The following repo covers a preview of the work I did under Claire Monteleoni and Moumita Saha. The goal of this project was to classify the rainfall type for a given day in advance. We then moved to the final goal, which was to classify spell detection (predicting a wet, dry, or normal day). In total, we utilized NOAA's reanalysis data and created a 4-dimensional time series grid stretching from 1948 to 2014.
Although we used a general pipeline from traditional methods, we ended up creating variants of Convolutional Neural Networks (CNN). Above, you can see the code for the models as well as programs to extract the .netcd4 data. We have also included results for anyone to see how the different variations between the architectures as well as hyperparamaters performed.