Abstract Despite significant advancements in NWP since its inception in the mid-twentieth century, numerous error sources still exist. Consequently, forecasts often exhibit systematic biases and lack sufficient accuracy. Postprocessing seeks to enhance the raw NWP forecasts by leveraging available observational data. To address the lack of observational data in many locations, methods using a diagnosed temperature lapse rate from multiple model grid points to correct the 2-m temperature bias due to the altitude difference between the model and the real terrain have been tested. As a step further, we introduce the neighborhood ensemble altitude (NEA) correction method, a novel approach that enhances the 2-m temperature forecasts without relying on observations. NEA involves finding the closest model point to the location of interest (LoI), forming a neighborhood ensemble, calculating the temperature lapse rate through linear regression of neighboring points’ altitude and lowest model-level temperature, and applying this lapse rate to all neighboring points to obtain altitude-corrected 2-m temperatures. The ensemble mean of these corrected 2-m temperatures forms the final forecast, accounting for the model uncertainty due to the inability to resolve all processes at the model grid-size scale. The method was thoroughly tested and validated using 1 year of 36 surface measurements in Croatia. Results are encouraging and show an increase in forecast accuracy of about 10% on average when compared to the default method of using the nearest model land point. Moreover, NEA’s performance is comparable to the advanced analog-based method, especially during daytime, making it a robust solution for diverse terrains and operational settings.
Keresturi et al. (Tue,) studied this question.