13.3 Physics Ensemble




Another component of the uncertainty in the concentration calculation is the variation that can be introduced by using different physical parameterizations, or model options. Most of these were already reviewed in the concentration modeling sections. The model normally runs with its default options, unless a SETUP.CFG namelist file is found. Although various physical parameterization options were reviewed in previous sections, the actual namelist parameters or their values were not discussed in detail. This information is available through the HELP menu button. In the model physics ensemble, a script is run that automatically sequences through the most common parameter variations. This automated process will be reviewed in this section. Start by retrieving the saved ensemble_control.txt and ensemble_setup.txt settings into the GUI menu. For this simulation, from the Setup Run menus, change the release height back to 10 m and change the output file from ensemble to a unique name such as ensphys.

  1. Before running a simulation, again delete all the ensemble files left over from the previous section. The Special Runs / Ensemble / Physics menu will sequence through 20 different calculations starting with the 3D-particle option, Gaussian and Top-Hat puffs, variations in how the mixed layer depth is calculated, various turbulence and stability equations, varying the ratio of vertical to horizontal turbulence, and varying the Lagrangian time scale. Note that these variations are hardcoded into the script and not all combinations may be valid for different meteorological data.

  2. The physics ensemble menu will stay open as long as the model is running. WARNING - it may take 2 hours or more to finish all 20 simulations using 10,000 particles. You may want to reduce the particle number release rate as in some of the previous examples to speed up calculation. Open the Advanced / Configuration Setup / Concentration / Menu #4 and change the particle release per cycle from 10000 to 2500.

  3. Press Execute Script to start the process. The puff simulations (002, 003, 004) are much slower than the other variations. As each simulation completes, the name and value of the namelist variable will be shown in the text box. The following variations are programmed in the script:

    • {file}.001 : initd = 0      initial distribution is particle (default=0)
    • {file}.002 : initd = 3      distribution gaussian horizontal-puff vertical-particle
    • {file}.003 : initd = 4      distribution top-hat horizontal-puff vertical-particle
    • {file}.004 : kpuff = 1     empirical puff growth rate (default=0 linear)
    • {file}.005 : kmixd = 1    mixed layer from temperature profile (default=0 input data)
    • {file}.006 : kmixd = 2    mixed layer computed from TKE profile
    • {file}.007 : kmixd = 1500 forced mixed layer depth to a constant value
    • {file}.008 : kmix0 = 50   minimum mixing depth (default=250)
    • {file}.009 : kzmix = 1     vertical mixing PBL average (default=0 use profile)
    • {file}.010 : tvmix = 0.5   vertical mixing scale factor (half as strong)
    • {file}.011 : tvmix = 2.0   vertical mixing scale factor (twice as strong)
    • {file}.012 : kdef = 1    horizontal turbulence from deformation (0 = scale to vertical)
    • {file}.013 : kbls = 2    stability derived wind/temperature profile (default=1 fluxes)
    • {file}.014 : kblt = 1    use Beljaars turbulence parameterizations (default=2 Kanthar)
    • {file}.015 : kblt = 3    use TKE field for turbulence (default TKER=0.18)
    • {file}.016 : tkern = 0.10    vertical turbulence anisotropy factor (0.18) = w'2/(u'2+v'2)
    • {file}.017 : vscales = 5.0     stable vertical Lagrangian time scale (default = 200 sec)
    • {file}.018 : vscaleu = 1000   unstable vertical Lagrangian time scale (default = 200 s)
    • {file}.019 : hscale = 5400     horizontal Lagrangian time scale (default = 10800 s)
    • {file}.020 : hscale = 21600   horizontal Lagrangian time scale (default = 10800 s)

  4. When the last simulation has finished, the calculations complete message will be shown. Then go to the Display / Ensemble / View Map menu and select the 90th percentile map and page forward to the 26th from 2100-0000 which shows a value greater than 10-9 near sampling station 510, Little Valley, NY.

  5. Now open the Box Plots menu and set the location for station 510 to 42.25 -78.80 which will show the box plots time series. The plot for the 27 03 (27th 00-03) shows the 90th percentile concentration (upper horizontal line) to be about 2x10-9, within the range shown on the map for that location. What the 90th percentile value means is that only 10% of the ensemble members (2) have concentrations greater than that value at that location.

  6. To determine which members produced the highest concentrations, look closely at the second graphic, the ensemble member plot, which shows that member 6 produced the highest concentration at that time period; the simulation where the mixed layer depth was computed from the TKE profile. As the script is running, the namelist variations for each member are written to the file ensemble.txt.

The results shown here represent a wide range of model options. The physics ensemble script (\hysplit4\guicode\conc_phys.tcl) could easily be customized for different model options depending upon the problem under consideration. As should be more obvious by now, there is never going to be just one right answer due to limitations in how well the meteorological data represent the true flow, variability from atmospheric turbulence, and how well the model parameterizations capture these processes. Do not delete the HYSPLIT ensemble output files, they will be used in another example.