Inspiration
There is currently no model that predicts tornadoes more than about 15 minutes in advance, so an even earlier warning system would be really helpful in keeping people safe in a disastrous event.
What it does
It scrapes current weather data from Mesonet weather stations, and that weather data is fed into a neural network that gives a probability of a tornado forming.
Data includes: current temperature, humidity, atmospheric pressure, wind speed, wind direction, precipitation, and more.
How we built it
We utilized python for data scraping and processing, and Keras with Tensorflow backend for the neural network training and testing.
Challenges we ran into
We couldn't find a significant correlation between current weather data like temperature, humidity, pressure, etc, and the formation of tornadoes. The model does perform better than random.
Accomplishments that we're proud of
The model does perform better than random, and we learned more about how neural networks are trained, and what type of data we should be looking for. The more data, the better!
What we learned
We learned new neural network models, and more features in the keras/tensorflow Machine Learning library.
What's next for Tornado Predictor
Creating a better model that will work well enough to be used in a realistic scenario.
Built With
- keras
- mesonet
- noaa
- python
- tensorflow
Log in or sign up for Devpost to join the conversation.