Weather Underground is very easy to poll since they have a well documented API. What you need is an algorithm you can trust.
1″ of rain – delay for 12 hours
2″ of rain – delay for 24 hours
Temp > 90 – water level 150%
This is less about programming and more about researched data. For a system like this to work, you have to be able to trust it will make the same decisions you would, or an expert would. It has to cover all possible scenarios otherwise it’s useless (if it misinterprets data and activates sprinklers while it’s raining, for example).
I think Weather Underground’s data can be trusted however I have not seen a solid algorithm, yet.
If you can show me or convince me something is worth while, I have no problem switching away from Yahoo and building it right into my web app. But, that is a lot of effort and unless I have tangible data showing the algorithm is legitimate I don’t think it would be worth it.
Update: A flow chart for example would be amazing. I could also be over thinking however trusting a program to modify your schedules seems scary (yard dies if it miscalculates while you’re on vacation). In my opinion, this is something you use until it messes up and then you never use it again. So, if we roll this out it needs to be perfect from the start.