Machine Learning and GIS Approaches to Spatial Prediction of Urban Air Temperature

Air temperature is measured directly at weather stations, with small spatial coverage. Weather observations, geographic information systems (GIS), and machine learning can be used to estimate the temperatures in these areas based on the conditions around them. This paper uses literature synthesis and case study analysis to review the data and methods used in spatial prediction of urban air temperature. It investigates the aspects of meteorological observation, spatial and environmental predictors of the city, GIS-based feature extraction, machine learning models, and temperature mapping. Three case studies in Zurich, Oslo, and Szeged demonstrate the application of high-resolution mapping, the predictive value of urban features, and seasonal and day/night variations in model results. The reviewed studies indicate that machine learning can represent complex relationships between air temperature and meteorological and spatial variables. GIS can bring support in this process by integrating data, extracting features, and mapping predictions. The choice of model and the resolution of the mapping depend on the research question being asked and the data available, and must match the station distribution, quality of input data, spatial scale, and local urban conditions. Future efforts should enhance observation coverage and data completeness, increase independent site validation, and test model transferability between cities.