Heliophysics

These examples apply the methods of the earlier chapters to problems in space physics. In each, the difficulty lies less in the model than in the data: gaps and changing cadence, strong dependence between neighbouring samples, rare events, the need for a physical baseline, and archives whose provenance limits what can be concluded. Each notebook ends by stating what its results support.

• Dst Forecasting: a one-hour-ahead forecast of the Dst index from hourly OMNI data, with year-based splits and persistence as the baseline to beat.

• Plasma-Sheet Modeling: saved predictions of a neural-network plasma-sheet model compared with the empirical TM03 model, for ion temperature and for density maps.

• SEP Occurrence Forecasting: a rare-event classification problem, the space-weather verification scores, and how the decision threshold changes the forecast.

• Coronal-Loop Reconstruction: recovering loop heights from projected shapes, and how the choice of split changes the answer from excellent to useless.