REGIONAL ENVIRONMENTAL PREDICTION OVER THE PACIFIC …cliff/RegEnvMass.pdf · The regional numerical prediction effort at the De-partment of Atmospheric Sciences at the University

1353OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

A n important issue for the weather and environ-mental prediction communities is the organiza-tion of modeling and associated activities.

Should environmental prediction be centralized at afew national centers, decentralized at local forecastcenters close to the user communities, or some com-bination of the two? This issue has become particu-larly timely as rapidly increasing local computer re-sources, the availability of state-of-the-art models, andincreasing access to observational and model data overthe Internet make local environmental prediction in-

creasingly viable (Mass and Kuo 1998). Today ap-proximately three-dozen sites in the United States arerunning mesoscale atmospheric models in real timeusing a range of modeling systems, including Thefifth-generation Pennsylvania State University–National Center for Atmospheric Research (PennState–NCAR) Mesoscale Model (MM5), the CoupledOcean–Atmosphere Prediction System (COAMPS),the Advanced Regional Prediction System (ARPS),the Regional Atmospheric Modeling System (RAMS),and the National Centers for Environmental

REGIONAL ENVIRONMENTALPREDICTION OVER THEPACIFIC NORTHWEST

BY CLIFFORD F. MASS, MARK ALBRIGHT, DAVID OVENS, RICHARD STEED, MARK MACIVER, ERIC GRIMIT, TONY

ECKEL, BRIAN LAMB, JOSEPH VAUGHAN, KENNETH WESTRICK, PASCAL STORCK, BRAD COLMAN, CHRIS HILL,NAYDENE MAYKUT, MIKE GILROY, SUE A. FERGUSON, JOSEPH YETTER, JOHN M. SIERCHIO, CLINT BOWMAN,

RICHARD STENDER, ROBERT WILSON, AND WILLIAM BROWN

The experiences of the Northwest Modeling Consortium demonstrate the potential of local

modeling as an important component of the future numerical

environmental prediction system.

AFFILIATIONS: MASS, ALBRIGHT, OVENS, STEED, MACIVER, GRIMIT,AND ECKEL—Department of Atmospheric Sciences, Universityof Washington, Seattle, Washington; LAMB AND VAUGHAN—Laboratory for Atmospheric Research, Washington StateUniversity, Pullman, Washington; WESTRICK AND STORCK—3-Tier, Inc., Seattle, Washington; COLMAN AND HILL—NationalWeather Service, Seattle, Washington; MAYKUT AND GILROY—Puget Sound Clean Air Agency, Seattle, Washington;FERGUSON—PUSDA Forest Service, Pacific Northwest ResearchStation, Seattle, Washington; YETTER AND SIERCHIO—Naval AirStation Whidbey Island, Oak Harbor, Washington; BOWMAN

AND STENDER—Washington State, Department of Ecology,Olympia, Washington; WILSON—U.S. EPA Region X, Seattle,Washington; BROWN—Washington State, Department ofTransportation, Seattle, WashingtonCORRESPONDING AUTHOR: Clifford F. Mass, Departmentof Atmospheric Sciences, University of Washington, Box351640, Seattle, WA 98195E-mail: [email protected]: 10.1175/BAMS-84-10-1353

In final form 22 May 2003© 2003 American Meteorological Society

1354 OCTOBER 2003|

Prediction’s (NCEP’s) Eta Model (see www.mmm.ucar.edu/mm5/mm5forecast/sites.html for a partiallisting). In addition, the use of real-time air quality andhydrological models is rapidly increasing.

The regional numerical prediction effort at the De-partment of Atmospheric Sciences at the Universityof Washington (UW) was initiated in 1995 as a single-domain forecast system applying MM5 with 27-kmgrid spacing. In the succeeding years, it has growninto a regional environmental forecast system thatincludes atmospheric, hydrologic, and air quality real-time prediction down to 4-km horizontal resolution;a wide range of real-time applications; and the col-lection of many telemetered observational networksin the Pacific Northwest. A significant contributor tothe success of the Northwest modeling effort has beenthe management and funding by the Northwest Mod-eling Consortium, a collection of federal, state, andlocal agencies. This paper reviews the scope and ap-proach of the Northwest modeling effort and exam-ines its implications as a national model.

HISTORY OF THE NORTHWEST RE-GIONAL PREDICTION EFFORT. The North-west regional prediction effort began in the early1990s when a group of Northwest air quality andweather prediction agencies identified the lack ofupper-air observations over Puget Sound as a majorobstacle for the diagnosis and prediction of localweather and air quality. Under the chairmanship ofN. Maykut of the Puget Sound Air Pollution ControlAgency, a Northwest Upper-Air Committee wasformed and proceeded to identify the Radian915-MHz radar wind profiler as a possible solution.The group then devised a novel funding approach:support in terms of dollars or other assets (land, per-sonnel) by a “consortium” of agencies. The profilerwas purchased in 1992 and remains operational to thisday under the care of the Seattle National WeatherService (NWS) office. At roughly the same time,M. Albright, a UW staff member and Washingtonstate climatologist, began construction of a regionalreal-time weather database for research and forecast-ing by collecting data from several Northwest weatherobservation networks into one UW computer server.In such a way a relatively dense mesoscale network wasbuilt at little cost, while coordination between differentnetworks reduced duplication of effort. The regionalobservational database (NorthwestNet) has growninto a collection of over two-dozen networks, includ-ing nearly a thousand stations over the Northwest.

During the late 1980s and early 1990s the lead au-thor and several of his students began research simu-

lations of weather features of the west coast of NorthAmerica using the Colorado State RAMS and theMM4/5 mesoscale models. Running with grid spac-ing down to 5 km, it was found that such mesoscalemodels could produce highly realistic mesoscale cir-culations, particularly those driven by orography, ifthe synoptic forcing was accurate. By 1994, relativelyfast single-processor UNIX workstations becameavailable, making it possible to run regional domainsat much higher resolution than used at national mod-eling centers such as NCEP, where the NCEP EtaModel was being applied at 80-km grid spacing. Basedon the promising research runs, the Northwest Up-per-Air Committee (soon to be renamed the North-west Modeling Committee) decided to support theevaluation of local numerical weather prediction(NWP). The initial evaluations completed byJ. Steenburgh (then a UW postdoc) were so promis-ing that in 1995 real-time prediction using a single27-km domain of the MM5 (with initialization andboundary conditions from the Eta Model) was begunusing a single processor, an Alphaserver 250. Thevalue of the real-time MM5 prediction system becameclear during the next year, as it successfully forecastimportant regional circulations (such as onshorepushes and coastal surges) for which the Eta Modellacked sufficient resolution.

Research runs had determined that realistic simu-lation of the major mesoscale features of the North-

Regular MembersNational Weather ServiceUniversity of Washington

Washington State UniversityUSDA Forest ServicePort of Seattle

U.S. NavyU.S. Environmental Protection AgencyWashington State Department of Ecology

Puget Sound Clean Air AgencyWashington State Department of Natural ResourcesWashington State Department of Transportation

Seattle City Light

Associate MembersOregon Department of Forestry

Corporate AffiliatesSun Microsystems, Inc.KAI Software, Inc.

TABLE 1. Membership of the Northwest ModelingConsortium.

1355OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

west required a grid spacing less than 15 km. Theseresults, coupled with the clear value of the 27-kmruns, inspired a jump to far higher resolution. Withfunding from a large collection of agencies (“the Con-sortium,” see Table 1) and an exceptional discountoffered by Sun Microsystems, the UW purchased aSUN E4000 server with 14 processors during the sum-mer of 1996. Using this powerful system, a new gridconfiguration was initiated with a large 36-km domainthat extends several thousand kilometers over theeastern Pacific and western North America, and anested 12-km grid over the entire Pacific Northwest.With the acquisition of upgraded processors the fol-lowing year, an additional 4-km nest was added overwestern Washington, making the Northwest effort thehighest-resolution NWP effort in the United States fora short period. Additional computers have been ac-quired during the past 3 yr, allowing the expansionof the 12- and 4-km grids, extension of the simula-tions to 72 h, and the addition of approximately two-dozen ensemble forecasts at 36- and 12-km grids spac-ing. During the last few years, the Pacific Northwestprediction effort has grown well beyond atmosphericmodeling and diagnosis to perform hydrological, airquality, and smoke dispersion modeling, as well asother applications.

MAJOR COMPONENTS OF THE NORTH-WEST ENVIRONMENTAL PREDICTIONSYSTEM. The Northwest real-time regional predic-tion system can be divided into four levels (Fig. 1).The top level includes all the observational and modelinputs required by the regional models and applica-tions. The second level contains the local atmosphericmodeling systems, while the third level encompassesthe local environmental modeling systems and appli-cations. The fourth level includes the Web pages anddistribution channels through which model outputand observations are provided to a diverse user com-munity. A Web portal to all components of the North-west Environmental Prediction System is foundonline at www.atmos.washington.edu/pnw_environ/.

Observational and model inputs. All available observa-tions that can be accessed in real-time (or near real-time) are decoded, quality controlled, placed on harddisks for several weeks to several years, and archivedon tape. This collection of observations, known asNorthwestNet, is acquired from approximately two-dozen networks (Table 2). A plot of the NorthwestNetsurface observations over the state of Washington isshown in Fig. 2. Other groups, including the MesoWestnetwork run by the University of Utah, have taken up

this idea of building a network of preexisting observa-tional networks. In fact, NorthwestNet observationsare being sent to MesoWest operationally, where theyare transferred to the NWS Western Region for dissemi-nation to regional NWS offices for use in the AdvancedWeather Information and Prediction System (AWIPS).

In addition to surface observations, the UW effortalso gathers all regional upper-air data, including ra-diosonde soundings, the Seattle 915-MHz profilertemperatures and winds, and the Aircraft Commu-nications Addressing and Reporting System (ACARS)aircraft observations—which are becoming an ex-traordinary rich source of mesoscale data aloft. Otherdata sources include all NWS Weather SurveillanceRadar-1988 Doppler (WSR-88D) radar data and sat-ellite imagery for the region.

For initialization of the real-time MM5 forecasts,including the ensemble runs, gridded analyses andforecasts are acquired operationally from a numberof major prediction centers such as NCEP, the Cana-dian Meteorological Center, the Australian Bureau ofMeteorology, the Taiwan Central Weather Bureau,the Met Office, and the U.S. Navy Fleet NumericalOceanography Center. Most of these datasets are ac-quired through file transfer protocol (ftp) servers. Forthe NCEP products a redundant feed uses theUNIDATA CONDUIT system, in which model gridsare distributed over the Internet through a few ma-jor sites using the UNIDATA Local Data Manager(LDM) system.

Local atmospheric modeling systems. PENN STATE–NCARMESOSCALE MODEL (MM5). Using 38 vertical levels andthree nested grids (Fig. 3), the UW real-time systemis run twice daily (0000 and 1200 UTC) over all threegrids with initialization and boundary conditions

FIG. 1. Schematic of the Northwest Environmental Pre-diction System.

1356 OCTOBER 2003|

from NCEP’s Global Forecast System (GFS) modelforecasts, which have proven to provide superior syn-optic guidance. The MM5 is also run twice a day, forthe 36–12-km domains, using the NCEP Eta Modelgrids, to provide early high-resolution guidance. Thistype of “cold start,” without any local data assimila-tion or spinup period, was used after tests showed thatmesoscale data assimilation using local data assets im-proved forecasts only during the first few hours. TheUW MM5 forecasts are run for 72 h over the 36- and12-km grids and for 42 h (6–48 h) over the 4-km do-main. For the 0000 UTC cycle the MM5 is run withGFS forcing out to 7 days for use in the NationalWeather Service Interactive Forecast Preparation Sys-tem (IFPS). Cumulus parameterization (Kain–Fritsch) is only applied over the outer domains. TheMM5 output is verified operationally against theNorthwestNet observations and is available online ingraphical form or by ftp transfer. More informationabout the UW high-resolution MM5 runs can be foundat its Web site (www.atmos.washington.edu/mm5rt/).

UW MESOSCALE SHORT-RANGE ENSEMBLE FORECAST SYSTEM.Operational evaluation of the UW high-resolutionforecasts has suggested that poor initialization overthe Pacific is a large source of prediction uncertainty.To evaluate such initial condition uncertainty (as wellas uncertainty due to model error) and to explore thepotential of probabilistic forecasts, MM5 ensembleforecasts were initiated in January 2000 using 5 mem-bers and continue today (expanded to 25 members).The UW ensemble system is based on running the36- and 12-km MM5 domains multiple times usingthe initializations and boundary conditions from anumber of operational modeling systems [e.g., NCEPEta and AVN models, the U.S. Navy’s OperationalGlobal Atmospheric Prediction System (NOGAPS)model, the Canadian Global EnvironmentalMultiscale (GEM) model, the Met Office andJapanese global models, the Australian Global As-similation Prediction (GASP) model, and the Tai-wanese global model]. The central idea is that thevariation in the initializations of major modeling sys-

tems provides a measure ofinitialization uncertainty. Ad-ditional members of the UWensemble system are createdby varying model physical pa-rameterizations (microphys-ics, boundary layer schemes,moist physics) and surfaceproperties (variations of seasurface temperature and soilmoisture within observationalerror). Furthermore, UW en-semble work has tested theapplication of initialization“mirrors,” whereby particularinitializations are reflectedaround the ensemble mean(see the Web site provided be-low for more details). This en-semble work has been facili-tated by the purchase ofrelatively inexpensive Linuxclusters through which the en-sembles can be efficiently andrapidly computed. Opera-tional for over 2 yr, the initialresults of the UW ensemblesystem are reviewed in Grimitand Mass (2002), and dailyforecasts are found online atwww.atmos .wash ington .edu/~emm5rt .ensemble/

1 U.S. SAO ASOS and AWOS hourly METAR observing network

2 Canadian SAO manual and automated hourly METAR observing network3 Land 6-hourly synoptic network4 Ship 6-hourly synoptic network

5 CMAN coastal marine automated network6 U.S. and Canadian fixed buoy network7 Drifting buoy network

8 Canadian coastal observing network9 U.S. coastal observing network10 U.S. NRCS SNOTEL network

11 USDA Forest Service and Bureau of Land Management RAWS network12 Northwest Avalanche Center mountain observing network13 USDA Agrimet network

14 Washington State University public agricultural weather network (PAWS)15 Hanford–Batelle network16 Automated Weather Source (AWS) schoolnet

17 Weather Underground personal weather station network18 University of Washington school network19 British Columbia RWIS network

20 Washington State DOT RWIS network21 Washington State Department of Ecology air quality network22 Washington State DOT ferry marine observing network

23 Puget Sound Energy temperature observing network24 Seattle City Light network25 U.S. Geological Survey hydromet network

26 U.S. National Ocean Survey marine network27 Approximately a half-dozen individual stations

TABLE 2. NorthwestNet observation networks.

1357OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

cgi. Figure 4 provides examples of individual en-semble forecasts and derived probability products.Local environmental modeling systems and applications.

REGIONAL HYDROLOGICAL PREDICTION SYSTEM. Beginning in De-cember 1997, 12- and 4-km output from the MM5has been used to drive a fully distributed hydrologi-cal model [the Distributed Hydrological Soil Vegeta-tion Modeling System (DHSVM)] that was devel-oped by Professor D. Lettenmaier and students at theUW Civil Engineering Department. The coupling ofthe MM5 and the DHSVM(initially completed by K.Westrick) was so successfulthat the number of simulatedwatersheds has been increasedfrom 1 to 26, encompassingmost of the river basins inwestern Washington state(Fig. 5a). Running at 150-mresolution, the real-timestreamflow forecasts are madedaily out through 60 h, usingexplicit channel routing thatprovides streamflow at any

point in the river networks. Thehydrological predictions are ac-cessible over the Web and dis-tributed to the NationalWeather Service forecast officein Seattle. For over a year, theUW hydrological predictionsystem was driven by the en-semble forecasts as well, pro-viding a collection ofhydrographs at numerous sites(see Fig. 5b). D. McDonnal ofthe Seattle NWS office hasbuilt an interactive display andanalysis system that allowsNWS forecasters to view theregional hydrological fore-casts and streamflow observa-tions. More information on

the UW hydrological effort can be found on thehydrometeorology Web page (http://hydromet.atmos.washington.edu/index.html) or in recent publications(Westrick et al. 2002; Westrick and Mass 2001).

SMOKE AND FIRE GUIDANCE. The suppression of wildfires,as well as the planning and control of prescribed burnsin forest and rangeland areas, requires detailed meteo-rological guidance, particularly over the mountainousNorthwest. To provide such information, the UnitedStates Department of Agriculture (USDA) Forest Ser-

FIG. 3. Model domains for theNorthwest real-time MM5 fore-casts run at the UW. The gridspacing is 36 km for the outerdomain, 12 km for the middledomain, and 4 km for the innerdomain. Terrain contours (graylines) are given every 300 m.

FIG. 2. NorthwestNet surface observation locations over Washington state.

1358 OCTOBER 2003|

vice is collaborating with theUW and the Consortium toprovide a wide range of fire-and smoke-related productsdriven by MM5 forecastsand regional data assets.Many of these products canbe accessed through theSmoke and Fire Web site(www.atmos.washington.edu/gcg/smokeandfire/), whichalso provides graphical dis-plays of the regional MM5forecasts, including meteo-grams and soundings atlocations around the North-west. The site includes guid-ance products of forecast firepotential (driven by MM5output), such as the Hainesand Fosberg fire indices, andventilation indices that com-bine MM5 winds, stability,and boundary layer depths(see Fig. 6a). As part of theForest Service BlueSkyproject the regional MM5grids are interfaced with aLagrangian “puff” model(CALPUFF) to predictsmoke distributions fromwild and prescribed fires(Fig. 6b). MM5 graphics andBlueSky smoke predictionsare available on EPA’sBlueSky-RAINS web site(www.B lueSkyRAINS .org). MM5–CALPUFF isalso being applied in an ag-ricultural smoke manage-ment system, devised atWashington State Univer-sity. In this system, MM5–CALPUFF predicts smokedispersion based on the hy-pothetical field-burningscenarios. In addition,Oregon Department ofForestry forecasters accessMM5 sounding predictionsto avoid adverse air qualityimpacts from prescribedburning.

FIG. 4. UW ensemble system products: (a) 3-h precipitation with color shad-ing ranging from heavy (red) to light (blue) and winds (wind barbs) for 48 hinto a forecast initialized 0000 UTC 11 Mar 2003 and (b) probability of 6-hprecipitation greater than 0.01 inch for a 48-h forecast initialized 28 Jul 2002.On the Web site, both the individual ensemble members and derived prod-ucts (e.g., ensemble spread, probabilistic forecasts) are provided.

1359OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

REGIONAL AIR QUALITY PREDICTIONS. Since the spring of 2001,real-time air quality forecasts have been made overwestern Washington by using MM5 forecasts to drivean Eulerian photochemical air quality model called

the California Photochemical Grid Model (CALGRID),which provides predictions of ozone, nitrogen oxides,and other species of interest. This system has beenbuilt by J. Vaughan, B. Lamb, and others at WSU in

FIG. 5. (a) Watersheds currently modeled operationally in the UW coupled hydrological prediction sys-tem. (b) Sample hydrographs showing stream flow on the Raging River near Fall City, WA. The bluerepresents the observed flow, the red (black) lines are based on forcing from the 12 (4 km) domain.Overlapping hydrographs are from consecutive hydrological forecasts.

1360 OCTOBER 2003|

cooperation with the UW Atmospheric SciencesMM5 group and the U.S. Environmental ProtectionAgency (EPA) sponsored Air Indicator Report forPublic Awareness and Community Tracking(AIRPACT) effort. The coupled MM5–CALGRIDsystem is run once a day at 4-km grid spacing for 24 husing hourly gridded emissions data provided by theWashington State Department of Ecology. Graphicaldisplays of the emissions data and CALGRID forecastsare available at the project Web site (http://airpact.ce.wsu.edu/index2.html). An example of a

CALGRID ozone forecast isfound in Fig. 7. Further de-tails regarding the North-west air quality predictionsystem are found in Vaughanet al. (2003).

ROAD WEATHER INFORMATION

SYSTEM. NorthwestNet re-gional observations and theUW MM5 forecasts arecombined to provide guid-ance for the traveling pub-lic and Washington StateDepartment of Transporta-tion (WSDOT) personnel.In addition, the Oregonstate land surface modelprovides forecasts of roadsurface temperatures formajor highways. ThroughWeb portals (www.wsdot.wa.gov/traffic/ or www.wsdot.wa.gov/rWeather/),people can view maps ofreal-time observations ofweather and road condi-tions as well as forecastsalong a particular highwaysection (Fig. 8). As part ofthe project, several dozenweather sensors have beenplaced along state highwaysand on Washington stateferries that cross the inlandwaters of the state (see“Ferry Weather” page inFig. 9).

Web pages and other distri-bution channels. Bothmodel output and observa-

tions are distributed through several channels. A widerange of graphical imagery of the MM5 high-resolu-tion runs, the UW ensemble forecasts, the UW ob-servational data collection, and of the output of thevarious regional applications and environmentalmodeling systems are available online (www.atmos.washington.edu/pnw_environ/). Model grids and ex-tracted model soundings are also provided throughftp access for major users, such as the NWS and theUSDA Forest Service.

FIG. 6. Smoke and fire products. (a) Ventilation index (18-h forecast) based onMM5 surface winds and low-level stability for 1200 UTC 28 Feb 2002. (b) The19-h forecast of surface concentrations of particulate matter (less than 2.5 µmin diameter) valid at 1900 UTC 3 Aug 2002, calculated using the CALPUFFLagrangian dispersion model forced by the 12-km northwest MM5 forecast.The ventilation index is the product of boundary layer wind speed and stabil-ity and has units of m2 s−1.

1361OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

COMPUTATIONAL INFRASTRUCTUREAND MANAGEMENT. With operational integra-tion of a number of models at high resolution and thepreparation of thousands of graphical and other prod-ucts each day, the computational demands for theNorthwest modeling effort are correspondingly large.Located at the Atmospheric Sciences Department atthe UW Seattle campus, the main computer resourcesinclude a 30-processor SUN 6500 server, a 20-proces-sor Athlon (1.2 GHz) Linux cluster, a 32-processorAthlon (1.5 GHz) Linux cluster, over 6 terabytes ofRedundant Array of Independent Disks (RAID) discstorage, 2 four-processor servers for integration of thehydrological and air quality models, and four addi-tional machines for pre- and post processing of modeldata and graphics generation (see Table 3). The ex-cellent scalability of the MM5 on large numbers ofprocessors has been a major factor in allowing high-resolution predictions.

The Northwest MM5 forecasting system hasproven to be highly dependable, providing predic-tions even when NCEP has been down. At the UWsuch robustness has been made possible by havingalternative sources of initialization and boundary con-

FIG. 7. The 16-h forecast from the coupled MM5–CALGRID system developed by Washington StateUniversity. High levels of ozone are forecast south-east and south of Seattle.

FIG. 8. Interstate 90 (I90) Travel Route Information Web page. (top) Camera imagery across theCascades, (lower) real-time weather observations and road surface temperatures. In addition, futureconditions across the mountains, driven by the UW MM5, can be viewed on this page and clicking onthe cross section at any point provides the appropriate NWS forecast.

1362 OCTOBER 2003|

dition grids, multiprocessor machines tolerant ofcomponent failures, RAID disk arrays, uninter-ruptible power supplies that keep models runningthrough power spikes and failures, and complex con-trol scripts that are tolerant of a wide variety of fail-

ure modes (e.g., missing grids)—not to mention dedi-cated monitoring.

The Northwest environmental prediction effort isrun by the Northwest Modeling Committee and isfunded by the Northwest Modeling Consortium (a listof members is provided in Table 1). The committeemeets quarterly and makes all major decisions regard-ing system development and the acquisition of newhardware and resources. Regular e-mail updates arealso used to apprise Consortium members of majormodel changes and for online discussion of importantissues between meetings.

THE USER COMMUNITY AND COMMER-CIAL “SPINOFFS.” The use of Northwest envi-ronmental prediction products has grown rapidlyover the past five years. The MM5 forecasts are usedoperationally by local NWS offices, military forecast-ers, private sector and media meteorologists, and fire-weather, air quality, transportation, and recreationalinterests—to name only a few. A typical day brings50,000–150,000 hits (from several thousand unique

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

SUN ES-2500, with four processors.

LINUX Cluster with 20 processors (10 boards with dual1.2 GHz Athlon processors)

LINUX Cluster with 32 processors (16 boards of dual1.533 GHz Athlon processors)

Compaq ES-40 server (4 EV-6 500 MHz processors with3.5 GB of memory)

SUN Ultra 10 for pre- and postprocessing

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

TABLE 3. Current computer resources for thenorthwest regional modeling effort.

FIG. 9. The Ferry Weather Page that shows weather observations along ferry routes as well as nearbyland-based data.

1363OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

users) on the MM5 Web page alone. On “interesting”weather days the number of hits can exceed 500,000.Occasionally, local TV weathercasters air UW MM5graphics, particularly during active weather situations.

Recently, the first commercial spinoff companybased on the UW real-time prediction technologieshas opened its doors: the 3-Tier Corporation, startedby K. Westrick and P. Storck (former UW studentsand staff members who built the UW hydrologicalprediction system). This firm offers real-time hydro-logical, meteorological, and wind energy forecastingservices, and runs both the MM5 and DHSVM hy-drological models operationally with a large Linuxcluster (see their Web site at www.3tiergroup.com/).

REGIONAL RESEARCH. Regional real-time fore-cast centers not only provide model predictions andderived applications for the local user community, butcan also serve as regional research hubs for modelevaluation and development. In addition, such effortscan facilitate the creation of new applications for lo-cal users based on regional model output and obser-vations. Regional model evaluation and research havethe added advantage of being completed by individu-als with an intimate knowledge of local weather fea-tures and data resources.

Daily real-time forecasts create large datasets thatmake possible the evaluation of model forecasts farbeyond what is possible in case studies, allowing subtlemodel biases and infrequent failure modes to be de-termined. Regional real-time prediction systems arepowerful test beds for improving mesoscale modeldynamics, physics, and data assimilation, advancesthat are often applicable nationally. Thus, they can behighly productive components of the U.S. WeatherResearch Program (USWRP) and can be valuablepartners to national centers such as NCEP and theFleet Numerical Meteorological and OceanographyCenter (FNMOC).

The Northwest environmental prediction systemhas facilitated research in a number of areas, as wellas spawning major field experiments. A partial list ofregional research efforts associated with the North-west modeling effort includes the following:

• Effects of increasing resolution: Using theNorthwestNet Observations, the MM5 forecasts at36, 12, and 4 km have been evaluated, with theresults published in several recent papers (Masset al 2002; Colle et al. 1999, 2000). The essentialfinding has been that using traditional objectivemeasures of forecast skill (e.g., mean absolute orrms errors), model errors decrease substantially as

grid spacing decreases from 36 to 12 km, with farless improvement as grid spacing is decreased to4 km. The latter result contrasts with subjectiveevaluations of mesoscale structures, which suggestconsiderable reduction in model error as grid spac-ing decreases below 12 km. One explanation is thatsmall timing and position errors preferentiallydegrade higher-resolution forecasts (which havetighter structures and more amplitude), even if thestructures are more realistic (Mass et al. 2002).

• Mesoscale short-range ensemble methodology andevaluation: Although a number of studies have ex-plored the value of ensembles for the centralUnited States under convective conditions (e.g.,Stensrud et al. 1999), relatively little evaluation hasbeen given to the value of ensembles over coastalregions of substantial terrain. The UW ensembleresearch effort has taken on such a study. Further-more, while most ensemble studies have made useof breeding or singular vector perturbation ap-proaches, the UW work is examining the applica-tion of initializations from multiple operationalcenters. The UW ensemble work has demon-strated a robust relationship between model spreadand skill (Grimit and Mass 2002). The UW en-semble research group is working closely with alarger collection of UW investigators from statis-tics, psychology, and the Applied Physics Labora-tory in a Department of Defense (DOD) initiativeto develop methods for evaluating uncertainty inmesoscale meteorological model prediction, im-proving statistical methods for dealing with uncer-tainty, and understanding how forecasters incor-porate uncertainty in their forecasts (www.stat.washington.edu/MURI/).

• Model microphysical parameterizations: Long-term verification of surface precipitation from theUW real-time system revealed significant prob-lems with the moist physics schemes in the MM5,particularly at the highest resolutions. A particu-lar problem has been overprediction along thewindward slopes of terrain. The lack of simulta-neous and extensive observations of both basic-state structures and microphysical parameters—needed to evaluate and improve model moistphysics—inspired the planning and initiation of amajor two-phase field experiment: the im-provement of Microphysical Parameterizationthrough Observational Verification Experiment(IMPROVE). In IMPROVE, aircraft flight leveland radar observations—in concert with surfaceradars, profilers, and other observing systems—provided a comprehensive description of frontal

1364 OCTOBER 2003|

systems approaching the Washington coast andorographic cloud and precipitation structures overthe central Oregon Cascades. IMPROVE datasetsare now being used to evaluate and improve mi-crophysical schemes in several mesoscale models.[More information is available online at http://improve.atmos.washington.edu; also see the over-view paper Stoelinga et al. (2003)].

• Boundary layer parameterization: Multiyear veri-fication of the Northwest MM5 forecasts using themedium-range forecast (MRF) Model planetaryboundary layer (PBL) parameterization has indi-cated substantial model biases, including excessivevertical mixing, stronger than observed low-levelwinds, and insufficient diurnal temperature range(Mass et al. 2002). Long-term and shorter-termverification of other PBL schemes available for theMM5 suggests similar (and other) problems. Inrecognition of these deficiencies, a joint projectbetween UW Atmospheric Sciences, the UW Ap-plied Physics Laboratory, and the USDA ForestService is evaluating and improving PBL schemesfor the MM5 and the Weather Research and Fore-casting (WRF) models.

LESSONS LEARNED: DOES REGIONALENVIRONMENTAL PREDICTION MAKESENSE? Seven years of real-time meteorological pre-diction at UW provide some perspective on the viabil-ity, value, and potential organization of regional pre-diction and its relationship to national forecastingcenters.

How can regional prediction augment and enhance na-tional prediction efforts? The Northwest modeling ex-perience has demonstrated that a regional effort canoptimize the forecasting system for an area and pro-vide additional value to the numerical weather prod-ucts from national centers. For example, while NCEPwas running the Eta Model at 32- and 22-km gridspacing, which is inadequate to model critical North-west weather features, the Northwest effort was run-ning the MM5 (driven by the Eta Model’s initial con-ditions and boundary conditions) at grid spacingsdown to 4 km. Equally as important, the NCEP EtaModel’s vertical coordinate system does not work wellfor high-resolution simulations near and over terrain,producing excessive flow blockage and a near absenceof realistic mountain waves or downslope windstorms. The sigma-coordinate MM5 was a betterchoice for the highly mountainous Pacific Northwest.Thus, regional prediction enhanced Eta forecasts overa region for which the Eta’s structure was not opti-

mal. With greatly varying physiography and meteo-rology across the United States, one-size-fits-all nu-merical weather prediction is not necessarily the bestapproach.

Local prediction centers, such as UW, can providefull-resolution model output for regional applicationsand other uses. In contrast, the National WeatherService has had difficulties providing full-resolutionmodel output to even its own offices. For many ap-plications (e.g., local trajectory, dispersion, or airquality models), the current 3-h temporal resolutionof NCEP model grids is not adequate and thus localmodel integration is required.

The creation of an integrated environmental pre-diction system encompassing atmospheric, air qual-ity, hydrological, and other predictions systems is notonly possible but a necessary step for dealing withcrucial societal needs, including national security. TheUW’s effort has shown the potential of coupling ahigh-resolution mesoscale atmospheric model with avariety of other models and applications. At present,a local prediction system is the only way to create suchintegration, since high temporal and spatial resolu-tion are required.

The Northwest effort has proven effective in iden-tifying and collecting local weather data for both lo-cal and national applications. Regional centers areoften aware of data sources not apparent to nationalcenters and can facilitate access to such data throughlocal contacts. In turn, the local centers can contrib-ute regional data to national entities. Furthermore,regional efforts can identify areas where data areneeded and work with local organizations to placeadditional observing assets. Intimate local knowledgeallows the identification of problematic observing sitesor the determination of where quality control algo-rithms are rejecting valid data.

As experienced in the Northwest, local predictioncenters can enjoy a close relationship with users, gar-nering quick feedback regarding model strengths andweaknesses. Such interactions encourage rapid im-provement in the modeling systems, as well as thetailoring of graphics and output to the needs of theuser community.

Local environmental prediction efforts can serveas active centers for model improvement and localresearch. Such efforts should be viewed as essentialcomponents of the USWRP, whereby regional cen-ters contribute improvements to national modelingsystems, develop new applications, and evaluate theirusefulness to regional users. After rigorous local test-ing, productive ideas can then feed back to nationalcenters such as NCEP, FNMOC, and the Air Force

1365OCTOBER 2003AMERICAN METEOROLOGICAL SOCIETY |

Weather Agency (AFWA). Regional centers can serveas repositories of local weather knowledge and act asfoci for research on the meteorological phenomena oftheir area.

Support and management of regional prediction efforts.The Northwest experience has shown that the con-sortium approach, whereby a collection of local, state,and federal agencies combine resources, is a viable,but time consuming, method to fund such efforts. Ithas taken several years for a level of trust and owner-ship to develop among the principals of such a jointenterprise, with patient, empathetic leadership beingcrucial. A strength of the Northwest consortium ex-perience has been the relative robustness of the fund-ing—although individual agency contributions varyconsiderably year to year, the aggregate total has beenfar steadier (typically $200,000–$300,000 per annumfor operational—nonhardware—support).

The Northwest regional effort grew due to the ser-endipitous combination of universities with the nec-essary technical skills (University of Washington andWashington State University) and a forward-lookinguser community willing to provide needed funding(the Consortium). There is no guarantee of the lon-gevity of the Northwest effort, nor the expectancythat similar cooperative efforts will spring up spon-taneously in all regions of the country in which theywould prove beneficial. For this reason, nationalorganization and at least partial funding will be nec-essary to make the vision of a network of regionalcenters a sustainable reality. One attempt to providesuch a national support structure is the USDA For-est Service funded Fire Consortia for Advanced Mod-eling of Meteorology and Smoke (FCAMMS).FCAMMS was created to foster the development ofreal-time regional modeling and consortia buildingat research laboratories in Riverside, California; EastLansing, Michigan; Athens, Georgia; Ft. Collins,Colorado; and Missoula, Montana, as well as theNorthwest effort (more information available onlineat www.fs.fed.us/FCAMMS). Regional prediction ef-forts could also be formed under the auspices of theU.S. Weather Research Program, or built into the Na-tional Weather Service (which is already divided intoregions).

There should be no sense of tension or competi-tion between regional and national prediction efforts.Both are required to effectively develop forecastingtechnology and to serve the user community. Onevision of the future of environmental prediction en-compasses similar regional forecasting centers acrossthe United States, with close ties to national centers.

Funding could come from a single federal agency,multiple federal agencies (e.g., NWS, DOD, EPA), orthrough federal–local partnerships as done in theNorthwest consortium. Another approach would befor the National Weather Service to take on the taskof creating and maintaining regional prediction cen-ters that would maintain close ties with NCEP.Precedent for such regional prediction centers alreadyexists in the regional climate centers, which are sup-ported by the National Oceanic and AtmosphericAdministration (NOAA).

THE FUTURE OF THE NORTHWEST PRE-DICTION EFFORT. With large numbers of Webhits and the continual demand for more products, thecontinued need for the forecast and diagnostic prod-ucts of the Northwest effort seems clear. During thenext few years the Northwest regional prediction ef-fort will evolve in a number of ways:

• The new WRF mesoscale model, the planned re-placement for both the MM5 and Eta Models, willbe evaluated in parallel runs with MM5 during2003. If verification scores show improved forecastskill, a switch will be made to WRF.

• During 2003 the Northwest MM5 and hydrologi-cal (DHSVM) models will be coupled to a real-time Puget Sound predictive system based on thePrinceton Ocean Model (POM), in concert withM. Kawase (UW, Oceanography) and the UWPuget Sound Regional Synthesis Model (PRISM)program.

• The Northwest effort will test real-time mesoscaledata assimilation. With the increasing availabilityof ACARS aircraft data during ascents and de-scents into Northwest airports, greatly increasednumbers of surface reports, the availability of ad-ditional radar data, and finally improved local dataassimilation analysis tools [e.g., three-dimensionalvariational (3DVAR) methods, the ensembleKalman filter], the time has come to reevaluateregional data assimilation over the Northwest. Inconcert with this work, the Northwest system willprobably move to a “warm start” with the modelbeing spun up before the nominal start time.

• The 4-km domain will be expanded to include allof Idaho. Ultra-high resolution domains (1.3-kmgrid spacing) may be added for Puget Sound andthe Columbia River Gorge. Local research efforts(e.g., Sharp and Mass 2002) have shown that forsome regional features, such as the Columbia RiverGorge, 1.3-km grid spacing is required to producerealistic structures.

1366 OCTOBER 2003|

• The short-range mesoscale ensemble system willbe expanded further and a new generation of re-gional ensemble-based probabilistic guidance willbe created. In concert with a multidisciplinaryDOD-sponsored research project, we will cali-brate, combine, and/or weight ensemble membersusing Bayesian and other approaches.

• The regional air quality prediction system will beexpanded to encompass the Portland, Oregon, ur-ban area and several air toxic compounds will beadded. Work is also under way to convert the sys-tem from using CALGRID to using the EPA Com-munity Multi-Scale Air Quality (CMAQ) modelthat will provide additional capabilities for fore-casting particulate matter in the region.

• Work will continue on improvements in modelphysics, including the moist physics and bound-ary layer schemes.

• Experimentation has begun on grid-based re-moval of systematic bias. All modeling systemspossess systematic bias, which is often removed atobserving sites using methods such as model out-put statistics. With the use of model grids for fore-cast dissemination and display, it is crucial to re-move such biases on the grid itself. After testing avariety of approaches, the Northwest modelingeffort will begin such bias removal on an opera-tional basis.

ACKNOWLEDGMENTS. The U.S. EnvironmentalProtection Agency, NWS Western Region, Puget SoundClean Air Agency, the USDA Forest Service, and theCOMET Outreach Program have provided continuoussupport over the past decade for both personnel and hard-ware. The UW PRISM program has provided sustainedfunding for the development of the real-time hydrologicalprediction system. The Office of Naval Research has sup-ported the verification and evaluation of real-time high-resolution forecasting, and local U.S. Navy groups (e.g.,Whidbey Island Naval Air Station) have given us impor-tant feedback and support. We acknowledge the coopera-tion of NCEP, the Fleet Numerical Meteorological andOceanographic Center, the Australian Bureau of Meteorol-ogy, the Canadian Meteorological Center, the Taiwan Cen-tral Weather Bureau, and the UK Met Office in providingoperational access to their model grids. Extraordinary dis-counts provided by SUN Microsystems have greatly facili-tated the local prediction effort. Finally, the invaluable as-sistance of department system computer staff (H. Edmon,M. Michelsen, D. Warren) was essential for creating and

maintaining the complex hardware/software infrastructurerequired for such a regional prediction center.

REFERENCES——, C. F. Mass, and K. J. Westrick, 2000: MM5 precipi-

tation verification over the Pacific Northwest duringthe 1977–99 cool seasons. Wea. Forecasting, 15, 730–744.

——, ——, and D. Ovens, 2001: Evaluation of the tim-ing and strength of MM5 and Eta surface trough pas-sages over the eastern Pacific. Wea. Forecasting, 16,553–572.

Grimit, E. P., and C. F. Mass, 2002: Initial results of amesoscale short-range ensemble forecasting systemover the Pacific Northwest. Wea. Forecasting, 17,192–205.

Mass, C., Y.-H. Kuo, 1998: Regional real-time numeri-cal weather prediction: Current status and futurepotential. Bull. Amer. Meteor. Soc., 79, 253–263.

——, D. Ovens, M. Albright, and K. Westrick, 2002:Does increasing horizontal resolution produce bet-ter forecasts? The results of two years of real-time nu-merical weather prediction in the Pacific Northwest.Bull. Amer. Meteor. Soc., 83, 407–430.

Sharp, J., and C. F. Mass, 2002: High-resolution forecastsfor the Columbia River Gorge. Bull. Amer. Meteor.Soc., 83, 1757–1762.

Stensrud, D. J., H. E. Brooks, J. Du, M. S. Tracton, andE. Rogers, 1999: Using ensembles for short-rangeforecasting. Mon. Wea. Rev., 127, 433–446.

Stoelinga M. T., P. V. Hobbs, C. F. Mass, J. D. Locatelli,B. A. Colle, R. A. Houze, Jr., A. L. Rangno, N. A.Bond, B. F. Smull, R. M. Rasmussen, G. Thompson,and B. R. Colman, 2003: Improvement of Micro-physical Parameterizations through ObservationalVerification Experiment (IMPROVE). Bull. Amer.Meteor. Soc., in press.

Vaughan, J., and Coauthors, 2003: A numerical daily air-quality forecast system for the Pacific Northwest.Bull. Amer. Meteor. Soc., in press.

Westrick, K., and C. Mass, 2001: An evaluation of a high-resolution hydrometeorological modeling system forthe prediction of a cool-season flood event in acoastal mountainous watershed. J. Hydrometeor., 2,161–180.

——, P. Storck, and C. F. Mass, 2002: Description andevaluation of a hydrometeorological forecast systemfor mountainous watersheds. Wea. Forecasting, 17,250–262.