Machine Learning (AI) & storm/weather related data

[ChristofferB]

I have recently started studying Machine Learning, which is a sort of Artificial Intelligence used for example in image recognition, self driving cars, speech recognition (Siri) etc. It is quite interesting and I think it could be really useful in terms of storm/tornado prediction in the future.

As a side project I am eager to start testing this with real data and it would be fun to do it on real storm data. So, I am looking for projects that have data that needs to be analyzed. I will do it for free.

The data needs to be:
- Inputs in terms of parameters. This could be for example moisture, temperature, shear etc.
- Measured output. This could be in form of Tornado/No Tornado or "Hail size" or "Wind speed".

What this will give is a prediction of Output given the Inputs. I know weather is far more unpredictable than this but as an example it could be that given the measured windspeed in the meso, the wind speed of the RFD, the moisture the system could predict the possibility of a tornado.

I am far, far from an expert in the field (just a student at the moment) but I will do my best. Let me know if I can help you!

 

[Alex Elmore]

It is quite interesting and I think it could be really useful in terms of storm/tornado prediction in the future.

You are correct, as there are already two projects that are AI-based I'm aware of that are currently being developed for the purpose of aiding in the forecasting severe weather. I know NSSL is working on a product called "severeprob" (I think that's how it's spelled). Doing a quick search, I can't any information on it to post here, which is probably because it's still experimental. I do know it uses Bayes' theorem to form probabilities of severe weather based on satellite, radar, and meso analysis data. The other project is another experimental product from the CIPS group at Saint Louis University. This one uses analogs and severe reports to build a logistic regression model to form probabilities of severe weather in the short- and long-term forecast range. Once I get permission from my adviser, I plan on posting the link to access the product in a separate thread.

There is a conference for this particular topic put on by the AMS:

https://annual.ametsoc.org/2018/ind...s-applications-to-the-environmental-sciences/

The conference this year has already happened, but you should be able to look at presenters' abstracts, which will hopefully give you a better idea of what's going on in the field, as well as people and places to look into for possibly working with.

 

[ChristofferB]

Alex: Thanks, that was really quite interesting. I believe AI in general and machine learning in particular will become more and more beneficial for weather predictions in the future so it will be interesting to follow. For my own means though, I believe this may be quite a bit above my knowledge level. I found some interesting tornado data on Kaggle.com that I will try my skills on to begin with.

If anyone, who is doing some research, and is in the need of finding patterns and predictions from its data - later on - find this thread. Send me a PM and I may be able to help you out.