8.4 Air Concentration Statistics




In this section we will review how the HYSPLIT binary air concentration file can be converted to the DATEM format which is compatible with several statistical analysis programs that can be used to compare the measured and calculated values. This is a continutation of the previous two sections, which should have been completed before proceeding with this section.

  1. A variation of the interpolation to a location is the Utilities / Convert to DATEM tab which opens a similar options menu which converts the HYSPLIT output to the DATEM (Data Archive of Tracer Experiments and Meteorology) format by matching each value in the measurement file with a calculated value. First browse to find the measurement file: /Tutorial/captex/captex2_meas.txt. Then press Create DATEM file and the calculated data file hysplit.txt is created in the working directory. Pressing Compute Statistics analyzes the data and opens the GUI viewer to the results file: statA.txt. The scatter plot button opens a graphic showing the measured and calculated values identified by station number. The paired measured and calculated values are also written to the file: dataA.txt.

  2. Now that we have a baseline performance statistics, it is possible to test various changes. For instance, instead of just using the NARR meteorology, we can also include the high resolution WRF (3 km every 15 min) model. If the configuration has been changed within the GUI, retrieve the captex_control.txt and captex_setup.txt files into the menu. Open the Concentration Setup menu and add (do not clear NARR) the meteorology file captex2_wrf03.bin which includes data over the Ohio region through 1200 UTC of 26 September. Save the changes and run the model. Although both meteorological data files are available for the first 19 hours, the HYSPLIT calculation will use the finer resolution of the two until 1200 UTC of the 26th and then switch to the NARR.

  3. Warning this run may take at least 10 minutes! After the run completes go through the first step to recompute the performance statistics. The new results shown in statA.txt indicate that when using higher resolution meteorology, at least during the first day, the model performance improved for several of the metrics: correlation (0.57 to 0.65), figure-of-merit (64% to 70%), NMSE (19 to 18), and RANK (2.82 to 2.89). The improved results are also visibile in the scatter diagram which shows more points near the 1:1 line than the NARR only calculation.

The comparison of model results with measured data can be used to test various model options and maximize the model's performance given various user set parameters, such as particle release rates, concentration grid resolution, and different model parameterizations. In addition to CAPTEX, experimental data from nine other experiments, are available from the DATEM web site.