In this section we will use the measurement data and the adjoint dispersion calculation (backward integration from the measurement location) to estimate the source location and magnitude of the emissions. Start by retrieving the previously saved captex_control.txt
and captex_setup.txt settings into the GUI menu.
- As in previous examples in this section, we will only use the sampling data from 3-hour duration collections. These occurred from 18Z September 25th to 12Z September 26th. Therefore, in the Concentration / Setup Run menu, change the run duration to 19 hours. Because we will be running the model for only 19 hours, fewer particles are needed. Open the Advanced / Configuration Setup / Concentration / Menu #4 and reduce the particle release number to 5000 to speed up the calculations. Save to close and exit all menus.
- Now press the Special Runs / Geolocation menu tab to open another data input menu composed of four steps. In this step you need to define the measured data file (DATEM format) that will be used to create the individual backward CONTROL files. You could manually create a subset of only the 3-hour sampling data from the master file captex2_meas.txt, but for convenience, this file is already provided captex2_3hr.txt. Use the browse button to find and select this file.
- In step 2 leave the defaults (measured value in the numerator of the source rate) and press the Execute button. After the completion message appears, 48 CONTROL files will have been created in the working directory corresponding to each of the measurements. As an example, file CONTROL.023 contains the information from station 318, where the sample was collected from 00Z-03Z on the 26th. Note this simulation, like all others, starts at 12Z on the 26th, running for -19 hours (negative = backward), with the particle emissions occurring from 03Z to 00Z. In this way all model simulation files contain the same number of output time periods. The emission mass equals the measurement. Zero measurements are assigned a nominally small emission rate (10-11). Now Execute Step 3 to run the dispersion simulations.
- When the calculations have completed, Execute Step 4 with the default menu values to see the six frames (one for each time period) of the weighted source sensitivity function, the mean of the 48 simulations. The last frame shows the pattern for the time period corresponding to the tracer release. These patterns are weighted by the measurements so that nearby samplers, with the highest measurements and smallest model dilution factors, provide the greatest weight to the pattern.
- As noted in an earlier section, the model computes C = D Q, where C is concentration, Q is the emission value (mass) and D is the model dilution (m-3) factor. Although not precisely correct, we must assume that D forward, the dilution factor from the source to the receptor, is the same as D backward, the dilution factor from the receptor to the source. Then the source term can be estimated from measurements (M) as the ratio of M/D:
However, the model only computes D in the numerator, not its inverse. If we set the source term as 1/M for the calculation, the model will compute D/M which equals 1/Q.
- You may want to run Special Runs / Ensemble / Cleanup before starting this subsection. To perform the 1/M calculation, it is necessary to repeat Steps 2 and 3, but this time selecting the Inverse radio-button. This will set the emission rate as the inverse of the measured value, an example of which is shown in the control file (004) for station 318. Only the 22 non-zero measurements are used in this approach.
- Rerun the calculations and display, and now the frame corresponding to the emission time shows a more complex pattern, with the smallest values corresponding to the larger emission rates. For instance, the grid location near the known source location (40N 84W) shows a value near the 10-17 contour. Use Convert to Station to determine that the exact value from input file cmean is 3.3x10-18. The inverse would be 3.0x1017 pg or about 300 kg.
- The post-processing of the source attribution output files results in several different files (prob??, cmax??, cmean) in the working directory which are discussed in more detail in the hysplit help files. In this application, all the graphics are created from cmean, the mean value at each grid point for all the output files with the .000 suffix.
The results shown here use measurement data in conjunction with the dilution factors computed by HYSPLIT to estimate source location and magnitude. For this particular example, the flow was relatively uniform so that all the samplers were in the same general downwind direction, resulting in estimates of source area that were larger than if we had sampled multiple releases that went in different directions. Multiple downwind directions would result in a better triangulation of the source region and a smaller area of uncertainty about the possible source location.