As you may know, 'interesting' weather hit czech republic. Heavy rains
followed by floods claiming lives. What is more interesting, the
weather forecasting went crazy, too. yr.no
normally works pretty well,
but these days, it oscillates crazily as model is recomputed with new
data. (fridays forecast basically said 'saturday mostly nice with
light rain in the morning, sunday rainy; saturdays forecast says
'heavy rain in the evening, only light rain on sunday).
Now, forecasts got better. We used to use simple 'sunday rainy at 20C'
predictions, then medard-online
came where you actually see data from
the model. Unfortunately browsing them is quite time consuming. yr.no
helps there: you select place and it shows you 3 day of prediction on
But it still lacks a lot: it only tells you expected values for the
predictions, and not the expected deviations (aka the ammount of
certainity in the prediction). "Easy" way to solve that would be to run the simulation few times, slightly varying input variables each time; then dispalying both mean values and deviations calculated.
....but I'm told that's not feasible, because weather forecast is already computationaly intensive, as is. OTOH, weather forecasting is already repeated, once every few hours, when new data become available. The solution may be as easy as displaying the "old" predictions, too: if they are similar to the "new" prediction, the prediction is probably reliable. If not, well...
This should be all easy to modify/check if weather modeling software was open source and did not require super computer... is there such beast? (I believe model running medard is opensource, but fortran and the supercomputer is probably needed... plus where to get the source data?)
On the similar wein... extremely short term forecasts (< 3 hours) should also be extremely reliable. I'd really like android to use its gps, then warn me if the rain is coming... Maybe it is as simple as predicting cloud motion from weather radar
? Is there maybe similar software/service already?