High-Resolution Estimation of Monthly Air Temperature from Joint Modeling of In Situ Measurements and Gridded Temperature Data

Surface air temperature is an important variable in quantifying extreme heat, but high-resolution temporal and spatial measurement is limited by sparse climate-data stations. As a result, hyperlocal models of extreme heat involve intensive physical data collection efforts or analyze satellite-derived land-surface temperature instead. We developed a geostatistical model that integrates in situ climate-quality temperature records, gridded temperature data, land-surface temperature estimates, and spatially consistent covariates to predict monthly averaged daily maximum surface-air temperatures at spatial resolutions up to 30 m. We trained and validated the model using data from North Carolina. The fitted model showed strong predictive performance with a mean absolute error of 1.61 ∘F across all summer months and a correlation coefficient of 0.75 against an independent hyperlocal temperature model for the city of Durham. We show that the proposed model framework is highly scalable and capable of producing realistic temperature fields across a variety of physiographic settings, even in areas where no climate-quality data stations are available.