Near real-time mapping of air temperature at high spatiotemporal resolutions in Tasmania, Australia

Abstract

Surface air temperature (Ta) is important for a wide variety of applications that require real-time monitoring of environmental change; yet, few studies have attempted to deliver such products at a resolution appropriate for local-scale analysis. We investigated the feasibility of producing hourly Ta maps in near real time at a spatial grid resolution of 80 m across the state of Tasmania (area: 68,401 km2), Australia. We first assimilate 267 non-telemetered logger recording sites for data recorded in the 1-year period from September 2013 to 2014 and statistically calibrate them to 43-telemetered Australian Bureau of Meteorology (BoM) weather station sites for real-time application. This was evaluated in the following year using held back recordings in addition to evaluating regression trees (RT) and thin plate splines (TPS) to interpolate the hourly Ta estimates. For real-time operational mapping, the system was fully automated in the R programming language and hosted on a cloud-based computing platform to assess performance over a 7-day period in February 2020. The calibration procedure yielded accurate results with the root mean square error (RMSE) ranging between 1.33 °C in summer and 1.29 °C in winter. The TPS interpolation method was optimal in summer with an RMSE of 1.35 °C compared to RT with 1.39 °C; however, RT performed better in winter with an RMSE of 1.34 °C compared to 1.42 °C. The mapping system was capable of producing spatial outputs within the hour of the BoM observations becoming available with the TPS interpolation proving to be more efficient at producing outputs in a timelier manner.