13.5 Source Location Statistics




In the previous section, we configured the model to simultaneously calculate the dispersion from a regular array of source locations, keeping the results independent of each other. Given that measurement data are available, it is possible to compute the performance statistics of each source location to determine the location with the best performance. If not continuing from the previous section, load the saved files matrix_control.txt and matrix_setup.txt and rerun the calculations if needed. At a minimum, the internal GUI variables need to be set accordingly.

  1. The comparison could be done manually, but there is a menu option to invoke an automated procedure. Press the Display / Source-Receptor / Stats menu tab to open the statistics menu. The source-receptor file srm.bin created in the last section is already defined in the GUI. In step 1 define the measured data file, because the measurements are in pico-grams set the units conversion factor at least to 1.0E+12 but even better would be 1.0E+15 so that the output becomes more realistic. Edit the output file names (steps 2 & 3) if required, then press the Execute button of Step 4 to generate the statistics for each source location. This will take a little time as each source location is extracted from the binary output file and the results are accumulated in the statistical summary text file. When processing is completed, the termination message file will be displayed.

  2. Select the Corr radiobutton to output the correlation coefficient as the primary statistic for viewing. The other statistical metrics can be influenced by the conversion factor (i.e. source term) but not the correlation. Press the Execute of Step 5 and the map of the correlation coefficient shows that the highest value (0.85) is associated with the release location at 43N 75W, matching the result from the last section. See the output file sumstat.txt from Step 4 for all the statistics associated with each release location.

  3. To improve the estimate of the emission rate from 43N 75W, we can use the concentration file SRM_fwrd.bin already extracted for this location in the last section. Open the Concentration / Utilities / already Convert to DATEM menu and enter the concentration file, measured data file, and the same conversion factor, or source-term in pg because the calculations used a unit-emission rate. Create the DATEM file and then compute the statistics which shows that the ratio of calculated to measured is 0.21 suggesting that the source term (conversion factor) needs to be around 5000 g/h rather than the 1000 g/h to remove the bias. However, a review of the scatter plot for the 1000 g/h case suggests that although the bias can be removed, a significant number of sampling locations will move toward over-prediction if all the points are shifted upward by a factor-of-five.

In this example, we used the measurement data in conjunction with model simulations from multiple locations to determine which location provides the best fit with the measurements. Determining a more precise estimate of the emission rate at the source location was complicated by some uncertainty in the model predictions. In the next section, assuming that the release location is known, we will try to determine the emission rate through more quantitative approaches that address model errors.

7 s