Spatial downscaling of 2-meter air temperature using operational forecast data

Abstract

We present a method for enhancing the spatial resolution of 2 m temperature (T2m) estimates. The method is based on operational forecast data supplied by the European Centre for Medium Range Weather Forecast. From the hourly and monthly average 2-meter temperatures a vertical gradient is determined by linear fitting to the temperature data in larger areas of 1°×1° or 2°×2°. Validation against data from more than 8000 meteorological stations worldwide shows that the estimates of annual average temperature at these points becomes significantly more accurate when applying the vertical gradients to correct the local temperature estimates to the elevation of the stations. When the elevation difference between forecast and station is larger than 300 m, the overall mean absolute deviation of the individual stations bias values decreases from 3.44 to 1.02 °C and the root mean square deviation decreases from 4.11 to 1.42 °C. The gradients have also been applied to the ERA-Interim reanalysis data and the validation results are similar. The vertical temperature gradients will be useful for studies in many fields, including renewable energy and the study of energy performance of buildings.