I use WRF (Weather research and forecasting) model for in regional scale meteorology simulation. And then, the WRF output files will be deemed as criterial input data applying for chemical transport model(e.g CMAQ, CAMx).

The meteorology information are very important for chemical species in the atmosphere. It will influence its emission, diffusion, transportation and removal process and thereby determine the distribution of chemical in horizontal and vertical scales(healthy effect and climate effect related).

Since the importance of meteorology mode output data, the model evaluation process is essential before applying WRF output file for further use.

Several papers I have read mainly has the similar method that introducing some land-based station data(e.g NCDC) within the simulation area and compare them using statistic method.

Usually, we use point-to-point statistic metrics to estimate the similarity of observed data and simulation data in nearest grid in value(MB, RMSE, NMB, NME) and temporal variation(R).

But I have some doubt here

(1) The land-based data is the physics parameters collected from one specific spot and the corresponding WRF output grid data is the average value of the whole area(e.g 9 km^2). They represent the meteorology condition in different scale. What's more, if the station spot is near or at the boundary of grid system, how do we choose the best grid indices?

(2) We always regard the land-based observation data as 'true' value. But the instruments also have its own errors. If the observed data has large bias with 'true' physics process, it should be considered as useless. How do we evaluate the observed data itself and analysis it with certain uncertainty?

(3) More detailly, the WRF output data 2-m temperature(RH) and 10-m wind direction/wind are directly used for point-to-point difference. Does it reasonable to compare in these altitude or the value has little difference in several meters range.

I think this problem not just exist in WRF model. The air quality model evaluation may has the same issue when comparing simulated value with observed concentration. Is there any more advanced method?

Any one has some thought about this question can share your valuable opinion here!

  • 1
    $\begingroup$ you may want to split up your question. Each would require a fairly lengthy answer :-) $\endgroup$
    – gansub
    Sep 22, 2016 at 10:34
  • $\begingroup$ Also, it might not be appropriate to post snapshots of a paper unless you have authorization. It might be better to copy the results if publicly available $\endgroup$
    – arkaia
    Sep 22, 2016 at 12:50
  • $\begingroup$ @arextxabeleta Sorry, that's my negligence. I'll edit it. $\endgroup$ Sep 22, 2016 at 14:04
  • 1
    $\begingroup$ If possible, it might be worth asking a native English-speaker to look over this question. I think it's a decent question, but it's quite difficult to read. $\endgroup$ Sep 27, 2016 at 8:54
  • 1
    $\begingroup$ Not a complete answer - I can't comment on the WRF-specific aspects - but re point 2: Regarding observation data as an accurate "true" value is wrong. The true value lies within the margins of error of the observation data. A nice diagram showing this is Figure 1 of dx.doi.org/10.1016/j.jmarsys.2008.03.011 $\endgroup$ Sep 27, 2016 at 9:00


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy