Regional Environmental Prediction over the Pacific Northwest


Clifford F.  Mass1, Mark Albright, David Ovens, Richard Steed, Eric Grimit and Tony  Eckel

Department of Atmospheric Sciences,

University of Washington,

Seattle, Washington 98195


Brian Lamb and Joseph Vaughan

Laboratory for Atmospheric Research

Washington State University

Pullman, Washington


Kenneth Westrick and Pascal Storck

3-Tier, Inc.

Seattle, Washington


Brad Colman and Chris Hill,

National Weather Service

Seattle, Washington


Naydene Maykut and Mike Gilroy

Puget Sound Clean Air Agency

Seattle, Washington


Sue Ferguson

USDA Forest Service Research Station

Seattle, Washington


CDR Joe Yetter and John M. Sierchio,

Naval Air Station Whidbey Island

Oak Harbor, Washington


Clint Bowman and Dick Stender

Washington State Department of Ecology

Olympia, Washington


Robert Wilson

U.S. EPA Region X

Seattle, Washington


William Brown

Washington State Department of Transportation

Seattle, Washington


Submitted to the Bulletin of the American Meteorological Society

August 2002



1 Corresponding author: 

Department of Atmospheric Sciences,

Box 351640,

University of Washington,

Seattle, Washington 98195



            This paper examines the potential of regional environmental prediction by focusing on a local forecasting effort in the Pacific Northwest.  A consortium of Federal, State, and local agencies have funded the development and operations of a multifaceted numerical prediction system centered at the University of Washington that includes atmospheric, hydrologic, and air quality models, as well as the collection of all available real-time regional weather data sources .  This paper reviews the Northwest modeling and data collection systems, describes the funding and management system established to support and guide it, and examines the value of regional environmental prediction.



1.         Introduction


            An important issue for the weather and environmental prediction communities is how modeling and associated activities should be organized.  Should environmental prediction be centralized at a few national centers, decentralized at local forecast centers close to the user communities, or some combination of the two?  This issue has become particularly timely as rapidly increasing local computer resources, the availability of state-of-the-art models, and increasing access to observational and model data over the Internet makes local environmental prediction increasingly viable (Mass and Kuo 1998).   Today (August 2002) approximately two-dozen sites in the U.S. are running mesoscale atmospheric models operationally (see for a partial listing), and the use of real-time air quality and hydrological models is rapidly increasing.

            One of the first regional numerical prediction efforts, and perhaps the most extensive, is found at the Department of Atmospheric Sciences at the University of Washington (UW).  Initiated in 1995 as a single-domain forecast system applying the Penn. State/NCAR mesoscale model version 5 (MM5) with 27-km grid spacing, the UW enterprise has grown into a regional environmental system that includes atmospheric, hydrologic and air quality real-time prediction down to 4-km resolution, as well as real-time access to many telemetered observational networks in the Pacific Northwest.  A significant contributor to the success of the Northwest modeling effort has been the management and funding by the Northwest Modeling Consortium, a collection of Federal, state and local agencies.  This paper will review the scope and approach of the Northwest regional modeling effort, tracing its development into an integrated and highly multidisciplinary environmental prediction system.  An appraisal of the Northwest modeling and data collection effort is provided, and its implications as a national model are examined.


2.         History of the Northwest Regional Prediction Effort


            The Northwest regional prediction effort began in the early 1990’s when a group of Northwest air quality and weather prediction agencies identified the lack of upper air observations over Puget Sound as a major problem for diagnosing and predicting local weather and air quality.  Under the chairmanship of Naydene Maykut of the Puget Sound Air Pollution Control Agency, a Northwest Upper Air Committee was formed and proceeded to identify the Radian 915-Mhz radar wind profiler as a possible solution.  The group then devised a novel approach to funding:  support in terms of dollars or other assets (land, management time) by a “consortium” of agencies.  (A listing of current Consortium members is found in Table 1.)  The profiler was purchased in 1992 and remains operational to this day.  At roughly the same time, Mark Albright, a UW staff member and Washington State Climatologist, began constructing a regional real-time weather database for analysis and forecasting by collecting data from several Northwest weather observation networks into one computer system.  In such a way a relatively dense mesoscale network could be built at little cost, while coordination between different networks reduced duplication of effort.  By 2002 the regional observational database (NorthwestNet) has grown to a collection of over a two-dozen networks, including nearly a thousand stations over the Northwest (more details in the next section). 

            During the late 1980’s and early 1990’s the lead author and several of his students began research simulations of regional weather features of the west coast of North America using the Colorado State RAMS and Penn. State/NCAR (MM5) mesoscale models.  Running with grid spacing down to 5 km, it was found that such mesoscale models could produce highly realistic mesoscale circulations, particularly those driven by orography, if the synoptic forcing was accurate.  By 1994, relative fast single processor UNIX workstations became available, making it possible to run regional domains at much higher resolution than used at national modeling centers such as NCEP (at the time the NCEP Eta model was being run at 80-km resolution).  Based on the promising research runs, the Northwest Upper Air Committee (soon to be renamed the Northwest Modeling Committee), decided to support the evaluation of local numerical weather prediction (NWP).   The initial results for coastal surges, Puget Sound convergence zones, and local diurnal circulations completed by Jim Steenburgh (then a UW postdoc) were so promising, that in 1995 real-time prediction using a single 27-km domain of the MM5 (with initialization and boundary conditions from the Eta model) was begun using a single processor Alphaserver 250.  The value of the real-time MM5 prediction system became clear during the next year, as it successfully forecast important regional circulations (such as onshore pushes and coastal surges) for which the Eta model lacked sufficient resolution.  A notable early  success was the accurate prediction of a major windstorm on 12 December 1995, for which the local MM5 simulation was closely duplicated the low’s intensity and track. 

Previous research runs had determined that realistic simulation of the major mesoscale features of the Northwest required a grid spacing under 15 km.  These results, and the clear value of the 27-km runs, inspired the Northwest Modeling Committee to support a jump to far higher resolution.  With funding from a large collection of agencies (which became known as “The Consortium”) and with an exceptional discount provided by Sun Microsystems, the UW purchased a SUN 4000 server with 14 processors during the summer of 1996.  Using this powerful new system, a new grid configuration was initiated with a large 36-km domain over the eastern Pacific and western North America, and a nested 12-km grid over the entire Pacific Northwest.  With the acquisition of upgraded processors the following year, an additional 4-km nest was added over western Washington, making the Northwest effort the highest resolution NWP effort in the U.S. for a several years.   An additional computer (4-processor Alphaserver ES-40) was acquired in 1999 that was used to expand the 4-km to include the entire State of Washington.   During the summer of 2000, a new SUN 6500 server was acquired with 23 processors, allowing the expansion of the 4-km domain to include all of Oregon and Washington, the improvement of model physics, and the expansion of the forecast period to 60 h.  Computational power was further enhanced with the acquisition of 7 additional processors for the 6500 and the arrival of some additional computers for pre and post processing.

            The operational evaluation of the UW high-resolution forecasts made it clear that the largest source of prediction error was poor initialization over the Pacific.  To get a handle on this initial condition uncertainty and to explore the potential for probabilistic forecasts, MM5 ensemble forecasts were initiated later in 2000 and continue (considerably expanded) today.  A key aspect of the UW ensemble system has been its use of gridded analyses and forecasts from major NWP centers for initialization and boundary conditions (see below for more details).  In order to get real-time guidance of the quality of the initialization over the Pacific, available offshore observations have been collected and compared, both graphically and numerically, to the initial fields of major forecasting models.  Described below, this Pacific Initialization System has been available on the web since 1998.

            During the last few years, the Pacific Northwest prediction effort has grown well beyond atmospheric modeling and diagnosis.  In 1998, the MM5 was coupled to a distributed hydrological model (DHSVM) run at very high (100 m) resolution over the Snohomish River drainage in an attempt to forecast river flow rates.  This work (completed by Ken Westrick) was so successful that the real-time hydrological prediction system has been expanded to all the major rivers draining into Puget Sound (more information below).  The Northwest effort has also turned to predicting air quality and the dispersion of smoke.  In cooperative with Brian Lamb’s air quality modeling group at Washington State University and the EPA-sponsored AIRPACT (Air Indicator Report for Public Awareness and Community Tracking) effort, 4-km MM5 forecasts have been coupled to the CALGRID air quality model over western Washington to provide predictions of ozone, nitrogen oxides, and other species of interest. 


3.         Major Components of the Northwest Environmental Prediction System       


            This section describes the major components of the Northwest real-time regional prediction system that is resident at the University of Washington.  As shown in Fig. 1 the Northwest system has three major levels.  The top level includes all the observational and model inputs required by the regional models and application programs.  The middle level encompasses the local modeling systems  (MM5, DHSVM, CALGRID, Ensembles). The final level includes all the web pages and applications through which the model output and observations are provided to a diverse user community.  A web portal to all components of the Northwest Environmental Prediction System is found at


Figure 1:  Schematic of the Northwest Environmental Prediction System


Observational and Model Inputs


            As noted above, a major Northwest initiative since 1992 has been to collect observations (mainly at the surface) from real-time meteorological networks in the Northwest into one operational database, termed “NorthwestNet.”  Spearheaded by UW’s Mark Albright, all available observations that can be accessed in real-time (or near real-time) are brought into a UW Atmospheric Sciences server, where they are decoded, quality-controlled, placed on hard disk for several weeks to several years, and later archived on tape.  An example of the coverage made possible by combining many data sources is found in Fig. 2, which shows surface observations over Washington State that are available in or near real-time.  Table 2 lists the current networks being ingested into NorthwestNet.  This idea of a network of observational networks has recently been taken up by other groups, including the MesoWest network centered at the University of Utah.  In addition to surface observations, the UW effort also gathers all regional upper air data, including radiosonde soundings, the Seattle 915-Mhz profiler temperatures and winds, and ACARS aircraft observations—which is becoming an extraordinary rich source of mesoscale data aloft.  Other data sources include all WSR-88D radar data and satellite imagery for the region.


Figure 2.  NorthwestNet observations over Washington State.


            For initialization of the real-time MM5 forecasts, including several dozen ensemble runs, gridded analyses and forecasts are ingested operationally from a number of major prediction centers such as NCEP (ETA, AVN, MRF model output), the Canadian Meteorological Center (GEM), the Australian Bureau of Meteorology (GASP), the Taiwan Weather Bureau, and Fleet Numerical Oceanography Center (NOGAPS).  Most of these data sets are acquired through ftp servers.  For the NCEP products a redundant feed uses the UNIDATA CONDUIT system in which model grids are distributed over the Internet through a few major sites using the UNIDATA Local Data Manager (LDM) system.


Computational facility


            As noted above, the Northwest environmental prediction system involves operational integration of a number of models run at high resolution, with the preparation of thousands of graphical and other products each day.  The computational demands have been corresponding large, necessitating the creation of a large computer facility.  Table 3 lists some of the major computational assets available for the Northwest modeling work.  The main computer resources include a 30-processor SUN 6500 server, a 20-processor Athlon (1.2 Ghz) Linux cluster, a 32-processor Athlon (1.5 GHz) Linux cluster, over 6 terabytes of RAID disc storage, two four-processor servers for integration of the hydrological and air quality models, and four additional machines for pre and post processing of model data and graphics generation.  The excellent scalability of the MM5 on large numbers of processors has been a major factor in allowing high-resolution prediction over the Northwest.   SUN Microsystems has been particularly helpful to the effort, providing extraordinary educational discounts.  Harry Edmon and Dave Warren, the department system programmers, have played a critical role in building and maintaining this heterogeneous facility.


Regional Mesoscale Modeling Systems


a.         MM5 Atmospheric Model


            The centerpiece of the Northwest regional prediction effort has been the Penn. State/NCAR mesoscale model, version 5—commonly known as MM5.  The current configuration includes three domains:  an outer domain with 36-km grid spacing that extends several thousand km upstream into the Pacific, a 12-km resolution domain that includes the entire Pacific Northwest, and a 4-km domain that covers Washington, Oregon and portions of British Columbia, Idaho, and California (Fig. 3).  Using 38 levels, the UW real-time system is run twice a day, being initialized and deriving boundary conditions from the NCEP Eta model’s operational forecasts.  This type of “cold start,” without any local data assimilation, was used after tests found that local data assets only improved forecasts during the first few hours.  The UW MM5 forecasts run for 60 hours for the 36 and 12-km grids and for 24 h (12 to 36h) over the 4-km domain.  Cumulus parameterization (Kain-Fritsch) is only applied in the outer domains.  The MM5 output is verified operationally against the observations collected in NorthwestNet. More information about the UW high-resolution MM5 runs can be found at its web site (

Figure 3:  Models domains for the Northwest real-time MM5 forecasts run at the University of Washington.  The grid spacing is 36-km for the outer domain, 12-km for the middle domain, and 4-km for the inner domain.


b.         UW Mesoscale Ensemble Forecast System


As noted above, the Northwest regional modeling effort has evaluated two complementary approaches to regional mesoscale prediction:  high resolution MM5 forecasts down to 4-km grid spacing, and an ensemble approach in which the MM5 is run multiple times using different initializations and model physics.  The UW ensemble system, built by graduate students Eric Grimit and Tony Eckel, with considerable inspiration from the National Weather Service’s Brad Colman, is based on running the 36-12 km MM5 domains multiple times using the initializations and boundary conditions from a number of operational modeling systems (e.g., NCEP Eta and AVN models, Navy NOGAPS model, Canadian GEM model, UKMET Office global model, Australian GASP model, Taiwanese global model).  The central idea is that the variation in the initializations of major modeling systems is a good measure of initialization uncertainty.   Additional members of the UW mesoscale ensemble system are created by varying model physics parameterizations (microphysics, boundary layer schemes, moist physics) and surface properties (variations of sea surface temperature and soil moisture within observational error), and the addition of initialization “mirrors,” whereby particular initializations are reflected around the ensemble mean (see web site below for more details).  Currently there are 40 members of the UW ensemble system.  This ensemble work has been facilitated by the purchase of a relatively inexpensive Linux clusters over which the ensembles can be efficiently and rapidly computed.  Operational for two years, the initial results of the UW system are reviewed in Grimit and Mass (2000), and daily forecasts are found on the web at (  On the web site, both the individual ensemble members and derived products (e.g., ensemble spread, probabilistic forecasts) are provided (see Fig. 4 below for an example).



Figure 4:  Six-hour precipitation(shading) and sea level pressure (blue isopleths) for 24 h

into a forecast from the UW mesoscale ensemble system.






c.  Regional hydrological prediction system


            Beginning in December 1997, 12 and 4-km output from the MM5 has been used to drive a fully distributed hydrological model (DHSVM, Distributed Hydrological Soil Vegetation Modeling System) that was developed by Professor Dennis Lettenmaier and students at the UW Civil Engineering Department.  During the last five years the number of simulated watersheds has increased from 1 to 26, encompassing most of the river basins in western Washington State (Fig. 5a).  Running at 150-m resolution, the real-time streamflow forecasts are made daily for up to 60 h, using explicit channel routing that provides streamflow at any point in the river networks. Supported by the National Weather Service and UW PRISM[1] programs, as well as Seattle City Light, the hydrological predictions are both accessible over the web and distributed electronically to the National Weather Service forecast office in Seattle.   For over a year, the UW Hydro system was driven by the ensemble forecasts as well, providing a collection of potential hydrographs at any location.  An example of a recent forecast hydrograph product for one site (Tolt River near Carnation) is found in Fig. 5b.  This graphic shows several hydrographs from recent consecutive runs, as wall as the observed strreamflow.   The UW hydrological prediction system has been built and maintained by Ken Westrick and Pascal Storck, previously UW students and staff members, who started a commercial venture (3-Tier, Inc) that offers a range of hydrological and energy related services.   More information on the UW hydro effort can be found on the hydro web page ( or in recent publications (Westrick et al, 2002; Westrick and Mass, 2000)





Figure 5:  Watersheds currently modeled operationally in the UW coupled hydrological prediction system (a).  Sample hydrographs showing streamflow on the Tolt River near Carnation, WA (b).



d.         Smoke and Fire Guidance


The suppression of wildfires, as well as planning and control of proscribed burns in National Forests requires detailed meteorological guidance.  To provide such information the USDA Forest Service has provided funding for the maintenance of the Northwest atmospheric modeling as well as support for the creation of a wide range of fire and smoke related products driven by the MM5 forecasts and regional data assets.  Many of these products can be accessed through the Smoke and Fire web site shown in Fig. 6a  (  This site provides graphical interfaces to viewing MM5 forecasts, including meteograms and soundings at locations around the Northwest.  It also provides forecast fire potential guidance that is driven by MM5 output, such as the Haines and Fosberg fire indices, and ventilation indices that combine MM5 winds, stability, and boundary-layer depths (see Fig. 6b below).   Another project has been to interface the MM5 grids with lagrangian “puff” models (such as CALPUFF and NFSPUFF) for predicting smoke distributions from wild and proscribed fires.






Figure 6.  Smoke and Fire web page that provides access to a wide range of fire and smoke related products, many of which are driven by the UW MM5 forecasts (a).  Ventilation index (18-h forecast) based on MM5 surface winds and low-level stability for 1200 UTC 28 February 2002



e.         Regional Air Quality Prediction


            Since the spring of 2001, real-time air quality forecasts have been made over western Washington by using MM5 forecast grids to drive a Eulerian photochemical air quality model (CALGRID).  This system has been built by Joseph Vaughn, Brian Lamb, and colleagues of Washington State University in cooperation with the UW Atmospheric Sciences MM5 group.  Support for this effort has come from the EPA EMPACT program and EPA Region X.  The coupled CALGRID/MM5 system is run once a day at 5-km grid spacing for 24-h using hourly gridded emissions data provided by the Washington State Department of Ecology.  MM5 forecasts are run through the CALMET meteorological processor before CALGRID integration.  Graphical display of the emissions data and CALGRID forecasts are available at the project web site (  An example of a CALGRID ozone forecast is found in Fig. 7.


Figure 7:  16-h forecast from the coupled CALGRID/MM5 system developed by Washington State University.  High levels of ozone are forecast southeast and south of Seattle.



f.          Road Weather Information System



            The Washington State Department of Transportation (WSDOT) has supported the development of a web-based system that uses the regional observational database (NorthwestNet) and the UW MM5 forecasts to provide guidance for WSDOT personnel and the traveling public.   In addition, the Oregon State land surface model is run over major highway routes to provide forecasts of road surface temperatures.  Accessed through Road Weather Information web portals ( or, the user can view real-time observations in either map form or can view the weather and road conditions along a particular highway section.    For example, Fig. 8 shows the web page for one highway segment: Interstate I90 that crosses the central Washington Cascades from Seattle to Ellensburg.  Such pages provide an up-to-the-minute view of weather conditions and road surface temperatures;  clicking on the cross section at any point provides the appropriate NWS forecast.  In addition, the use can move forward in time (using the MM5 forecasts) to view predicted highway and weather conditions. As part of the project, several dozen additional weather sensors have been placed along state highways and on Washington State ferries that cross the inland waters of the state. A web page that displays ferry weather and nearby observations are shown in Fig. 9.


Figure 8:  I90 Travel Route Information web page.  The top panel shows cams across the Cascades, while the lower panel provides real-time weather observations and road surface temperatures.  In addition, future conditions across the mountains, driven by the UW MM5, can be viewed on this page.



Figure 10:  The Ferry Weather Page that show weather observations along ferry routes as well as nearby land-based data.



g.  Pacific Initialization and Forecast Verification System


            Mesoscale model forecasts are only as good as the large-scale guidance used for their initialization and boundary conditions.  Thus, a crucial aspect for any mesoscale forecast system is the development of software for appraisal of the initialization and forecast fidelity of the synoptic scale model in which it is embedded.  Such appraisal is of particular importance for the Northwest, where poor initializations over the data-sparse Pacific often produce unrealistic synoptic forecasts that can greatly degrade mesoscale model skill over the western U.S.  To provide information about the quality of Pacific initializations and forecasts for a number of major modeling systems (e.g., NCEP ETA, AVN; Navy NOGAPS, Canadian GEM models), an operational Pacific verification system was developed by a UW student (Brett Newkirk) and  UW Research Scientist Lynn McMurdie.  This system takes in a wide range of Pacific observations (buoy and ship reports, ACARS flight level data, land surface sites, cloud and water vapor track winds) and interpolates model initializations and forecasts to the observation sites.  This information is used to create graphical comparison products (as shown in Fig. 10), tabular output, and long-term error summaries for model initializations and short-term (up to 48-h) forecasts.  Thus, this system allows an analyst to evaluate the skill of model initializations and recent forecasts in an efficient manner, and to learn which models are typically most skillful in the area of interest.  The real-time Pacific Verification Web site is found at


Figure 10:  Initialization verification at the surface for 1200 UTC 4 August 2002 from the NW Verification web page.  Observed and model (NCEP eta) surface winds are shown by green and red barbs, respectively.  Also shown in model sea-level pressure (blue isopleths) and the simultaneous GOES West IR image.  Yellow numbers give the pressure error at each observing site.


4.         The User Community and Commercial “Spin-offs”


The user community for the Northwest environmental prediction effort has grown rapidly over the past five years, with the MM5 forecasts being used operationally by local National Weather Service offices, military forecasters, private sector and media meteorologists, air quality and transportation interests, as well as recreational users—to name only a few.  The forecasts from the MM5 and other northwest modeling systems (ensembles, hydrologic and air quality) are disseminated  through web pages as well as Internet access to model grids and extracted model soundings (using ftp).   A typical day brings 15,000-50,000 hits (from approximately 1000 unique users) on the MM5 web page alone.  On “interesting” weather days the number of hits can exceed 150,000 from several thousand unique users.   Interestingly, non-members of the consortium, Environment Canada’s British Columbia’s offices, are consistently the biggest web access  “customer.”  Occasionally, the UW MM5 graphics have been shown on local TV weathercasts, particularly during active weather situations.

Recently, the first commercial spin-off company based on the UW real-time prediction technologies has opened it doors:  the 3-Tier Corporation.  Started by Ken Westrick and Pascal Storck (previously students and staff members at the UW), this firm is offering real-time hydrological, meteorological, and wind energy forecasting services.  With a large Linux cluster, the firm is able to run both the MM5 and DHSVM hydrological models operationally to provide real-time feeds to its customers (3-Tier web site:


5.         The Research Component


Regional real-time forecast centers can not only provide high-resolution model predictions for the local user community, but can also serve as regional research hubs where local model evaluation and development can occur.  In addition, local efforts can facilitate the creation of new applications for regional users based on local model output and observations.  It may be argued that research into the application of model output and new observing systems has been acutely lacking.  Daily real-time forecasts create large data sets that make possible the evaluation of model forecasts far beyond what is possible in case studies, allowing subtle model biases and infrequent failure modes to be determined.  Regional real-time prediction systems are powerful test beds for improving regional model dynamics, physics, and data assimilation, advances that are often applicable beyond the local area.   Such regional evaluation and research can be completed by individuals with an intimate knowledge of local weather features and data resources and can be highly productive components of the U.S. Weather Research Program (USWRP).

            The Northwest environmental prediction system has facilitated research in a number of areas, as well as spawning major field experiments.  A partial list include:


Effects of increasing resolution:  Using the NorthwestNet observations,  the MM5 forecasts at 36, 12, and 4-km have been evaluated, with the results published in several recent papers (Mass et al 2002, Colle et al. 1999, 2000).  The essential finding has been that using traditional objective measures of forecast skill (e.g., mean absolute or rms errors) error scores decrease substantially as model grid spacing decreases from 36 to 12-km resolution, with far lesser improvement as resolution is decreased to 4 km.  This result is in contrast to subjective evaluations of mesoscale structures, which appear to improve as grid spacing is decreased.  Small timing and position errors preferentially degrade higher resolution forecasts, even if the structures are more realistic (Mass et al 2002).


Mesoscale Model Microphyscal Parameterizations:  Long-term verification of surface precipitation from the UW real-time system has revealed significant problems with the moist physics in the MM5, particularly at the highest resolutions.  A particular problem has been overprediction along the windward slopes of terrain and a dry bias over the lee slopes.  The lack of simultaneous and extensive observations of both basic state structures and microphysical parameters—needed to evaluate and improve model moist physics-- inspired the planning and initiation of a major two-phase field experiment, IMPROVE (Improvement of Microphysical Parameterization through Observational Verification Experiment).  In IMPROVE, aircraft flight level and radar observations--in concert with surface radars, profilers, and other observing systems—provided a comprehensive description of frontal systems approaching the Washington coast and orographic cloud and precipitation structures over the central Oregon Cascades.  Sponsored by the USWRP, the National Science Foundation, and the National Weather Service, IMPROVE data sets are now being used to evaluate and improve microphysical schemes in mesoscale models.  For more information, check 


Ensemble Prediction:  In addition to evaluating the benefits of high-resolution mesoscale predictions, the UW effort has also maintained an active operational research program in mesoscale ensemble prediction.  Using the system described above, the UW ensemble work has demonstrated  a robust relationship between model spread and skill, and has found that over the Northwest initial condition uncertainty is a more important source of model variability than variations in physics.  The UW ensemble research group is working closely with a larger collection of UW investigators from statistics, psychology, and the UW Applied Physics Laboratory in a DOD Multidisciplinary University Research Initiative (MURI) on “Integration and Visualization of Multi-Source Information for Mesoscale Meteorology: Statistical and Cognitive Approaches to Visualizing Uncertainty.”  This research project is making use of the Northwest real-time modeling and observational assets to develop methods for evaluating uncertainty of mesoscale meteorological model prediction, improving statistical methods for dealing with uncertainty, understanding how forecasters incorporate uncertainty in their forecasts, and for integrating and visualizing multi-source information from model output, observations, and expert knowledge.


Boundary Layer Parameterization:  Multi-year verification of the Northwest MM5 forecasts using the MRF boundary layer parameterizations has indicated substantial model biases, including excessive vertical mixing, stronger than observed low-level winds, and insufficient diurnal temperature range.  Long-term and shorter-term verification of other PBL schemes available for the MM5 suggest similar problems.  In recognition of these deficiencies, the USDA Forecast Service has sponsored a multi-year research project for the evaluation and improvement of MM5 PBL schemes to be completed by Jim McCaa and Ralph Foster of the UW Applied Physics Laboratory.      


6.         Lessons Learned:  Does Regional Environmental Prediction Make Sense?


            Seven years of real-time meteorological prediction at the UW provides some perspective on the value of regional numerical forecasting efforts:












One vision of the future of environmental prediction encompasses similar regional forecasting centers across the U.S., with close ties to national centers such as NCEP (NWS) and FNMOC (Navy).   The funding of these centers could come from a single Federal agency (e.g., NWS), multiple Federal agencies (e.g., NWS, DOD, EPA), or through Federal/local partnerships as done in the Northwest consortium.  Precedent for such regional prediction centers already exists in the regional climate centers, which are supported by NOAA (Department of Commerce).  Regional prediction centers would be seen as local foci of prediction, research, data archival and dissemination, with expertise in both regional weather issues and numerical forecasting techniques, and would serve as an integral part of the U.S. Weather Research Program (USWRP).  The Northwest regional effort grew due to the serendipitous combination of universities with the necessary technical skills (University of Washington and Washington State University) and a forward-looking and relatively well-funded user community (e.g., the Consortium).   There is no guarantee of the longevity of the Northwest effort, nor the expectancy that similar cooperative efforts will spring up spontaneously in all regions of the country in which they would prove beneficial.  For this reason, national organization and at least partial funding will undoubtedly be necessary to make the above vision a sustainable reality.


7.         The Future of the Northwest Prediction Effort


            During the next few years the Northwest regional prediction effort will be evolve in a number of ways:





Table 1:  Membership of the Northwest Modeling Consortium


  Regular Members


  National Weather Service

  University of Washington

  USDA Forest Service

  Port of Seattle

  United States Navy

  U.S. Environmental Protection Agency

  Washington State Department of Ecology

  Puget Sound Clean Air Agency

  Washington State Department of Natural Resources

  Washington State Department of Transportation

   Seattle City Light


Corporate Affiliates


  Sun Microsystems, Inc.

  Kuck & Associates, Inc.


Table 2:   NorthwestNet Observation Networks


 1    United States SAO ASOS and AWOS hourly metar observing network

 2    Canadian SAO manual and automated hourly metar observing network

 3    Land 6 hourly synoptic network

 4    Ship 6 hourly synoptic network

 5    CMAN coastal marine automated network

 6    US and Canadian fixed buoy network

 7    Drifting buoy network

 8    Canadian coastal observing network

 9    US coastal observing network

 10  US NRCS SNOTEL network

 11  USDA Forest Service and Bureau of Land Management RAWS network

 12  Northwest Avalanche Center mountain observing network

 13  USDA Agrimet network

 14  Washington State University public agricultural weather network (PAWS)

 15  Hanford - Batelle network

 16  Automated Weather Source (AWS) schoolnet

 17  Weather Underground personal weather station network

 18  University of Washington school network

 19  British Columbia RWIS network

 20  Washington State DOT RWIS network

 21  Washington State Department of Ecology air quality network

 22  Washington State DOT ferry marine observing network

 23  Puget Sound Energy temperature observing network

 24  Seattle City Light network

 25  US Geological Survey hydromet network

 26  US National Ocean Survey marine network

 27  Approximately a half-dozen individual stations


Table 3:  Current Computer Resources for the Northwest Regional Modeling Effort


SUN ES-6500 with 30 processors and 4 GB of memory

SUN ES-2500, with 4 processors.

LINUX Cluster with 20 processors (10 boards with dual 1.2 Ghz Athlon processors)

LINUX Cluster with 32 processors (16 boards of dual 1.533 Ghz Athlon processors

Compaq ES-40 server (4 EV-6 500 Mhz processors with 3.5 GB of memory)

SUN Ultra 10 for pre and post processing

Approximately 6 terabytes of RAID disk storage, 2 terabytes of non-RAID storage






Colle, B. A., K. J. Westrick, and C. F. Mass, 1999: Evaluation of MM5 and Eta-10 precipitation forecasts over the Pacific Northwest during the cool season . Weather and Forecasting, 14, 137-154


Colle, B. A., C. F. Mass, and K. J. Westrick, 2000: MM5 precipitation verification over the Pacific Northwest during the 1977-1999 cool seasons. Weather and Forecasting, 15, 730-744.


Colle, B.A., C. F. Mass, and D. Ovens, 2001: Evaluation of the timing and strength of MM5 and Eta surface trough passages over the eastern Pacific. Weather and Forecasting,16, 553-572


Grimit, E. P., and C. F. Mass, 2002: Initial results of a mesoscale short-range ensemble forecasting system over the Pacific Northwest, 17, 192-205


Mass, C. and Y.-H. Kuo, 1998: Regional real-time numerical weather prediction:  current status and future  potential. Bull. Amer. Meteor. Soc., 79, 253-263


Mass, C., D. Ovens, M. Albright, and K. Westrick, 2002: Does Increasing Horizontal Resolution Produce Better Forecasts?: The Results of Two Years of Real-Time Numerical Weather Prediction in the Pacific Northwest. Bull. Amer. Meteor. Soc., 83, 407-430.


Sharp, J. and C. F. Mass, 2002:  High-resolution forecasts for the Columbia River Gorge.  Accepted for publication in the Bulletin of the Amer. Meteor. Soc.


Westrick, K. and C. Mass, 2000: An evaluation of a high resolution hydrometeorological modeling system for the prediction of a cool-season flood event in a coastal mountainous watershed. Journal of Hydrometeorology, 2, 161-180


Westrick, K. J., P. Storck, and C. F. Mass, 2002, Description and evaluation of a hydrometeorological forecast system for mountainous watersheds. Weather and Forecasting, 17, 250-262.

[1] Puget Sound Regional Synthesis Model, a program supported by  University of Washington internal funding.  More details at