Modeling the Lower Arctic Atmosphere to Evaluate NOAA Prediction Tools

With the fast rate of warming in the Arctic, the ability to both model and predict cloud formation and weather in the Arctic region becomes all the more essential. Modeling cloud formation in the Arctic poses two unique challenges. First, clouds are a sub-grid phenomenon, and are typically represented in models using a large number of parameters representative of physical processes. Second, atmospheric physics of the Arctic differs from physics represented in mid-latitude forecast systems. In the Arctic, for the majority of the year, clouds function as a source of warming through longwave radiation which depends on both cloud height and phase (liquid water and/or ice). In this study, we analyze the output of the NOAA-ESRL Coupled Arctic Forecast System (CAFS) at Oliktok Point and Barrow, Alaska to identify model biases during the Profiling at Oliktok Point to Enhance YOPP Experiments (POPEYE) campaign. We investigate the differences between predictions and observations of temperature inversions (at the surface and at cloud-top), wind speed, and relative humidity as a function of lead time and height. To gain a better understanding of the underlying relationships between model variables, we implement principal component analysis (PCA) on the surface observations and CAFS model output. We found that over several lead days, the model gains skill in predicting cloud-top inversions, with the main error being the height of the inversion. These results suggest that the model can develop accurate clouds over many lead days, but that the height of a given cloud may differ from observations. By analyzing cloudy days on a monthly basis using PCA, we found that the majority of the observational variance is due to: solar radiation variables (e.g. shortwave downwelling and sensible heat flux), cloud-related variables (e.g. water vapor mixing ratio), and pressure/wind speed. The model and observations are distinguishable in subspaces of these three principal components, indicating a potential bias in model cloud formation due to radiation and/or wind speed/pressure.