Localised Magnetic Substorm Forecasting using Machine Learning

We use a prevailed technique to extract image features and classify 4 seasons of aurora all sky images, combine these with solar wind and interplanetary magnetic field (IMF) data and use this as a basis to forecast the onset of geomagnetic substorms local to the imager. To prove the viability of our model, we successfully reproduce the results of a previous study which used only solar wind and IMF data to forecast global substorm onsets. Although this viability test proves successful and we independently confirm the previous model, our expanded model fails to deliver the necessary performance required for it to be used for accurate localised substorm forecasting.

In the accompanying publication we describe our methods and show the results that we obtained.

This dataset contains the processed files that can be used to replicate our results. Instructions how to apply the data and on the accompanying code can be found here:
http://tid.uio.no/plasma/LOCATE/