Dear Didin,

In my opinion, I think that for the parameter of wind direction, a simple statistical analysis for the frequency of occurrence data by speed interval and directional interval, would be well enough. This could be done, for instance, by creating a table with percentage frequencies of speed intervals (in columns) and of direction intervals (in rows), for your chosen time interval and then finding a correlation with observational data.

As it concerns the evaluation of the model performance, considering that you are going to use observational data, taken from anemometer, I would suggest to use the cut-off speed of anemometer.

Dear Dimitrios, thank you very much for your suggestion, do you have recommended cut-off speed of anemometer? 1 m/s or 3 m/s? many thanks

Dear Didin, It really depends on the type and the model of anemometer. You have to find the model of the anemometer which has been used for taking the observations, or if you don't know, you can search on the internet for some technical information about anemometers.

If the wind speed is low, you do not expect MM5 and another local wind source to have the same wind direction. Actually at 0 m/s, wind direction is undefined. It is in this respect easier to work in the wind vector components (e.g., north-south and east-west) to evaluate wind quaity. Alternatively, you culd define a speed under which you are not much interested in direction performance, e.g., 4 m/s, and accept no inputs below this threshold. See, e.g., Stoffelen, A., Towards the True Near-Surface Wind Speed: Error Modelling and Calibration Using Triple CollocationJ. Geophys. Res., 1998, 103, 7755-7766. for more details (Ch. IV of http://dspace.library.uu.nl/bitstream/handle/1874/636/full.pdf?sequence=1 ).

http://dspace.library.uu.nl/bitstream/handle/1874/636/full.pdf?sequence=1

Thank you Dimitrios and Ad Stoffelen for your kind constructive and useful input, best regards

You've also asked for criteria. In applications to air pollution modeling I've seen criteria such as within 30 degrees of observed wind direction. After more careful review these criteria turned out to be more lenient than I thought since they were being applied to daily average WD. Hourly WD of modeled fields that passed the test could be off as much as 120 degrees. Depending on scale and application sometimes even 10 degree shift can be significant.

@Mehmet: I'd think that the (x,y) position error of pollution is also handled more conveniently using the cartesian wind vector components.

Dear Mehmet Talat Odman, yes we used MAGE within 30 degrees (Emery et al.2001) for hourly dataset but seems hard to meet the criteria. By applying cut-off wind speed for observation we could get better MAGE but not that significant. We have not tried for daily average. Many thanks for your input.

Dear Didin，The WRF is more suitasble to simulate wind diractiona nd espeed. Especall for wind diraction, however, if you need to have accuratly simulate the wind variability, the best way is assimilate observation data.

Dear Chenghai Wang, do you have comparative study of WRF and MM5 in simulating win direction and its performance evaluation. I do agree with you as we are shifting to WRF to provide met fields to drive our air quality models. However for data assimilation, i think we do not have enough good ground observation here in term of spatial distribution, thank you for your kind input and suggestion, best