Evaluating the Rapid Refresh Numerical Weather Prediction Model in the Arctic

The Rapid Refresh (RAP) model is a rapidly updating model that has been run operationally over portions of the Arctic since 2012, but its performance has never been extensively analyzed at high latitudes. In this project, we evaluated version 4 of the RAP from July 2017 through June 2019 against observations at three Arctic locations: the ARM sites at Utqiagvik (formerly Barrow) and Oliktok Point in Alaska and at Summit Station in the center of the Greenland Ice Sheet. Using data from radiosondes and surface stations, we investigated the model’s skill in predicting surface conditions, vertical meteorological profiles, and surface radiation fields at forecast hours 1-18, as well as at initialization (hour 0). While the winds profiles generally agreed with the observations, we discovered significant seasonal biases in boundary layer thermodynamic fields at the Alaskan coastal sites, due to the presence or absence of sea ice. Looking into the RAP’s representation of surface-based inversions, we found that the RAP usually captures this feature well, but is unable to represent very sharp inversions, instead generating deeper inversions with more conservative lapse rates. Further, we identified issues in the model’s simulation of clouds, likely owing to the complex mixed-phase nature of many Arctic clouds. Interestingly, the change in cloud biases with forecast hour was relatively small, which could suggest that the issues are more associated with model initialization. These findings highlight the need to continue to improve the RAP physics to accurately simulate Arctic phenomena so that the model can provide more accurate predictions in the region.