding 300 stations, proving that data-driven network optimisation can substantially reduce deployment costs. The benefits were most pronounced in topographically complex or thermally stable nocturnal conditions, where optimised layouts mitigated significant systematic errors. These outcomes emphasise that spatial intelligence in sensor placement can effectively substitute for dense measurements. However, a full representation of sub-grid processes still requires additional fine-scale meteorological inputs. The approach offers city planners a practical, open-source workflow for designing efficient temperature-monitoring networks that support resilient, data-informed, and climate-responsive urban development.