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

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 Predictor Teams

A short-term solar flare prediction model is built using predictor teams rather than an individual set of predictors. The information provided by the set of predictors could be redundant. So it is necessary to generate subsets of predictors which can keep the information constant. These subsets are called predictor teams. In the framework of rough set theory, predictor teams are constructed from sequences of the maximum horizontal gradient, the length of neutral line and the number of singular points extracted from SOHO/MDI longitudinal magnetograms. Because of the instability of the decision tree algorithm, prediction models generated by the C4.5 decision tree for different predictor teams are diverse. The flaring sample, which is incorrectly predicted by one model, can be correctly forecasted by another one. So these base prediction models are used to construct an ensemble prediction model of solar flares by the majority voting rule. The experimental results show that the predictor team can keep the distinguishability of the original set, and the ensemble prediction model can obtain better performance than the model based on the individual set of predictors.     

Solar Flare Prediction Using Advanced Feature Extraction, Machine Learning, and Feature Selection

Novel machine-learning and feature-selection algorithms have been developed to study: i) the flare-prediction-capability of magnetic feature (MF) properties generated by the recently developed Solar Monitor Active Region Tracker (SMART); ii) SMART’s MF properties that are most significantly related to flare occurrence. Spatiotemporal association algorithms are developed to associate MFs with flares from April 1996 to December 2010 in order to differentiate flaring and non-flaring MFs and enable the application of machine-learning and feature-selection algorithms. A machine-learning algorithm is applied to the associated datasets to determine the flare-prediction-capability of all 21 SMART MF properties. The prediction performance is assessed using standard forecast-verification measures and compared with the prediction measures of one of the standard technologies for flare-prediction that is also based on machine-learning: Automated Solar Activity Prediction (ASAP). The comparison shows that the combination of SMART MFs with machine-learning has the potential to achieve more accurate flare-prediction than ASAP. Feature-selection algorithms are then applied to determine the MF properties that are most related to flare occurrence. It is found that a reduced set of six MF properties can achieve a similar degree of prediction accuracy as the full set of 21 SMART MF properties.     

Automatic Solar Flare Tracking Using Image-Processing Techniques

Measurement of the evolution properties of solar flares through their complete cyclic development is crucial in the studies of Solar Physics. From the analysis of solar H images, we used Support Vector Machines (SVMs) to automatically detect flares and applied image segmentation techniques to compute their properties. We also present a solution for automatically tracking the apparent separation motion of two-ribbon flares and measuring their moving direction and speed in the magnetic fields. From these measurements, with certain assumptions, we inferred the reconnection of the electric field as a measure of the rate of the magnetic reconnection in the corona. The automatic procedure is a valuable tool for real-time monitoring of flare evolution.     

Automatic Solar Flare Detection Using Neural Network Techniques

We present a new method for automatic detection of flare events from images in the optical range. The method uses neural networks for pattern recognition and is conceived to be applied to full-disk Himages. Images are analyzed in real time, which allows for the design of automatic patrol processes able to detect and record flare events with the best time resolution available without human assistance. We use a neural network consisting of two layers, a hidden layer of nonlinear neurodes and an output layer of one linear neurode. The network was trained using a back-propagation algorithm and a set of full-disk solar images obtained by HASTA (HSolar Telescope for Argentina), which is located at the Estación de Altura Ulrico Cesco of OAFA (Observatorio Astronómico Félix Aguilar), El Leoncito, San Juan, Argentina. This method is appropriate for the detection of solar flares in the complete optical classification, being portable to any Hinstrument and providing unique criteria for flare detection independent of the observer.     

Automatic Solar Flare Detection Using MLP,RBF, and SVM

The focus of automatic solar-flare detection is on the development of efficient feature-based classifiers. The three principal techniques used in this work are multi-layer perceptron (MLP), radial basis function (RBF), and support vector machine (SVM) classifiers. We have experimented and compared these three methods for solar-flare detection on solar H images obtained from the Big Bear Solar Observatory in California. The preprocessing step is to obtain nine principal features of the solar flares for the classifiers. Experimental results show that by using SVM we can obtain the best classification rate of the solar flares. We believe our work will lead to real-time solar-flare detection using advanced pattern recognition techniques.     

The Evolution of Sunspot Magnetic Fields Associated with a Solar Flare

Solar flares occur due to the sudden release of energy stored in active-region magnetic fields. To date, the precursors to flaring are still not fully understood, although there is evidence that flaring is related to changes in the topology or complexity of an active-region’s magnetic field. Here, the evolution of the magnetic field in active region NOAA 10953 was examined using Hinode/SOT-SP data over a period of 12 hours leading up to and after a GOES B1.0 flare. A number of magnetic-field properties and low-order aspects of magnetic-field topology were extracted from two flux regions that exhibited increased Ca ii H emission during the flare. Pre-flare increases in vertical field strength, vertical current density, and inclination angle of ≈ 8° toward the vertical were observed in flux elements surrounding the primary sunspot. The vertical field strength and current density subsequently decreased in the post-flare state, with the inclination becoming more horizontal by ≈ 7°. This behavior of the field vector may provide a physical basis for future flare-forecasting efforts.     

Phase Relationship Between Sunspot Number, Flare Index and Solar Radio Flux

To understand better the variation of solar activity indicators originated at different layers of the solar atmosphere with respect to sunspot cycles, we carried out a study of phase relationship between sunspot number, flare index and solar radio flux at 2800 MHz from January 1966 to May 2008 by using cross-correlation analysis. The main results are as follows: (1) The flare index and sunspot number have synchronous phase for cycles 21 and 22 in the northern hemisphere and for cycle 20 in the southern hemisphere. (2) The flare index has a noticeable time lead with respect to sunspot number for cycles 20 and 23 in the northern hemisphere and for cycles 22 and 23 in the southern hemisphere. (3) For the entire Sun, the flare index has a noticeable time lead for cycles 20 and 23, a time lag for cycle 21, and no time lag or time lead for cycle 22 with respect to sunspot number. (4) The solar radio flux has a time lag for cycles 22 and 23 and no time lag or time lead for cycles 20 and 21 with respect to sunspot number. (5) For the four cycles, the sunspot number and flare index in the northern hemisphere are all leading to the ones in the southern hemisphere. These results may be instructive to the physical processes of flare energy storage and dissipation.     

Prediction of Solar Cycle 24 Using Sunspot Number near the Cycle Minimum

Correlations between monthly smoothed sunspot numbers at the solar-cycle maximum [R _max] and duration of the ascending phase of the cycle [T _rise], on the one hand, and sunspot-number parameters (values, differences and sums) near the cycle minimum, on the other hand, are studied. It is found that sunspot numbers two?–?three years around minimum correlate with R _max or T _rise better than those exactly at the minimum. The strongest correlation (Pearson’s r=0.93 with P<0.001 and Spearman’s rank correlation coefficient r _S=0.95 with P=9×10^?12) proved to be between R _max and the sum of the increase of activity over 30 months after the cycle minimum and the drop of activity over 30 or 36 months before the minimum. Several predictions of maximal amplitude and duration of the ascending phase for Solar Cycle 24 are given using sunspot-number parameters as precursors. All of the predictions indicate that Solar Cycle 24 is expected to reach a maximal smoothed monthly sunspot number (SSN) of 70?–?100. The prediction based on the best correlation yields the maximal amplitude of 90±12. The maximum of Solar Cycle 24 is expected to be in December 2013?–?January 2014. The rising and declining phases of Solar Cycle 24 are estimated to be about 5.0 and 6.3 years, respectively. The minimum epoch between Solar Cycles 24 and 25 is predicted to be at 2020.3 with minimal SSN of 5.1?–?5.4. We predict also that Solar Cycle 25 will be slightly stronger than Solar Cycle 24; its maximal SSN will be of 105?–?110.     

Solar Flare Occurrence Rate and Probability in Terms of the Sunspot Classification Supplemented with Sunspot Area and Its Changes

We investigate the solar flare occurrence rate and daily flare probability in terms of the sunspot classification supplemented with sunspot area and its changes. For this we use the NOAA active region data and GOES solar flare data for 15 years (from January 1996 to December 2010). We consider the most flare-productive 11 sunspot classes in the McIntosh sunspot group classification. Sunspot area and its changes can be a proxy of magnetic flux and its emergence/cancellation, respectively. We classify each sunspot group into two sub-groups by its area: ??Large?? and ??Small??. In addition, for each group, we classify it into three sub-groups according to sunspot area changes: ??Decrease??, ??Steady??, and ??Increase??. As a result, in the case of compact groups, their flare occurrence rates and daily flare probabilities noticeably increase with sunspot group area. We also find that the flare occurrence rates and daily flare probabilities for the ??Increase?? sub-groups are noticeably higher than those for the other sub-groups. In case of the (M+X)-class flares in the ??Dkc?? group, the flare occurrence rate of the ??Increase?? sub-group is three times higher than that of the ??Steady?? sub-group. The mean flare occurrence rates and flare probabilities for all sunspot groups increase with the following order: ??Decrease??, ??Steady??, and ??Increase??. Our results statistically demonstrate that magnetic flux and its emergence enhance the occurrence of major solar flares.     

Cosmic Ray Intensity Variations in Relation with Solar Flare Index and Sunspot Numbers

From the monthly data of cosmic ray intensity (CRI), sunspot numbers (SSN) and solar flare index (SFI), an attempt has been made to study the relationship between CRI and solar activity (SA) parameters SSN and SFI. The correlation between SA parameters and CRI for different neutron monitoring stations having low, middle and high cut-off rigidity has been investigated. The anti-correlation between SA and CRI is found to exist with some time lag. Based on the method of minimizing correlation coefficient and time-delayed component method, the observed time-lag between SA parameters (SSN and SFI) and CRI has been found to be large for odd solar cycles in comparison to even solar cycles. The results of time-lag analysis between CRI and SSN and between CRI-SFI have also been compared. The findings of correlative study between CRI and SSN are in agreement with earlier results, while the CRI-SFI relationship provides new insights to understand the solar modulation of cosmic rays.     

Forecasting Solar Flares Using Magnetogram-based Predictors and Machine Learning

We propose a forecasting approach for solar flares based on data from Solar Cycle 24, taken by the Helioseismic and Magnetic Imager (HMI) on board the Solar Dynamics Observatory (SDO) mission. In particular, we use the Space-weather HMI Active Region Patches (SHARP) product that facilitates cut-out magnetograms of solar active regions (AR) in the Sun in near-realtime (NRT), taken over a five-year interval (2012?–?2016). Our approach utilizes a set of thirteen predictors, which are not included in the SHARP metadata, extracted from line-of-sight and vector photospheric magnetograms. We exploit several machine learning (ML) and conventional statistics techniques to predict flares of peak magnitude \({>}\,\mbox{M1}\) and \({>}\,\mbox{C1}\) within a 24 h forecast window. The ML methods used are multi-layer perceptrons (MLP), support vector machines (SVM), and random forests (RF). We conclude that random forests could be the prediction technique of choice for our sample, with the second-best method being multi-layer perceptrons, subject to an entropy objective function. A Monte Carlo simulation showed that the best-performing method gives accuracy \(\mathrm{ACC}=0.93(0.00)\), true skill statistic \(\mathrm{TSS}=0.74(0.02)\), and Heidke skill score \(\mathrm{HSS}=0.49(0.01)\) for \({>}\,\mbox{M1}\) flare prediction with probability threshold 15% and \(\mathrm{ACC}=0.84(0.00)\), \(\mathrm{TSS}=0.60(0.01)\), and \(\mathrm{HSS}=0.59(0.01)\) for \({>}\,\mbox{C1}\) flare prediction with probability threshold 35%.     

Short-Term Correlation of Solar Activity and Sunspot: Evidence of Lifetime Increase

We perform a nonlinear study of the short-term correlation properties of the solar activity (daily range) in order to reveal their long-life variations. We estimate the lifetime of the high-frequency component of a Markov-type signal when the high-frequency component is modulated by a slowly varying multiplicative factor. This treatment is applied to different series of solar activity: Wolf Sunspot numbers (WSN), Sunspot Group numbers (SGN), and Royal Greenwich Observatory (RGO) sunspot group series. We obtain that all the lifetime estimates exhibit similar temporal variations that agree with the variations of the sunspot lifetimes directly measured from the RGO data and those of the sunspot areas. An increase of lifetimes by a factor 1.4 is observed from 1915 to 1940. At the same time, a stable ratio is observed between the sunspot group’s maximal area and the lifetime, confirming the Gnevyshev–Waldmeier-type relationship. The analysis identifies also time intervals where the homogeneity of the different time series may be questioned.     

基于深度学习的太阳F_10.7辐射通量的短期预报研究

F_10.7太阳辐射通量作为输入参数被广泛运用于大气经验模型、电离层模型等空间环境模型,其预报精度直接影响航天器轨道预报精度.采用时间序列法统计了太阳辐射通量F10.7指数和太阳黑子数(SSN)的关系,给出了两者之间的线性关系,在此基础上提出了一种基于长短时记忆神经网络(Long and Short Term Memory,LSTM)的预报方法,方法结合了54 d太阳辐射通量指数和SSN历史数据来对F_10.7进行未来7 d短期预报,并与其他预报方法的预报结果进行了比较,结果表明:(1)所建短期预报7 d方法模型的性能优于美国空间天气预报中心(Space Weather Prediction Center, SWPC)的方法,预测值和观测值的相关系数(CC)达到0.96,同时其均方根误差约为11.62个太阳辐射通量单位(sfu),预报结果的均方根误差(RMSE)低于SWPC,下降约11%;(2)对预测的23、24周太阳活动年结果统计表明,太阳活动高年的第7 d F10.7指数预报平均绝对百分比误差(MAPE)最优可达12.9%以内,低年最优可达2...     

Flare Prediction Using Photospheric and Coronal Image Data

The precise physical process that triggers solar flares is not currently understood. Here we attempt to capture the signature of this mechanism in solar-image data of various wavelengths and use these signatures to predict flaring activity. We do this by developing an algorithm that i) automatically generates features in 5.5 TB of image data taken by the Solar Dynamics Observatory of the solar photosphere, chromosphere, transition region, and corona during the time period between May 2010 and May 2014, ii) combines these features with other features based on flaring history and a physical understanding of putative flaring processes, and iii) classifies these features to predict whether a solar active region will flare within a time period of \(T\) hours, where \(T = 2 \mbox{ and }24\). Such an approach may be useful since, at the present time, there are no physical models of flares available for real-time prediction. We find that when optimizing for the True Skill Score (TSS), photospheric vector-magnetic-field data combined with flaring history yields the best performance, and when optimizing for the area under the precision–recall curve, all of the data are helpful. Our model performance yields a TSS of \(0.84 \pm0.03\) and \(0.81 \pm0.03\) in the \(T = 2\)- and 24-hour cases, respectively, and a value of \(0.13 \pm0.07\) and \(0.43 \pm0.08\) for the area under the precision–recall curve in the \(T=2\)- and 24-hour cases, respectively. These relatively high scores are competitive with previous attempts at solar prediction, but our different methodology and extreme care in task design and experimental setup provide an independent confirmation of these results. Given the similar values of algorithm performance across various types of models reported in the literature, we conclude that we can expect a certain baseline predictive capacity using these data. We believe that this is the first attempt to predict solar flares using photospheric vector-magnetic field data as well as multiple wavelengths of image data from the chromosphere, transition region, and corona, and it points the way towards greater data integration across diverse sources in future work.     

A Flare and Umbral Flashes in a Sunspot

We studied the evolution and dynamic processes in the chromosphere above a sunspot umbra. A relatively rarely occurring phenomenon of bright long-lasting emission observed in the umbra of a unipolar sunspot of the AR 9570 group on August 11, 2001 was investigated. It was found that during the course of the observation, emission was spreading, gradually occupying nearly the entire sunspot umbra. Based on the analysis of the observations from other observatories, we arrived at the conclusion that the bright emission was a sympathetic flare that occurred in the sunspot umbra. It was assumed that there occurred an interaction with a neighboring, rapidly evolving group that exhibited subflares on the day of observation. In the same umbra, there was taking place an oscillatory process of the type of umbral flash (observations from August 11 and 12, 2001). The characteristics of the oscillatory process in the presence of the flare were studied. As the bright emission propagated in the sunspot umbra, brightness fluctuations ceased to be seen in the umbral flashes against the background of this brighter emission. The character of velocity variations did not change substantially, although the oscillation amplitude did decrease.     

A Prediction of the Peak Sunspot Number for Solar Cycle 23

We use a precursor technique based on the geomagneticaa index during the decline (last 30%) of solar cycle 22 to predict a peak sunspot number of 158 (± 18) for cycle 23, under the assumption that solar minimum occurred in May 1996. This method appears to be as reliable as those that require a year of data surrounding the geomagnetic minimum, which typically follows the smoothed sunspot minimum by about six months.     

An Early Prediction of Maximum Sunspot Number in Solar Cycle 24

A non-linear coupling function between sunspot maxima and aa minima modulations has been found as a result of a wavelet analysis of geomagnetic index aa and Wolf sunspot number yearly means since 1844. It has been demonstrated that the increase of these modulations for the past 158 years has not been steady, instead, it has occurred in less than 30 years starting around 1923. Otherwise sunspot maxima have oscillated about a constant level of 90 and 141, prior to 1923 and after 1949, respectively. The relevance of these findings regarding the forecasting of solar activity is analyzed here. It is found that if sunspot cycle maxima were still oscillating around the 141 constant value, then the Gnevyshev–Ohl rule would be violated for two consecutive even–odd sunspot pairs (22–23 and 24–25) for the first time in 1700 years. Instead, we present evidence that solar activity is in a declining episode that started about 1993. A value for maximum sunspot number in solar cycle 24 (87.5±23.5) is estimated from our results.     

Modeling Solar Spectral Irradiance and Total Magnetic Flux Using Sunspot Areas

We show that daily sunspot areas can be used in a simple, single parameter model to reconstruct daily variations in several other solar parameters, including solar spectral irradiance and total magnetic flux. The model assumes that changes in any given parameter can be treated mathematically as the response of the system to the emergence of a sunspot. Using cotemporal observational data, we compute the finite impulse response (FIR) function that describes that response in detail, and show that the response function has been approximately stationary over the time period for which data exist. For each parameter, the impulse response function describes the physical evolution of that part of a solar active region that is the source of the measured variability. We show that the impulse response functions are relatively narrow functions, no more than 3 years wide overall. Each exhibits a pre-active, active, and post-active region component; the active region component dominates the variability of most of the parameters studied.     

On Mid-Term Periodicities in Sunspot Groups and Flare Index

In this work we study the mid-term periodicities (MTPs), between 1 and 2 years, of the sunspot groups and the flare index (FI), by separating the data into hemispheres and spectral bands (SBs) according to the most significant periodicities presented by these phenomena. We found that the MTP of sunspot groups has a diminished power during the Modern Minimum and an increased power during the Modern Maximum, with the exception of cycle 20. For flares, the MTP has a diminished power during the low activity cycle 20, and an increased power during cycles 21 and 22. Therefore, for both sunspot groups and FI, cycle 20 shows a very diminished power followed by the active and higher-power cycles 21 and 22; cycle 23 shows a weaker power than cycles 21 and 22. It is uncertain whether MTP can be a precursor of a long-term minimum of solar activity or not, as has been previously suggested. Also, there is no one-to-one correlation between the cycle intensity and the importance of MTP. Concerning the quasi-biennial periodicities and the theory of two kinds of dynamos, we notice the tendency that higher-power cycles mean weaker coupling in the model. Concerning the hemispheric north-south asymmetry, for sunspot groups the southern hemisphere dominates in most of the SBs, while for FI the northern hemisphere dominates for all the SBs. Additionally, the time lag found between the two hemispheres indicates that the degrees of coupling in the photosphere for sunspot groups and in the corona for flares are between moderate and strong. Finally, the modulation shown by the MTP time series suggests that these periodicities are the product of chaotic quasi-periodic processes and not of stochastic processes.