Interpretable Machine Learning Methods for forecasting the SYM-H Index
Abstract
Forecasting geomagnetic indices is crucial for mitigating potential effects of severe geomagnetic storms on critical infrastructures such as power grids. In particular, the SYM-H index describes the symmetric geomagnetic disturbance of the horizontal component of the magnetic field at midlatitude on the Earth's surface with a 1-minute resolution. Machine learning methods, particularly neural networks, have been shown to outperform physics-based models for forecasting geomagnetic activity. In this work, we develop gradient boosting machines (GBMs) for forecasting the SYM-H index multiple hours ahead using different combinations of solar wind and interplanetary magnetic field (IMF) parameters , derived parameters, and past SYM-H values. An advantage of GBMs is that feature contribution scores for a prediction at any given time can be extracted efficiently, thereby making predictions transparent and interpretable. Using feature importance scores, we show that predictions of the SYM-H index from GBMs are consistent with physical understanding while also providing insight into the complex relationship between the solar wind and Earth's ring current. We also perform a direct comparison between GBMs and existing neural networks for forecasting the SYM-H index by training, validating, and testing them on the same data. Through this comparison, we show that GBMs have a lower overall root mean squared error and are more adept at predicting the magnitude of strong geomagnetic storms.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNG44A..02I