Important caveats regarding meteorological data used to ...



What Tim wanted…

I suspect there is confusion about two things. First, what happens if a forecast office calls ARL and asks for a model run - what model is used. And second, how well can the model (presumably ETA12) replicate near surface flows.

and how well it is likely to perform in different situations.

This will be inserted as a button that NWS forecasters can select - and it will go to a short presentation by you or Bob. I say Bob just because he is here and might be able to record it sooner than you.

This seems to have some urgency in NWS.

I suggest that you contact Stokols for more on procedures.

Introduction:

The National Weather Service is part of the front line of defense when toxic chemical, biological, or radiological releases into the atmosphere occur. Because the dispersion of such substances is highly dependent on weather conditions, emergency response teams depend on the weather service for guidance and information on how quickly a toxic substance release will be transported, and at what concentrations.

For small scale (0-10 km), short duration (less than one hour) releases of hazardous material, the CAMEO/ALOHA dispersion model, described in the web cast on Atmospheric Dispersion Models, is available to many local jurisdictions, including emergency responders.

For large scale (10 km or greater, but less in a light wind regime), long duration (greater than one hour) releases, the HYSPLIT model is used. Typically, HYSPLIT is run at NCEP or by WFOs, and the information is then passed on through a secure connection to the emergency manager needing the information.

The purpose of this supplemental webcast is to:

• Review when the HYSPLIT dispersion model is useful and when it is not

• Explain the procedure for obtaining special HYSPLIT runs from the Eta-12 at its native grid resolution from NCEP

• Discuss the potential shortcomings of the Eta-12 in forecasting near-surface conditions as a result of meteorological data interpolation and model forecast errors, and

• Review a case where problems with the Eta-12 forecast results in less than optimal HYSPLIT model results, to illustrate the kinds of problems that can result from Eta-12 model error.

Quick review of situations calling for use of HYSPLIT

To quickly review, the HYSPLIT model is used in instances where:

• The materials released are neutrally-buoyant

• An estimate is needed of air concentration or exposure over time at specific locations or areas downwind

• Temporal and/or spatial variations in meteorological conditions are expected

• Estimates of dry and/or wet deposition of the pollutant downstream from the source are required

• A forecast of pollutant plume position using the most recent NWS/NCEP forecast data is desired

Substance concentrations from point or line sources of pollutant in HYSPLIT can only be well-predicted at a minimum distance greater than 1/4 the resolution of the data driving the HYSPLIT model (or about 10-km, whichever is more). For example, in the case of the Eta-12 at its native grid resolution, the minimum distance would be 10 km or more from the point or line source. If interpolated 80 km fields from the same Eta-12 are used, however, good forecast concentrations can be obtained only at distances of 20 km or more from the point or line source.

HYSPLIT model output can be obtained in any of three ways:

1. By running the model locally at the weather forecast office (WFO). The model code can be obtained from the NOAA Air Resources Lab (ARL) web site at ,

2. By accessing the non-operational READY web site at the NOAA ARL, web address , or

3. By contacting the National Weather Service’s National Centers for Environmental Prediction, or NCEP, senior duty meteorologist (SDM) to request an emergency HYSPLIT run.

Running HYSPLIT locally results in a quick turnaround time, but the meteorological data available to run the model is at relatively coarse horizontal resolution. This is also true of HYSPLIT calculations on the READY site. Additionally, the READY site is not maintained on a 24 hour per day, 7 day per week schedule, and in the case of multiple emergencies, may be very slow to respond. On the other hand, NCEP provides 24 hour per day, 7 days per week HYSPLIT run capability to support the emergency management community, using the most recent 12-km operational Eta model forecast data at its native grid resolution to give a fast initial response. Additionally, a higher resolution mesoscale model over a preset domain surrounding the release site is run to provide more detailed meteorological data for an additional HYSPLIT run. These preset domains can be seen in the graphic.

Graphic: nmm_hysplit_regions.gif

[pic]

WFO procedure for requesting a HYSPLIT run

A weather forecast office (WFO) will generally be informed of an atmospheric release of hazardous material and asked for a dispersion forecast by an emergency manager or another federal agency such as the Nuclear Regulatory Commission (NRC). The senior forecaster on duty at the WFO must follow these procedures to request an emergency HYSPLIT run from NCEP:

1. Obtain from the manager or NRC the necessary information on the hazardous substance release, including

a. Location (latitude, longitude, or nearby airport)

b. An identifier for the incident

c. Size and or height of release

d. Time release began

e. Duration of release (if significantly different than one hour.) An expected duration should be estimated if the release is ongoing.

f. Other information about particle type, size, chemical composition of release, etc. are optional.

Contact information for output dissemination to the emergency manager should also be obtained (fax number, ftp address, e-mail address for attachments, etc.).

2. Call the NCEP Senior Duty Meteorologist (SDM) at NCEP (e-mail: sdm@).

3. Provide the SDM with emission release parameters. The following parameters are required:

a. Location of emission, latitude and longitude or a nearby airport

b. Starting time of emission

c. Duration of emission (if significantly different from one hour default duration)

d. Height and/or vertical extent of emission

Additional optional parameters are:

e. Particle size (diameter in microns)

f. Particle density (in grams per cubic centimeter)

g. Emission rate (in units of mass per hour)

h. Deposition velocity (amount of pollutant deposited on the ground per unit time)

If the optional parameters are not known, HYSPLIT will run with default values of particle size 0.0, particle density 0.0, a 1.0/hour emission rate, and 0.0 deposition velocity. These default values will result in treatment of the pollutant like a gas. For discrete particles such as soot, dust, or smoke, particle size and density should each be set at 1.0 if these values are not known.

4. The SDM will notify WFO when the HYSPLIT run is complete and the output is available on the web over a secured connection. The SDM will provide the WFO with access information, including password, for obtaining graphical (and raw data) output. Output should be available on the NCEP secure web site within 10-15 minutes of the call.

5. The WFO will obtain data and graphical output from the run within a few minutes directly from the Web site, using the access information from the SDM.

a. Output data on the web is at 0.05° (about 4-km) resolution for the first 12 hours, and 0.25° (about 20-km) resolution for later valid forecast times.

b. Graphics of trajectories from the release location and plume/concentration details are provided on the web.

c. Click on the day you want to view, then the state (or the identifier assigned above) and run time for GIF or Geographical Information System (GIS) downloads. The experimental graphical user interface also includes postscript (PS) files, compressed GIS files in TAR format, and the ability to loop GIF files.

6. The WFO will send the information obtained to the requestor using the contact information previously obtained.

Introduction to limitations on meteorological data

For ground-based hazardous material releases, near-surface meteorological conditions, including the 10-meter winds, 2-m temperatures, and surface fluxes (which determine the stability and thus the vertical dispersion of material) are important determinants of the horizontal and vertical spread of the material. NWP forecasts of these meteorological variables are inputs to the HYSPLIT model, but the NWP models are limited in a number of ways in their representation and forecast of near-surface conditions. Frequently, the meteorological forecast data available to run HYSPLIT has been spatially averaged, resulting in representativeness problems for modeling dispersion in, for example, areas of complex terrain.

Additionally, all meteorological data used to force HYSPLIT is available only at coarse temporal resolution, which can result in errors in rapidly changing conditions, even if the NWP model itself can faithfully emulate these conditions. NWP models also have limitations in their ability to accurately forecast the onset and spatial scales of phenomena that may affect dispersion, such as sea-breezes and mountain-valley circulations. Finally, the meteorological data provided is bi-linearly interpolated in space and linearly interpolated in time to the location and time of a hazardous material release and during the forecast of the subsequent trajectory and diffusion of materials. If important variations in the meteorological fields occur between grid points or between meteorological data output times, trajectory and concentrations errors may result.

The users of HYSPLIT trajectory and plume output data must be aware of these limitations and their potential effect on the HYSPLIT forecasts. The purpose of this section is to illustrate the above limitations using one of the important meteorological data used in HYSPLIT; the 10-meter winds.

Effect of interpolation on meteorological data

Let’s look at the effect of horizontal interpolation of meteorological data from the native grid to coarser grids. First, we’ll look at a graphic of 10-m winds from the Eta-12 at different spatial resolutions in a relatively uniform wind field, centered on Wausau, WI, from a 6 hour Eta-12 forecast valid at 18 UTC 14 March 2003.

(Graphic: WI_eta204080.gif)

[pic]

Topography in this area is relatively uniform, and the synoptic scale situation results in steady south to southwesterly flow over the entire domain. The smaller red wind barbs represent the 10-m winds from the Eta-12 interpolated to 20-km resolution, while the pink and blue barbs the same winds interpolated to 44-km and 80-km resolution, respectively. Note that the wind field is little-changed as a result of the interpolation at differing resolutions, because of the relative uniformity of the winds over the domain. Also because of the relatively uniform wind field, the bi-linear interpolation done by HYSPLIT to a hazardous material release point or trajectory would not result in significant interpolation errors. All else being equal, similar forecasts should result from HYSPLIT regardless of the resolution of the meteorological data used.

What effect does interpolation have on 10-m winds in areas with more complex wind fields? An example from the topographically diverse Puget Sound area is shown in this graphic, from a nine-hour forecast valid 21 UTC 13 March 2003. We show the same interpolated wind fields over this domain as in the Wisconsin graphic, using the same color conventions. Over the Pacific Ocean, 10-m winds at all resolutions are consistent. However, there are differences among the wind fields in the Strait of Juan de Fuca, where the 80-km wind field completely misses the easterlies coming through the Strait, and the wind velocity and direction in the 44-km field are changed from those at 20-km. More subtle speed and direction variations induced by terrain and evident at 20-km are also smoothed out, such as along the east slopes of the Washington Cascades and around Mt. Rainier southeast of Seattle. (Highlight each as they are mentioned in graphic.) Spatial interpolation from the 80-km grid will lead to errors in these areas.

(Graphic: pnw_eta204080.gif)

[pic]

The effect of interpolation of the 10-meter winds to a release point can be seen by comparing the 20-km winds to the 44- and 80-km resolution winds. As one would expect, the 10-m wind errors introduced by interpolation are particularly large in the same areas where the wind fields show the largest differences from each other.

Ability of Eta-12 to forecast near surface winds

Now that we’ve covered potential effects of interpolation on the meteorological data, the next question is, how well does the Eta perform in forecasting 10-m winds? The following graphic shows the verification for the same time as above indicated by large dark green barbs, against the 10-m winds from the 20-km area averaged grid of the 12-km Eta in smaller red barbs. It appears that at least some aspects of the wind field were captured fairly well, including the topographically influenced winds in the area surrounding the Puget Sound. There are problems resolving the topographic effects of the Olympic Mountains and Vancouver Island in the Strait of Juan de Fuca, however, with for example, 20-kt westerly winds on the south shore of Vancouver Island where the model shows east to southeast winds at about 10-kts. Gap winds through the Columbia Gorge between Washington and Oregon are also not well forecast by the Eta-12. It also appears that in this case the 10-m winds are generally too strong, at least over land.

(Graphic: pnw_eta.gif)

[pic]

Verification for the WI 10-m wind forecast we showed previously can be found in the graphic below. Here we see sub-grid scale effects on wind velocity and direction that the Eta model cannot pick up (even at 12-km). Note the observed lake breezes on the east coast of WI adjacent to Lake Michigan and in the northwest corner of WI off of Lake Superior, neither of which seem to be adequately depicted. Also note that there seems to be no obvious systematic positive or negative velocity bias in this case.

(Graphic: npl_etaver.gif)

[pic]

Squared speed errors in the Eta-12 for forecast 10-m wind are computed by NCEP and posted on the web for each month. An example from December 2002 is shown below for the Northern Midwest, which includes WI. On average, the forecast 10-m wind velocity over the northern Midwest region has a positive bias, which appeared during May 2002 and has persisted at least through December.

(Graphic: V10M_00_diurnal_DEC2002_NMW_Eta.jpg)

[pic]

Monthly wind biases for different US regions can be monitored through the NCEP Mesoscale Modeling Branch website at emc.ncep.mmb/research/nearsfc/statsbyregion.html

Example of the effect of near-surface wind errors on HYSPLIT forecast

What kind of effect do 10-m wind errors have on HYSPLIT results? To examine this question, we will use a case where the Eta-12 was predicting that a cold-air damming event affecting the Washington DC area would end on the morning of March 21st, 2003, resulting in strong south to southwesterly winds developing at the surface. In actuality, the warm air above the shallow cool layer of air near the ground never mixed down to the surface, leaving the area in a weak easterly flow through much of the afternoon of March 21st. Above this shallow cool layer, there was indeed a southwest flow of warm air, as can be seen in a special sounding taken at Washington Dulles International Airport at 18Z.

(Graphic: 2003032118.72403.skewt.gif)

[pic]

The model forecast was quite different because of a problem with its simulated clouds allowing too much solar radiation through to the surface. Additionally, the current version of the Eta-12 clears cloud water from the lowest 100-mb of the grid column, where much of the actual cloud was located. The Eta-12 BUFR sounding for Dulles from the 12Z March 21 2003 run, valid at 18Z on March 21, 2003 shows a predicted shelter temperature and dewpoint about 7( and 2(C warmer than the verification, with predicted southerly winds mixed down to the surface.

(Graphics: iad_eta.gif)

[pic]

The 10-m wind forecast and verification for the region around Washington DC at 18Z on 21 March can be seen on the left and the right in this graphic. The barbs represent wind direction and speed in knots.

(Graphics: fcst.gif, sfcob.bay.10mw06.gif)

[pic][pic]

Note that the actual winds in the DC area were light and variable, rather than from the south-to-southwest at 10-15 knots.

To test the effect of the error in predicting the demise of the cold air damming situation on the prediction of trajectories and pollutant concentrations in HYSPLIT, we ran two 24-hour simulations for a 10-m elevation, three-hour hazardous substance release at Washington Reagan International Airport (DCA) at 18Z on March 21. One HYSPLIT run used meteorological data from the Eta-12 forecast at 12Z March 21 (interpolated to 40-km), while the other used data from the Eta Data Analysis System or EDAS also interpolated to 40-km, which we will use as a surrogate for the observations. It should be noted, however, that while better than the forecast, the EDAS 10-m wind field at 18z was adversely affected by the Eta forecast error previously discussed and still has too much southerly flow. This can be seen in the graphics below. Therefore, the true error in trajectories and concentrations would have actually been even worse than those we will see.

(Graphics: anal.gif, sfcob.bay.10mw06.gif)

[pic][pic]

An important improvement in the analysis, however, is in the strength of the inversion at the top of the cool layer, which can be seen by comparing soundings from the forecast versus the analysis below 850-hPa for the Eta-12 grid point nearest to DCA.

(Gaphics:kdca_sndf006.gif, kdca_sndanal.gif)

[pic][pic]

Better emulation of the inversion will trap the hazardous material in the cool layer where the winds are weaker, and limits the northward and eastward spread of the associated plume as a frontal system and wind shift to the west moves through the area between 00 and 01Z on the 22nd.

The resulting difference in 24-hour trajectories can be seen in the graphics below, with the forecast trajectory on the left and the analysis trajectory on the right. These graphics and the pollutant concentration plots which will follow were generated using the NOAA Air Resources Lab Real-time Environmental Applications and Display sYstem (READY). The forecast and analysis data were only available at 40-km resolution, but the results are not significantly affected. Six hourly positions for the trajectory are marked with red triangles, and the maps were scaled differently because of the smaller trajectory in the analysis.

(Gaphics: dcaeta_traj001.gif, dcaedas_traj001.gif)

[pic][pic]

Note that the analysis trajectory only moves about 1/2( north of DCA, and then moves eastward after the winds shift to the west-northwest after the frontal passage, reaching the upper eastern shore of MD at 12 hours, Cape May at 18-hours, and then subsequently out to sea. On the other hand, the forecast trajectory has move a full degree northward at 6 hours, and then does a more gradual turn to the east, located just east of Trenton NJ at 12 hours and then subsequently out to sea. By 24 hours, the forecast trajectory is about about 39.5N, 70W, while the analysis trajectory is only at about 38.8N, 72.9W, a difference of about 235-240 km! Also, note that the location of the trajectory in the vertical is also different. The analysis indicates a trajectory that better hugs the ground, consistent with the trapping of a release in the existing cool near-surface layer. The forecast, on the other hand, lifts the trajectory up to about 300-m at 6 hours, and the trajectory at all subsequent times is higher than the analysis trajectory.

What effects does the Eta-12 error have on concentrations in the hazardous material plume? Animations of the forecast and analysis concentrations were made and are shown below. The frames, which start at 20Z March 21, are two-hour averages of the concentration of pollutant over the same 24-hour period as for the trajectories above. The pollutant in both cases is initially a line source from 10-100 meters in height, and assumes the same rates for dry deposition (settling of particles due to gravity) and wet removal (washout by precipitation). Because the source extends from 10 to 100 meters and because of dispersion processes, the plumes actually make more progress than the trajectories initially at 10-m might indicate. Concentrations are as indicated in the color code in the bottom of the graphic. Units are fraction of total quantity of hazardous substance released (3 units in this case). The maximum concentration in the plume is indicated by the red square, with the value indicated next to the concentration color coding.

Graphics: dcaeta_cloop.gif and dcaedas_cloop.gif

[pic][pic]

Even after two hours, we note some differences in concentration and extent of the plume. The maximum concentration in the analysis plume is 7-8 times greater than in the forecast, and there has been somewhat more dispersion to the north and east in the forecast plume.

The differences in plume location grow at each two-hour interval. By 00Z on March 22 (frame 3), the forecast plume extends over the PA line west of Philadelphia and has reached the Allentown/Bethlehem/Easton PA area and the NJ state line. The analysis plume, on the other hand, has drifted more eastward over northern DE, southeastern PA and southwest NJ and has just reached the southern part of Philadelphia. Concentrations seem to be higher, with the maximum concentration between Washington and Baltimore at about 4 times the forecast concentration, but about the same location.

By 02Z, the forecast plume has reached New York City, while the analysis plume is just reaching the southern and central NJ coast. The forecast and analysis concentration maxima are near Lancaster PA and just south of Baltimore MD, respectively, with the analysis concentration 20 times that of the forecast.

By 06Z, the concentration maximum is just south of Long Island in the forecast plume, which now extends from the Philadelphia area to the eastern tip of Long Island. The analysis plume, on the other hand, has a maximum concentration, still about 20 times that of the forecast plume, just east of Atlantic City. The plume is still affecting the eastern part of the Washington DC metropolitan area and extends eastward over the southern 2/3 of NJ into the Atlantic. However, NYC is outside the northern edge of the plume.

It takes until 14Z on March 22 for the analysis plume to completely pass off the NJ coast. The upwind end of the forecast plume by that time is leaving the eastern tip of Long Island.

Case summary

We have examined a case where a bad Eta-12 forecast results in significant differences in the trajectory and concentration of hazardous substance forecast by the HYSPLIT dispersion model. The Eta-12 error was limited to the lowest 100-hPa, and resulted from eroding a shallow layer of cool air resulting from cold air damming too quickly in the Washington DC metropolitan area. This affected both 10-m winds, the stability of the planetary boundary layer or PBL, and the existence and intensity of a temperature inversion at the top of the cool air.

All these near-surface factors came into play in the prediction of the trajectory and the concentrations in the plume downwind from the pollutant source. In the run forced by the model forecast, a vigorous PBL was erroneously forecast to develop, with south winds forecast to mix down to the surface. These errors were attributed to Eta-12 model problems with cloud interaction with solar radiation, resulting in too much solar radiation reaching the surface. The resulting hazardous material plume traveled further north, and showed lower concentrations because of dilution of the material through a larger depth, plus additional mixing from above the PBL.

The run force by Eta analysis data, however, better (but still not perfectly) emulated the persistence of the near-surface cool air and the inversion, resulting in slower and further south downwind transport, less mixing into the plume from above because of the strong temperature inversion, and thus higher concentrations of pollutant within the plume.

The higher concentrations and longer residence time of the hazardous substance near its source in the analysis run highlights the importance of assessing even short-term model forecast performance in predicting near-surface wind and stability fields.

General Summary

We’ve reviewed when the HYSPLIT dispersion model can be used to predict both the trajectory of a hazardous material release and the resulting plume concentrations, and how to obtain HYSPLIT runs from NCEP and the SDM.

Additionally, we’ve discussed Eta-12 meteorological data as it is used to force the HYSPLIT model, including the effect of:

• Meteorological data interpolation in time and space and

• The effect of model forecast errors on dispersion model results.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download