Biases in near-surface meteorological conditions over Alaska within the HRRR model revealed using observations from the U.S. Climate Reference Network

Abstract

The Arctic atmosphere intricately impacts midlatitude weather and climate across diverse timescales. Thus, an accurate representation of the Arctic and its surrounding regions in numerical weather prediction models is critical across many applications, including fire weather, aviation forecasts, marine forecasts, etc. Doing so requires that models properly simulate near-surface radiation, temperature, and moisture fields. However, knowledge of model performance over the Artic is limited largely due to the paucity of observations. To help close this gap, we used five years of observations obtained from 1 Jan 2021 through 31 Dec 2025 from 22 stations comprising the U.S. Climate Reference Network in Alaska to evaluate how well the 1-h through 48-h forecasts from version 4 of the High-Resolution Rapid Refresh (HRRR) model represent incoming shortwave radiation ( SW d ), surface (i.e., skin) temperature ( T s ), near-surface air temperature ( T a ), and humidity (RH) on interannual, seasonal, and diurnal timescales. We found that the HRRR had a net positive SW d bias but, counterintuitively, had negative T s and T a biases during the summer. In the winter, the T s and T a biases were positive. Overall, the T s and T a biases were smallest from 06 – 15 UTC and largest from 18 – 00 UTC. They were also largest on the subset of days with the largest T a changes and SW d . In closing, this work uses a unique data set over a region with few observations to identify biases in a widely-used operational forecasting model. Knowledge of these biases will be important to guiding improvements within future modeling systems.