i live in ga where we have hill billy overload so i have noticed something (just an observation):
im currently analyzing weather data in my area to help me understand the most effective way to schedule the iro (and provide a base line i use to measure efficacy of the iro) and what i have found is that pws(es) that do not report pan evaporation levels seem to be the least accurate/most volatile. i assume because they carry a higher price tag and every yahoo in the state is not throwing these up to report to various weather services.
i have found this site to be useful in validating other pws(es). http://www.ncdc.noaa.gov/cdo-web/datatools/findstation
my program discards and pws(es) that are more than 1 sigma out of standard deviation (where absent data gets the highest possible values) and the pattern that is beginning to emerge is the absence of pan evap data.
i hope this helps some, at first, the site i provided is un-intuitive but after you go 8 seconds on a link click rodeo, you will find all of the available court certified weather stations and then you can cross reference the pws(es) you desire with these highly maintained/calibrated weather stations.
sorry i cant tell you exactly how to determine the station you should use, but if it was explainable, these guys would have already coded an algorithm for it. hell my method is flawed as well, but i have given myself a false sense of security in my methods/data so i speak it as truth.