The numerous recent breakthroughs in machine learning (ML) make imperative to carefully ponder how the scientific community can benefit from a technology that, although not necessarily new, is today living its golden age. This Grand Challenge review paper is focused on the present and future role of machine learning in space weather. The purpose is twofold. On one hand, we will discuss previous works that use ML for space weather forecasting, focusing in particular on the few areas that have seen most activity: the forecasting of geomagnetic indices, of relativistic electrons at geosynchronous orbits, of solar flares occurrence, of coronal mass ejection propagation time, and of solar wind speed. On the other hand, this paper serves as a gentle introduction to the field of machine learning tailored to the space weather community and as a pointer to a number of open challenges that we believe the community should undertake in the next decade. The recurring themes throughout the review are the need to shift our forecasting paradigm to a probabilistic approach focused on the reliable assessment of uncertainties, and the combination of physics-based and machine learning approaches, known as gray-box.

/pdf/the-challenge-of-machine-learning-in-space-weather-12iowq5jx4.pdf

The Challenge of Machine Learning in Space Weather: Nowcasting and Forecasting

Solar cycle predictions are needed to plan long-term space missions, just as weather predictions are needed to plan the launch. Fleets of satellites circle the Earth collecting many types of science data, protecting astronauts, and relaying information. All of these satellites are sensitive at some level to solar cycle effects. Predictions of drag on low-Earth orbit spacecraft are one of the most important. Launching a satellite with less propellant can mean a higher orbit, but unanticipated solar activity and increased drag can make that a Pyrrhic victory as the reduced propellant load is consumed more rapidly. Energetic events at the Sun can produce crippling radiation storms that endanger all assets in space. Solar cycle predictions also anticipate the shortwave emissions that cause degradation of solar panels. Testing solar dynamo theories by quantitative predictions of what will happen in 5 – 20 years is the next arena for solar cycle predictions. A summary and analysis of 75 predictions of the amplitude of the upcoming Solar Cycle 24 is presented. The current state of solar cycle predictions and some anticipations of how those predictions could be made more accurate in the future are discussed.

https://www.leif.org/EOS/Pesnell_2012.pdf

Solar Cycle Predictions (Invited Review)

Solar flares originate from the release of the energy stored in the magnetic field of solar active regions, the triggering mechanism for these flares, however, remains unknown. For this reason, the conventional solar flare forecast is essentially based on the statistic relationship between solar flares and measures extracted from observational data. In the current work, the deep learning method is applied to set up the solar flare forecasting model, in which forecasting patterns can be learned from line-of-sight magnetograms of solar active regions. In order to obtain a large amount of observational data to train the forecasting model and test its performance, a data set is created from line-of-sight magnetogarms of active regions observed by SOHO/MDI and SDO/HMI from 1996 April to 2015 October and corresponding soft X-ray solar flares observed by GOES. The testing results of the forecasting model indicate that (1) the forecasting patterns can be automatically reached with the MDI data and they can also be applied to the HMI data; furthermore, these forecasting patterns are robust to the noise in the observational data; (2) the performance of the deep learning forecasting model is not sensitive to the given forecasting periods (6, 12, 24, or 48 hr); (3) the performance of the proposed forecasting model is comparable to that of the state-of-the-art flare forecasting models, even if the duration of the total magnetograms continuously spans 19.5 years. Case analyses demonstrate that the deep learning based solar flare forecasting model pays attention to areas with the magnetic polarity-inversion line or the strong magnetic field in magnetograms of active regions.

https://iopscience.iop.org/article/10.3847/1538-4357/aaae00/pdf

Deep Learning Based Solar Flare Forecasting Model. I. Results for Line-of-sight Magnetograms

Solar flares are powered by the energy stored in magnetic fields, so evolutionary information of the magnetic field is important for short-term prediction of solar flares. However, the existing solar flare prediction models only use the current information of the active region. A sequential supervised learning method is introduced to add the evolutionary information of the active region into a prediction model. The maximum horizontal gradient, the length of the neutral line, and the number of singular points extracted from SOHO/MDI longitudinal magnetograms are used in the model to describe the nonpotentiality and complexity of the photospheric magnetic field. The evolutionary characteristics of the predictors are analyzed by using autocorrelation functions and mutual information functions. The analysis results indicate that a flare is influenced by the 3-day photospheric magnetic field information before flare eruption. A sliding-window method is used to add evolutionary information of the predictors into machine learning algorithms, then C4.5 decision tree and learning vector quantization are employed to predict the flare level within 48 hours. Experimental results indicate that the performance of the short-term solar flare prediction model within the sequential supervised learning framework is significantly improved.

Short-Term Solar Flare Prediction Using a Sequential Supervised Learning Method

Review and prospect of guidance and control for Mars atmospheric entry

Solar flare forecasting model supported with artificial neural network techniques

As Earth-based radio tracking navigation is severely limited because of communications constraints and low relative navigation accuracy, autonomous optical navigation capabilities are essential for both robotic and manned deep-space exploration missions. Image processing is considered one of the key technologies for autonomous optical navigation to extract high-precision navigation observables from a raw image. New image processing algorithms for deep-space autonomous optical navigation are developed in this paper. First, multiple image pre-processing and the Canny edge detection algorithm are adopted to identify the edges of target celestial bodies and simultaneously remove the potential false edges. Secondly, two new limb profile fitting algorithms are proposed based on the Least Squares method and the Levenberg-Marquardt algorithm, respectively, with the assumption that the perspective projection of a target celestial body on the image plane will form an ellipse. Next, the line-of-sight (LOS) vector from the spacecraft to the centroid of the observed object is obtained. This is taken as the navigation measurement observable and input to the navigation filter algorithm. Finally, the image processing algorithms developed in this paper are validated using both synthetic simulated images and real flight images from the MESSENGER mission.

/pdf/image-processing-algorithms-for-deep-space-autonomous-4p7lbu4t4h.pdf

Liu Zhang

Papers

Solar flare forecasting model supported with artificial neural network techniques

Image Processing Algorithms For Deep-Space Autonomous Optical Navigation

MCAV/IMU integrated navigation for the powered descent phase of Mars EDL

Longitudinal distribution of major solar flares during 1975–2005

A summary of the applications of a weighted average method determining times of solar cycle extrema