Estimating air temperature with high spatio-temporal resolution in urban areas during heatwaves using genetic programming algorithm combined with multi-source datasets

The increasing frequency of heatwave events poses new threats to the health of urban residents. This effect can be exacerbated by the urban heat island (UHI) phenomenon. Air temperature is widely utilised in public health to quantify and analyse nonaccidental mortality attributable to heatwaves in urban areas throughout the world. Therefore, monitoring air temperature at the city level is important for identifying high-risk areas during heatwaves. However, measuring the spatial distribution patterns of air temperature in urban areas is challenging due to the lack of weather stations. The coarse spatial resolution of existing global and regional climate models is insufficient to detect the changes in microclimates, especially in complex-topography areas. In this study, a downscaling method for acquiring the 1-km hourly daytime air temperature data is proposed. It aims to produce a regression model by adopting Genetic Programming (GP) algorithm to estimate air temperature. Using multi-source datasets is considered to combine the advantages of spatial and temporal resolution from different datasets. This research used six weather stations from UK Met Office to assess the regression model obtained from seven satellite- and model-based products. The products consist of six satellite-based datasets retrieved from Aqua Moderate Resolution Imaging Spectroradiometer (MODIS), Terra MODIS, Shuttle Radar Topography Mission (SRTM) and Landsat 8, and one model-based dataset from the newly released ERA5-Land produced by the European Centre for Medium Range Weather Forecasts (ECMWF). The study demonstrates the potential of the proposed model in retrieving high-resolution urban air temperature. The regression model validation showed good results with an R-squared value of 0.992, an RMSE of 0.001 °C, an MAE of 0.322 °C and an NSE of 0.989. The novelty of the study is threefold: (a) unlike previous studies that only estimated the spatial distribution patterns of maximum daily temperatures in urban areas, this study is the first to produce estimations at a one-hour time granularity; (b) it innovatively combines multi-source datasets with GP algorithm to explore possible downscaling models; and (c) it makes the model more reflective of the temperature distribution of extremely hot days than others considering that the regression model is obtained based on data during heatwaves. This study provides a general framework for obtaining hourly air temperature data in urban areas, which could provide theoretical support for heatwave-related decisions. Simultaneously, it can help public health scholars improve the estimation process of mortality caused by heatwave events.