Interpolation, Machine Learning, or Meteorological simulation? A comparison analysis for spatio-temporal estimation of meso-scale urban air temperature

Fine resolution spatio-temporal maps of near-surface urban air temperature (<i>T</i>_<em>a</em>) distribution provide crucial data inputs for sustainable urban decision-making, personal heat exposure, and climate-relevant epidemiological studies. However, due to the low density of reference meteorological stations, data at high spatial resolution that accounts for the complexity of the urban built environment is not always available. Recent availability of IoT weather station data now allows for high resolution urban <i>T</i>_<em>a</em> mapping using Machine Learning (ML). Here, we employ a network of NetAtmo crowd-sourced weather station data and XGBoost Gradient Boosting algorithm to predict/map daily <i>T</i>_<em>a</em> at nearly 1 km spatial resolution in Warsaw (Poland) during warm months (Jun - Sep). The output from the ML approach was used to map sub-daily (four times per day) variability in <i>T</i>_<em>a</em> for Warsaw at nearly 1 km² spatial (0.008°) resolution. See the related materials, describing the details of collecting/producing the data.