14.1 Fire Smoke

Previous

HOME


Next

Although it is fairly simple to configure a simulation that follows the smoke from one or more fires that are represented by point sources, it is a little more complicated to predict PM 2.5 air concentrations. This additional step requires information of the fuel loading as well the burn area. In a prescribed burn simulation, the area to be burned is known. In a wildfire situation, we need to estimate the area burning. In this example, we will try to simulate a wildfire that occurred near the Okefenokee Swamp in Southeast Georgia during the week of 16 April 2007.

  1. The fire started due to strong winds on 16 April 2007 toppling a tree over a power line. By the next morning, the fire locations were easily detected by various satellite monitoring systems. Although the data for this fire are no longer on-line, the procedure can be used for current fires. All relevant files for this example have been saved in the \Tutorial\smoke directory. Examining the meteorological data on the 16th suggests the fire may have started around 1800 UTC near the time of the maximum wind speed and the satellite HMS fire detects for the next day shows about 20 locations around 31.15N 82.40W.

  2. Before proceeding, if this tutorial is being run through the web, it will be necessary to download the WRF meteorological data files for the 16th through 20th of April for this example. Any pre-configured CONTROL files used in this example will need to be edited to point to the location of the downloaded data file. For convenience, an extract of the five daily WRF files have been merged into a single file in the smoke directory.

  3. The next step is to convert the general information we already know about the fire into some specific estimates of the emission. After entering the latitude and longitude, the READY web site has an option under the dispersion simulation data entry page to do a prescribed burn calculation. The details of the simulation are configured on the next page, where the area to be burned should be entered. However, this being a wildfire, that information is unknown. If each satellite fire detect represents one pixel or about 1 km2, we make the assumption that only 10% of that pixel area is burning, then each pixel represents 100,000 m2 and with 20 fire detects, the total area would be 2,000,000 m2 which should be entered into the burn area text entry box. Submit the dispersion run and immediately the simulation input files become available. Download the emission rate EMITIMES file. The emissions file contains values at 3 minute intervals and find the row with the maximum emission rate:

    • 2007 04 16 01 09 0003 31.150 -82.400 0. 0.24E+09 0.20E+07 0.86E+11

    These calculations are a component of NOAA's Air Quality Forecast System and the emission processing is a component of the Blue Sky modeling framework. Note that a more recent version of the emission processing routines are used for the operational forecast.

  4. The value of 0.24E+9 g/hr will be used as the constant emission rate for the simulation. This value could be entered directly into the CONTROL file, but instead we will use an edited version of this EMITIMES file so that the information on the heat released can be used by the model to compute the plume rise. A more detailed discussion of the plume rise options can be found in the Advanced Tutorial. Edit the file by removing all lines except the first three and the data record shown above. The data record should be valid for 96 hours, with emissions starting at 18 UTC and valid for 72 hours and change the units of the emission from g/h to ug/h (x 1.0E+06) so that the model output can be compared directly with monitoring data. The resulting file should only contain four records.

  5. Now there is enough information to configure the HYSPLIT input files by either retrieving the smoke_control.txt and smoke_setup.txt files into the menu, or manually setting the required information. Note that the namelist was set to do a Gaussian-particle simulation, setting the particle number to 10,000 but limiting their age to 24 hours, and reading the EMITIMES file. The Gaussian option was selected to provide somewhat smoother transitions between hourly output periods. After all the changes have been made, save the results to close the menus and run the model.

  6. Hourly maps can also be plotted from the concentration display contours menu by setting the display units to ug and forcing the user set contours to 500+200+100+50+20+10 with the no contour line option. Running this will create 96 graphic frames. A MODIS AOD satellite image from April 17th clearly shows the smoke plume. The model prediction from 15-16 UTC on the 17th is the nearest in time to the MODIS pass (near solar noon). Comparing the model prediction and observation side-to-side shows a good match to plume position and extent. A batch file is provided to process just the one output frame valid at the time of the MODIS pass.

  7. The model results can be compared with PM2.5 monitoring data from the AirNow network. The hourly data from the Mayo Clinic (30.2556 -81.4533), a site near Jacksonville Florida, has been reformatted to be more easily displayed. After the run completes, open the concentration utilities Convert to station menu and set the Mayo site latitude-longitude, concentration multiplier to 1.0, and set the supplemental measured data file to \Tutorial\smoke\mayo_pm25.txt. Press the yellow Extract Data button to extract the model predictions at that location and then press Plot Data to show the predicted (red) and measured (black) time series. The model predictions are very comparable to the measured data with predicted arrival times matching the observations almost exactly.

The results shown here used just one approach to determine the magnitude of the smoke emissions from a fire. There are other approaches such as the stand-alone Fire Emission Product Simulator that can be use to estimate the emissions from a variety of different fire scenarios, meteorological conditions, and fuel loadings. FEPS can be used conjunction with VSMOKE, a Gaussian dispersion model for computing very short-range smoke plumes. A further advantage of this approach is that VSMOKE automatically outputs a properly formatted EMITIMES file which can then be used for longer-range HYSPLIT simulations.

2 m 21 s     External Meteorology Links:    WRF200704.bin