When we generate spatial analysis plots and compare two models’ results using AMET, not every station data is plotted on both plots. But rather, based on my evaluation of the source code, if the difference between the observation and model is greater than the error threshold (which looks like +/- 3 degree for temperature, +/- 3 g/kg for mixing ratio), that station is not plotted on the graph. Is there a way to plot “all” data points, regardless of the difference between the observation and model data on the graph, so we can see larger variations beyond the threshold level? Can we choose the threshold levels in the “run_spatial_surface.csh” script? The threshold level in that script is:
------------ begin cut ---------------------
And threshold of number of data points at a
single obs site, below which a site’s statistics are NA.
This excludes sites with limited obs.
24NDAYS0.5 would only do sites with 50% of data over
number of days
setenv THRESHOLD 6
----------- end cut ------------
refers to total number of days. Can we use this parameter to control plotting all data points, or do we need additional parameters?
Are you seeing this behavior with just the run_spatial_surface.csh script, or with other scripts as well?
AMET should not be applying any criteria to remove sites by default (all sites should be plotted). The exception is on the AQ site of AMET where a completeness criteria is applied for spatial plots. By default, this threshold is set to 75%, meaning that at least 75% of the obs must be non-missing for a site to be plotted. I’m not sure if any similar criteria is applied on the Met side of AMET.
The only other reason I can think of that you may see sites not being plotted is if the value to be plotted falls outside the set plotting range. On the AQ site, by default the scale is set to capture all values, with extremes plotted as gray in some cases. But all sites should be plotted. Not sure how it works on the Met side of AMET.
We have discussed this via email a few weeks back.
Wyat #2 above is correct. In this case WRF had very poor model performance in areas in and around NV. The stats were not plotted because they fell outside default range/levels for the statistics. In this case bias. MPAS had much lower bias and was plotted. I showed how to alter default bias, error and IOA plotting levels in the spatial_surface.input file.