Spatial interpolation of urban air temperatures using satellite-derived predictors

Abstract

Air temperatures in urban environments are usually obtained from sparse weather stations that provide limited information with regard to spatial patterns. Effective methods that predict air temperatures (Tair) in urban areas are based on statistical models which utilize remotely sensed and geographic data. This work aims to compute Tair predictions for diurnal and nocturnal time intervals using predictive models that do not exploit information on Land Surface Temperatures. The models are developed based on explanatory variables that describe the urban morphology, land cover and terrain, aggregated at 100 m × 100 m resolution, combined with in situ Tair measurements from urban meteorological stations. The case study is the urban and per-urban area of Heraklion, Greece, where a dense meteorological station network is available since 2016. Moran’s eigenvector filtering and an autoregressive moving average residual specification are implemented to account for spatial and temporal correlations. The statistical models display satisfactory predictive performance, with mean annual Mean Absolute Error (MAE) equal to 0.36 °C, 0.34 °C, 0.42 °C and 0.54 °C, for 11:00–12:00, 14:00–15:00, 22:00–23:00 and 02:00–03:00 (UTC + 2), respectively. The minimum (maximum) MAE for the estimated datasets is 0.22 °C (0.81 °C). The mean annual MAE for all Tair interpolations is 0.42 °C, the mean annual Root Mean Square Error (RMSE) is 0.49 °C and the mean annual bias < 0.01 °C. The time intervals of the analysed measurements coincide with the acquisition times of MODIS and Copernicus Sentinel 3 over Heraklion; hence, the derived estimates can be used in future spaceborne calculations of urban energy budget parameters.