The way Forecast.io works is you get a fairly significant amount of data for the upcoming day and even week, however as time progresses, the data changes because weather information changes. Especially where I live, you can have the forecast (even on TV) say sunny all day and then it changes to rain in the afternoon, so how ‘stale’ can you let your displays go?
So for “accurate” forecasts, I would suggest a 1-5 min update per player. How many calls that is to the app depends on your setup, if all the players are in the same region (eg. a campus network), you can set them all to the same GPS coordinates and minimize the calls by caching although I don’t think it’s entirely fair nor accurate to do so, Forecast prides itself on hyperlocal data, which means even across 5 km your data could be different.
1000 calls per day is still 50 / day for 20 clients. That’s one every half hour within the ‘free’ range although it’s cutting it close. Even so, the cost is $0.0001/call for every one above 1000 even if you have to pay for a 1000 extra calls, you’re talking 10c/day or $3/month for your entire network.