Inside this Issue

• Global Tsunami Model
• Alaska Weather Modeling
• Geochemical Interfaces
• VRcussion
• Clusters at ARSC
• Intern Summer Research Program
• Visualizing Frost Boils
• Permafrost Visualization

Dr. Don Morton, University of Montana at Missoula
Eric Stevens, National Weather Service

Story by Lorien Nettleton

Anyone who has tried to plan a weekend on Alaska’s Denali Highway will tell you the weather forecasts can at times seem maddeningly insubstantial. Even in Fairbanks, the projections can consist of a window of probable temperature, cloud-cover, and perpetual chance of precipitation.

But don’t blame the forecaster – weather prediction in Alaska comes with its own unique challenges. Rugged topography can have consistent small-scale effects on local weather – for example, weather systems in the interior can be significantly modified by the Alaska Range. Complicating matters is a data network for the state that is less dense than those in the Lower 48.

A total of seven Doppler radars cover the landmass in Alaska, whereas more than fifty cover a comparable area in the Lower 48. Weather forecasts made from observations collected at these real-time data input locations can extend into the near future with a fair amount of accuracy, but for anything more than a day or two in the future, forecasters must consider a wide variety of models to make their predictions. These models generally cover large areas of land and ocean, and can provide large-scale weather trends, but they also tend to overlook variation in localized weather conditions caused by physical geographic features. “An example is the local winter temperature inversion,” Stevens says. “You might need resolution as high as one kilometer to illustrate the difference in temperature from the West Ridge of the UAF Campus to Fairbanks International Airport, which are not far apart geographically, but can be dramatically different in temperature.”

With recent experiments using high-resolution, small-scale weather models running on Nelchina, ARSC’s Cray XD1™ supercomputing cluster, new understanding of close-in atmospheric behavior are continually being discovered to better account for the effect of small-scale topographical features.

Dr. Don Morton, Associate Professor of Computer Science at The University of Montana at Missoula, was inspired to establish high-resolution regional weather models to capture important small-scale dynamics of complex environments when, as a pilot in Missoula, he became familiar with a persistently violent wind coming out of Hellgate Canyon. While the Hellgate Winds are a constant obstacle to local pilots, they are rarely represented on numerical weather forecasts.

Using the Weather Research and Forecasting model (WRF), a modeling product produced by the Mesoscale Meteorology division of NCAR, Morton has experimented with coupling the small-scale forecasts with the larger atmospheric models in the Missoula region. Most numerical weather models support domains on a scale of a few meters to thousands of meters, and by nesting a small domain within a larger model, the computational load to drive its simulation is reduced.

Morton has also applied the WRF model to an Alaska test region, centering a three-kilometer resolution grid near Delta Junction nested within a nine-kilometer grid over northern Alaska. One of the aims of this specific application is to test the ability of the model to predict known local weather phenomena, specifically the two significant wind regimes at Delta Junction. The first is the Tanana Valley Jet, a frequently violent gust of katabatic, or down-hill glacier wind, that bullets down the Tanana Valley from southeast to northwest. The second is the Chinook wind, which pours out of Isabel Pass in the Alaska Range to hit Delta Junction from the south. In the winter season the Tanana Valley Jet is a cold wind, while the Chinook often brings warmer temperatures. Both wind regimes make life difficult for pilots and anyone who simply prefers calm air. Though they are regular weather patterns, the Tanana Valley Jet and the Chinook are not predicted by current numerical weather forecasts.

Morton’s aim is to explore the possibilities of high-resolution predictive numerical weather modeling, to see if a more detailed view of a small area can help generate a feedback loop to inform the big picture. His experiments with developing a weather model at a detail high enough to discern hills from valleys and the subsequent weather behaviors they produce are ongoing. Typical weather models have resolutions that are fairly coarse — about 40 kilometers, allowing them to cover larger areas with fewer grid points, reducing computational load. Because the WRF architecture allows users to set up parameters with resolutions as high as hundreds of meter grids, the possibility for high-resolution models over large areas would require considerable computing resources to run. Based on Morton’s experiments, some scientists speculate that a nested grid like the one used to predict the Tanana Jet could be applied to the entire state of Alaska using high-resolution regions where required. Greg Newby, Acting Chief Scientist for ARSC, sees potential to utilize ARSC resources to drive such a computationally efficient model, making the results available as a National Weather Service forecasting tool.

“We’d like to someday see a regular forecast powered by computers at ARSC incorporated into the set of models from which NWS makes their Alaska forecasts,” Newby says.

The Present

“Numerical Weather Prediction (NWP) Models are an important part of how forecasters put together their predictions,” says Stevens. “They’re not the only tool, but they are an important one, and the models are improving all the time as computers get more powerful and data processing becomes faster.”

But even models are by no means the ultimate in predictive accuracy when it comes to something as given to apparently whimsical change as the atmosphere. Current weather predictions are commonly a combination of real-data and models. Model output is displayed on a map, and NWS forecasters are able to modify the results based on knowledge of the landscape and their experience.

It is up to the forecaster to interpret the model output, and along with satellite, radar and weather observations, adjust model predictions for wind, surface pressure or precipitation to fit known local weather effects.

“Someone with one watch knows what time it is. Someone with two watches can never be sure,” says Stevens. “This is what’s happening with weather models. We have access to several American models, along with European models, British models and Canadian models, and over time their solutions tend to diverge. The role of the human is to determine which model to use in each case, or to sort through results and come up with one unified forecast.”

Benefits

Detailed forecasts could benefit United States Department of Agriculture Forest Service Fire and Aviation Management personnel by providing advanced wildfire fighting tools. Currently, fire prediction models are run on a laptop and generate wind information in mountains and valleys using grided wind data from coarse-resolution models. Accurate, localized forecasts of complex weather patterns, coupled with the weather generated by wildfires can be combined with realistic, high-resolution wind information to give firefighters added information and assist them in battling the burns.

The widespread development of detailed, high-resolution weather models would also open the door to greater accuracy in other research. Many studies depend on atmospheric weather models as a component of their research, and a complementary weather model of substantial detail would further refine these studies.

Dr. Kate Hedstrom, Oceanographic Specialist at ARSC, and her colleague Enrique Curchitser with Lamont-Doherty Earth Observatory of Columbia University, have been developing a coupled ice and ocean model of the Bering Sea that incorporates wind, temperature, precipitation and evaporation to provide predictive conditions of ice and ocean currents in key fishery areas. One gap in the accuracy of the National Climate and Environmental Prediction (NCEP) model they use is weather.

The Future

There is a debate in the operational forecast community about how to best apply the increasing computational power of today’s supercomputing technology. The first argument is to make the models run at a finer resolution, enabling forecasters to analyze the complex weather-generating topography so characteristic of Alaska and other states with rugged, mountainous topography. The second strategy, known as ensemble forecasting, instead runs coarser resolution models many times, but with slightly different initial conditions. Once complete, the outputs can be examined for similarities, which will allow the researchers to predict weather patterns with more certainty.

“If you run a model 100 times, and 80 of those results are close to the same, you can have reasonable confidence that the result would make a fine prediction,” Stevens says. “On the other hand, if the 100 solutions are all wildly divergent, we infer that noise in the initial conditions is diminishing the predictability of the atmosphere’s behavior. Confidence in a forecast made under such conditions is usually low.”

According to Stevens, the National Weather Service’s forecasts have moved toward a single highly detailed forecast over the last few years. Meteorologists still consider ensemble output when producing forecasts, but the forecasts themselves reflect the deterministic, high-resolution, single-model approach. In this paradigm shift, models such as the one Morton is deploying will play a larger role.