How is the forecast derived?


Where can I find information on how the forecast data shown in the Tempest app is derived and if/how my station and those around me are used as input to the forecast?

Thank you.

1 Like

That’s a good question. It is a proprietary forecast direct from WeatherFlow, so the exact details are probably part of their intellectual property (I would assume). There is some detail here:, but perhaps @dsj could add something further?


Hi @tim.feeney This is indeed a good question!

The forecast data displayed in the Tempest app is produced by WeatherFlow. Our modeling systems start with data from a variety of third-party sources, including global and regional numerical models from US and European weather agencies, as well as WeatherFlow’s in house regional numerical models. The forecasts are then enhanced using the rich source of WeatherFlow observations in WeatherFlow’s proprietary machine learning (AI) forecasting platform to improve the forecast for your specific location.

Improvements are made to the local forecast as the machine learning (AI) ingests observations from your Tempest System and other observations in your area. We pay a great deal of attention to the quality of the observations coming from your Tempest and our Continuous Learning AI seeks to improve the quality of your unique observations. You will notice advantages in accuracy over other forecasts starting about ~60 days from installation, and then you will see additional improvements going forward as we learn more about your local weather.


Has the new feature of using stats from your station to improve your local forecast been activated

Yes, see above and these Forecast FAQs

Thanks. Nice to know it has been activated. Remember it being stated that it was being held back. Glad its working. So if i have my SKY/AIR running and my new Tempest running, the stats from my 2 stations plus stations close to me will help give me a more precise local forecast. How far from my station would they go. 1,2,4 or 5 miles away or farther

@WFstaff, Will AI help with forecast details in places like Panama City Beach? Seems like the NWS will miss the finer details in their forecast like the timing of storms for example. Let’s say the predict the storms arriving after 2p but instead arrive around 8a. Or will AI will just help with the numerical stuff in forecasting like temp, wind, and that’s basically it?

1 Like

Great question @jgentry

The current method for forecast improvement focuses on using AI to tune our model output to the backyard/neighborhood level for base forecast parameters. We are also working on using observations in the model to affect the model forecast itself.


I’m curious how advanced your QC process is? Seems like, as with any citizen weather network, data issues will be prevalent with bad station citing. This is especially apparent with wind speeds, yielding a low bias to the model wind field.

but that’s where the AI would comes in handy, depending on the way you look at it.
The AI might notice that the prediction of the temperature is always too low for whatever your situation is. It might increase the predicted temperature (depending on all the other factors like wind, humidity, time of day etc, after all it is an AI). So next day, your prediction matches your measurements.
However, your temp might be high because you placed the tempest on a black flat roof that gets super hot in the sun. That would raise the temperature. Would that ever match the temperature predicted by your national weather service? Nope. Would that be a valid temperature? Yes, that IS the temperature measure at your location, can the AI predict that higher temperature. Sure it can. Is that better? Now that is a big question as you might argue that you would prefer your measured temperature to match a temperature as if your station was situated in an ideal location. I’m not sure I would like that, perhaps yes, perhaps no. But AI could be used to tune that. So instead of changing the forecast, change the measurement??

If it adjusted the measurement, the forecast comes closer to the measured value.

One other factor to think of, is that even when you have an ideal setup, accurate as it gets, you still might always be off in the forecast, because of a nearby river or mountain or something like that. In those cases it would be a super great idea if the AI could change the forecast to a better one.