I'm customizing WRF for operational prediction in East Africa. A recent suggestion is to compare with other models possibly UK's unified model, COSMO and AROME with WRF. Is this suggestion appropriate for a Ph.D study?
If you are "customizing" WRF for your region, the first thing to do would be a robust validation of WRF with existing observations over the region. Try to play around with various parameterization schemes and see which combination of schemes suits best to give closest results to observations. Of course, none of the options will be perfect. Once you are done with the best-suited schemes, go ahead and modify them in the code. There's a good book called parameterization schemes by David Stensrud which you might like to look into.
I agree that for a typical PhD a single optimal configuration of an operational prediction system should be enough. Probably you can identify those aspects of operational prediction that might be most relevant on your area/region and focus on systematically identify the best settings and parameterizations within the model.
My experience is more related with COSMO model than WRF. As WRF, it is also a free code for research or operational weather agencies on developing countries. Please check the first link I am including where it is explained whose services are free. In particular, "DWD also provides lateral boundary data (up to 120h) to these services in real time via the internet free of charge up to four times daily, based on the analyses at 00, 06, 12 and 18 UTC." which can be useful for your projects.
In the second link I am providing you may check several operational weather services based on COSMO. There are mainly located on Europe but anyway in case you consider to use COSMO, it is interesting to compare the computational resources used on every case. There is also information about verification of the COSMO model including possible tools.
Also the COSMO model is also used as a regional climate model, so in the third link you may find studies over Africa (also East Africa). They might be useful to identify specific problems of modelling to be ascertain on your region.
Finally, please check the training stuff, I am also providing a link, and I have attached .pdf document (please check the Page 3, might be useful in your case).