The Diurnal Cycle of Winter Season Temperature Errors in the Operational Global Forecast System (GFS)

Forecasts from NOAA's Global Forecast System (GFS) and the High-Resolution Rapid Refresh (HRRR) weather models are matched to surface observations for the winter season of November 2019 to March 2020 at 210 airports across the United States. The 2-m temperature errors, conditioned on observed weather conditions such as cloud cover amount and wind speed, are used to determine the nature of systematic model biases. We observe a strong diurnal cycle in 2-m temperature errors in the GFS in conditions with <= 50% and <= 25% sky cover, with a 1 degrees C warm bias at night and a 2 degrees C cold bias during the day. The HRRR, which uses a different set of physical parameterizations, does not have a clear diurnal cycle in errors under the same conditions. These results highlight the utility of weather-conditional comparisons across the diurnal cycle to diagnose sources of model weaknesses and to target model improvements. Plain Language Summary We evaluate the output from weather forecast models compared to observations at 210 airports across the United States during the November 2019 to March 2020 winter season. We focus on near-surface air temperature errors in the Global Forecast System (GFS) and High-Resolution Rapid Refresh (HRRR) weather models for different times of day and subsets of observed weather conditions. The GFS is 1 degrees C too warm at night and 2 degrees C too cold during the day in conditions with <= 50% and <= 25% cloud cover. The daily high and low temperatures have smaller errors in the HRRR model, which has different algorithms than the GFS model. Model refinement and development efforts would benefit from a focus on accurate representation of the diurnal cycle of temperatures as this basic characteristic of weather can reveal strengths and weaknesses in the model physics. Key Points National Oceanic and Atmospheric Administration (NOAA)'s global forecast system (GFS) model struggles to adequately represent the diurnal cycle of temperatures under observed conditions of <= 50% and 25% cloud cover NOAA's high-resolution rapid refresh (HRRR) model uses a different physics suite and does not have a strong diurnal cycle of temperature errors under the same conditions Examination of errors using similar weather conditions helps to constrain the portion of model physics that can yield larger forecast errors