Towards an operational model for estimating day and night instantaneous near-surface air temperature for urban heat island studies: outline and assessment

Near-surface air temperature (NSAT) is key for assessing urban heat islands, human health, and well-being. However, a widely recognized and cost- and time-effective replicable approach for estimating hourly NSAT is still urgent. In this study, we outline and validate an easy-to-replicate, yet effective, operational model, for automating the estimation of high-resolution day and night instantaneous NSAT. The model is tested on a heat wave event and for a large geographical area. The model combines remotely sensed land surface temperature and digital elevation model, with air temperature from local fixed weather station networks. Achieved NSAT has daily and hourly frequency consistent with MODIS revisiting time. A geographically weighted regression method is employed, with exponential weighting found to be highly accurate for our purpose. A robust assessment of different methods, at different time slots, both day- and night-time, and during a heatwave event, is provided based on a cross-validation protocol. Four-time periods are modelled and tested, for two consecutive days, i.e. 31st of July 2020 at 10:40 and 21:50, and 1st of August 2020 at 02:00 and 13:10 local time. High R2 was found for all time slots, ranging from 0.82 to 0.88, with a bias close to 0, RMSE ranging from 1.45 °C to 1.77 °C, and MAE from 1.15 °C to 1.36 °C. Normalized RMSE and MAE are roughly 0.05 to 0.08. Overall, if compared to other recognized regression models, higher effectiveness is allowed also in terms of spatial autocorrelation of residuals, as well as in terms of model sensitivity.