Analysis of Persistent Bias and Suggested Improvements in Forecasting Temperatures Patterns over Canaan Valley in West Virginia with the National Blend of Models

Complex terrain offers a unique challenge for numerical weather prediction (NWP) models, as the dynamical and physical effects of terrain on diurnal flow regimes can drastically influence the distribution of state variables such as temperature. High variability in terrain within mountainous regions can lead to sharp gradients in these variables, which must be accurately modeled at a high resolution to avoid significant errors in NWP model forecasts.

One particular cause of strong temperature gradients is cold pooling, in which cold air sinks into a basin protected from surrounding higher elevations to create potent nocturnal inversions. Cold pooling can also occur due to radiational cooling on clear nights with calm surface winds, as minimal turbulence removes an otherwise limiting factor to the extent of cooling overnight. This process is particularly effective with relatively low dew points or the presence of powdery snow cover.

Previous research has been extensively conducted regarding cold pools over the Western United States. A case of particular interest has been Peter Sinks, UT, a natural sinkhole with a diameter of about one kilometer, where temperatures have dropped below -50°C in the most extreme cases. However, research is limited for similar weather patterns east of the Mississippi River due to the generally lower relief of the topography.

One of the strongest sites for cold pooling in the Eastern United States is Canaan Valley, WV. With a basin size of about 80 square kilometers and a floor elevation greater than 3,100 ft, it is the valley with the highest elevation in eastern North America. Surrounding ridges over 4,000 ft in elevation allow for cold pooling on clear, calm nights, resulting in overnight temperatures much colder than areas immediately surrounding the valley floor. The Valley receives hundreds of thousands of annual visitors, highlighting the importance of understanding diurnal temperature patterns over the area.

This study seeks to characterize the various temperature patterns over Canaan Valley with respect to forecast grids from the National Blend of Models (NBM). Often used for operational forecasts, the NBM is a post-processing system that combines objectively analyzed observations with gridded model output statistics (MOS) and raw NWP model output. NBM has historically struggled with representing Canaan Valley temperatures, with the most extreme cases exhibiting a warm bias of over 15°C. Observations were gathered from a published National Weather Service (NWS) Cooperative Observer (COOP) weather station and several professionally sited and equipped weather stations maintained by Virginia Tech University.

A previous study we conducted in 2022 characterized discrepancies in two case studies of overnight Canaan Valley cold pools with respect to biases in the High-Resolution Rapid Refresh (HRRR) model and the UnRestricted Mesoscale Analysis (URMA) used for NBM bias correction. It was found that most observations from the northernmost valley floor station were assimilated into URMA, but the lack of representation in the gridded output suggests that NonLinear Quality Control (NLQC) applied a low weight to the station.

This research expands on the previous results, offering various case studies of Canaan Valley temperatures with good and poor performances by the NBM and URMA. This study also investigates discrepancies within NWP model forecasts from HRRR and offers a characterization of these cold pool environments with respect to synoptic flow regimes. The primary objective of this research is to synthesize these recent cases and offer potential improvements for the handling of future Canaan Valley cold pool events by the NBM and URMA.

Preliminary investigations indicate that the HRRR, a component to both the NBM and URMA, also struggles to represent cold pool events over the Valley. However, observation analysis and subsequent bias correction in the NBM should ideally correct this initial bias to produce a more refined product. The current NLQC algorithm within URMA appears to ignore the station at the center of the valley floor and favor stations surrounding the Valley based on the results of the previous study. As such, one possible solution could be to adjust the algorithm for NLQC to account for terrain variability, as the centrally located station is nearly 1,000 ft. lower in elevation compared to nearby stations on the Valley rim. Another potential solution might be to conduct post-analysis of daily low temperature observations from the COOP station located in the southern, more developed end of the Valley, in addition to including DY007.