EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Unsupervised classification of the solar wind using Self-Organizing Maps

Jorge Amaya, Romain Dupuis, and Giovanni Lapenta
Jorge Amaya et al.
  • KU Leuven, CmPA, Mathematics, Leuven, Belgium (

During the past decades different methods of classification of the solar wind have been proposed. These include simple models separating the “fast” from “slow” flows (Arya and Freeman, 1991, Yordanova et al., 2009, among others), complex empirical methods grouping its properties in multiple categories associated to its origin in the solar atmosphere (Xu et Borovsky, 2015), and more recently probabilistic classifications based on Gaussian processes (Camporeale et al., 2017).

Solar wind classification serves four main purposes: 1) statistical analysis of different wind types, 2) interpretation of observations in the magnetosphere, 3) diagnostics of physical processes in the Sun, and 4) identification of solar cycle effects on the Earth’s plasma environment. In this work, instead of using empirical methods, we use the machine learning technique known as Self-Organizing Maps (SOM) to automatically classify the solar wind at 1AU, without human intervention, using observations gathered by the ACE mission.

The ACE spacecraft has been continuously recording solar wind data for the past 22 years. We use hourly averaged solar wind parameters from the ACE Science Center in CalTech for this study. Each entry in this database can be considered as a single point in a multi-dimensional (ND) space. SOM techniques transform all the points in this space into a single 2D space with a small number of L x L nodes. The nodes are the 2D representation of the cloud of points in the ND space, grouping together around each node, points with similar properties. The nodes in this 2D map are interrelated, maintaining a structural topology that is useful for their interpretation. Each one of the nodes in the SOM map can viewed as one of the possible L x L classes. We go one step further, automatically grouping together nodes in the map that are close in the ND space, reducing the total number of classes to only a few. We compare the results obtained using SOM with the methods introduced above, showing the similarities and differences. We show that the SOM technique, which does not rely on human intervention, can be used to properly describe the different types of solar wind conditions observed in a full solar cycle.

How to cite: Amaya, J., Dupuis, R., and Lapenta, G.: Unsupervised classification of the solar wind using Self-Organizing Maps, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-15568,, 2020

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 1 – uploaded on 05 May 2020, no comments