An evaluation of vertical profiles using a limited area numerical weather prediction model over Baffin Island, Nunavut and surrounding area during autumn 2007

In this century, a large warming trend is expected to occur in the Arctic during the cold season (autumn and early winter) due to loss of summer sea ice, enhanced absorption of solar radiation into the upper-ocean, and subsequent autumnal release of heat to the atmosphere. The Arctic is also projected to see pronounced seasonal changes in precipitation. For meaningful and practical model simulations, leading to detailed analyses of changing weather and climate phenomena, model accuracy assessments are essential. In this study, data from the 2007 Storm Studies in the Arctic (STAR) field project is used to investigate the skill of the Global Environmental Multi-Scale model at 2.5 km grid resolution (GEM-LAM 2.5) in a high latitude mountain environment (Baffin Island, Nunavut in the eastern Canadian Arctic). Emphasis is placed on four specific cases derived from several dropsonde vertical profiles released from a research aircraft. Results provide a description of model errors and proficiencies for different synoptic conditions and surface environments, including orography, open water and sea ice, combined with a general assessment of model performance. Overall the model showed a tendency to overestimated temperature (~1 to 2°C), with the exception of profiles over sea ice where difficulties representing temperature inversions resulted in both positive and negative bias in the vertical profile. The model generally over-predicted moisture (~0.1 to 0.3 g kg^-1), but this was not consistent. Over open water, standard errors for moisture were much larger for cyclonically driven events compared to weakly forced events, and in a high sea ice cover environment the model showed a greater tendency to underestimate moisture content (-0.2 g kg^-1). In some profiles the model also had difficulty with moisture over land in dry layers (too dry). Wind speed was usually underestimated, resulting in model predictions of weaker upslope processes compared to observations. Errors in wind direction were large (> 90°) when sea ice was present and along coastal margins, but in most cases were ≤ 20°. In cases where multiple cloud layers were present, this feature was well represented by the model, but dry layer depth did not always match observations. In addition cloud-tops were usually overestimated (200 to 300 m) and cloud-bases were too low (-50 to -500 m). Model errors were shown to have implications for cloud and precipitation production and their forecast. Large errors occurred above or in proximity to an ice-covered surface and it is important to learn if these errors are a common in the model. Based on evidence from the four case studies, results confirm regional variation in GEM-LAM 2.5 performance in the Arctic, along with variability in its ability to characterize high and mid-latitudes mountainous environments.