This will be an early warning system for "You need to get home and unplug your computer before it's hit by lightning," or "It's going to rain soon." (Yes, I have $100 worth of protective equipment between the wall and my computer, but there are factors that are more-or-less outside my control.) There may well be all sorts of apps and such for this, but I've become obesessed / fascinated exploring the problem for myself.
Input began with the (US) National Weather Service radar (station ID FFC) based near (or in) Peachtree City, GA. One problem with it is that it works at roughly 98% uptime, not the literal or proverbial 99.x%. It has died at least twice in a week or two in mid-August, 2019. Once it was down for around an hour at least.
The other "problem" is that it certainly hints at lightning, but you can't be sure from its data.
Enter Lightning Maps or what turns out to be much more computer-readable data which, as I remember, comes from Blitzortung. Those sites show you lightning, and I've never seen them down. One of the related links gives data as a PHP array that works with the "eval()" function. It's the first time I ever used that function. (After verifying it works, I used another method because eval is just too dangerous.) That array gives you time, longitude and latitude. My observation is that the location is almost always accurate to within 5 - 10 miles, although it has some non-important quirks like showing several points for the same strike. Some of those points are almost certainly incorrect because I can't see or hear the strike when the system says I should.
In any event, it's good data. One "problem" is that I want to see the clouds forming and rain coming. The other problem is that it appears those (closely related) systems are run by hobbyists. I don't know what they are paying, although I could look into it, but I don't want to be reading their data too often. Eventually, I would like to work with them to lessen the load on their system, but that's another project.
The next possible input is GOES satellite data. I've seen some downtime, but not with the data I wanted. One path of research, which for now is a "problem," is I haven't figured out which combination of bands (wavelenths) is best to see thunderstorms. Candidates are bands 7, 11, 14, and 15. For different storms, it seems different bands are best.
GOES would give me the same or perhaps better data than the doppler radar (NWS), probably with better uptime.
Yet another possible input is CWOP--example on my site, FindU example, definition of CWOP. Possible advantages to this are very local readings, and there are probably indicators one can determine for thunderstorms. My observation is that humidity goes quickly up before a storm, and possibly that the barometer goes quickly up. (I thought it would go down.) This could be a subject for Bayesian analysis / machine learning.
One possible problem with the FindU data is that it may also be funded by hobbyists, although I suspect there is a government subsidy by now. I also read recently that the CWOP data goes to a more-or-less government site after a five minute delay.
I've color-processed the doppler radar so that I can get dbz units. I've also identified my approximate pixel on the map. So I have my code telling me how far away every dbz is--light rain, medium rain, heavy rain, etc., and also what direction it's in. I'm doing the same with the lightning data.