Predicting the energetic proton flux with a machine learning regression algorithm

TL;DR

A simple and efficient machine learning regression algorithm is presented that is able to forecast the energetic proton flux up to 1 hr ahead by exploiting features derived from the electron flux only, which could be helpful in improving monitoring systems of the radiation risk in both deep space and near-Earth environments.

Abstract

The need of real-time of monitoring and alerting systems for Space Weather hazards has grown significantly in the last two decades. One of the most important challenge for space mission operations and planning is the prediction of solar proton events (SPEs). In this context, artificial intelligence and machine learning techniques have opened a new frontier, providing a new paradigm for statistical forecasting algorithms. The great majority of these models aim to predict the occurrence of a SPE, i.e., they are based on the classification approach. In this work we present a simple and efficient machine learning regression algorithm which is able to forecast the energetic proton flux up to 1 hour ahead by exploiting features derived from the electron flux only. This approach could be helpful to improve monitoring systems of the radiation risk in both deep space and near-Earth environments. The model is very relevant for mission operations and planning, especially when flare characteristics and source location are not available in real time, as at Mars distance.

Figures (4)

• Figure 1: EPHIN proton/electron differential channels P8 (blue), E150 (orange) and E300 (green) during the event of October 25, 2000. The black vertical lines mark the onset in proton and electron data. In this event, the electrons measured in E150 and E300 differential channels arrive 01:30 hours earlier than protons in the P8 channel. Furthermore, the time profiles are very similar.
• Figure 2: RF algorithm diagram adapted from janosh_tikz_2023. The paths highlighted in red represent the series of decisions made by each tree in the forest to make the final prediction. The RF prediction is the average of the trees final decisions. Each node represents a logical operation on a single feature. For example, a node may be asking if the variable E300 $<$ 100 (cm$^{2}$ s sr MeV)$^{-1}$. If yes go to the next node on the right, otherwise go to the next node on the left. The graph and the logical operations are inferred from data during the training phase of the model.
• Figure 3: A 3-months sample dataset used for testing/visualization purposes. Green line represents the target of the model (true values), while blue line represents the output of the model (forecasting). Both are representative of the logarithm of the proton flux integrated from 7.8 MeV and 53 MeV energy channels (P8, P25, P41; see Table \ref{['Tab:tab1']}). The proton flux is given in pfu, while the logarithm is adimensional. The inset shows a closer look around the ESP event of 2012-03-08, which is not well reproduced by the model. This is due to the fact that the electrons do not exhibit the ESP signature (see the light blue line, which shows the track of the E300 electron differential channel).
• Figure 4: (Left panels) Comparison between the predicted time series and the target one for three samples: 23, 18 and 31 (from top to bottom). (Right panel) correlation plot between the predicted and the target points. The grey line represents the bisector (not the linear regression line). Note that the sample 31 is the worst possible example (see Table \ref{['Tab:tab_summary']}). However, even though the model overestimates the flux, the trend is reproduced quite well during the enhancement of 2016-07-23.