Influence of Extreme Solar Events on Geomagnetic Variations Using Machine Learning by IRJET Journal

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

Influence of Extreme Solar Events on Geomagnetic Variations Using Machine Learning

Shivam tiwari1, Dr. Achyut Pandey2, Dr. Laxmi Tripathi3

Dr. Yash Kumar Singh4 .

2Govt. T.R.S. College Rewa (M.P.) 1,3,4Govt. Model Science College Rewa (M.P.)

Abstract

Solar activity produces substantial modifications in Earth's near-space environment with extreme events such as solar flares andcoronalmassejections(CMEs)leadingthecharge.Solarflaresandcoronalmassejectionscreategeomagneticstormsthat interrupt satellite operations and communication systems along with power infrastructure by disrupting the interplanetary magnetic field and solar wind. The research examines how extreme solar events impact geomagnetic field changes using a machine learning methodology. Using a combination of solar wind parameters, geomagnetic indices (Dst, Kp), and flare classifications, we develop predictive models to forecast geomagnetic disturbances. Long Short-Term Memory (LSTM) networksandRandomForestclassifiersareappliedtohistoricaldatatoidentifyandanticipatestorm-scalefluctuationslinked to solar triggers. The results show that machine learning methods can effectively capture complex, nonlinear interactions in the solar-terrestrial system and provide meaningful forecasts of geomagnetic responses. This work contributes to space weather research by demonstrating how artificial intelligence can enhance early-warning systems for solar-induced geomagneticactivity.

Keywords:Extremesolarevents,geomagneticstorms,machinelearning,solarwind,spaceweatherforecasting

1. Introduction

The geomagnetic field of the Earth, a vital defense against detrimental cosmic radiation and solar particles, is constantly moldedanddisturbedbysolaraction.Ofallthenumerouscausesofgeomagneticvariations,extremesolarflaresandCMEsare the most severe external causes of geomagnetic disturbances. These events release huge energies and charged particles into theheliosphere,usuallyleadingtogeomagneticstormswhentheyinteractwiththeEarth'smagnetosphere[1].Theheightened frequency and magnitude of these solar phenomena, especially during phases of solar maximum, necessitate the creation of goodmodelscapableofpredictingtheirterrestrialimpact.

Historically, the study of solar-driven geomagnetic perturbations has utilized statistical correlations of solar wind variables and geomagnetic indices. Although these approaches have been highly effective in yielding key insights, they tend to fail to account for the nonlinear and temporally dependent characteristics of the solar-terrestrial system. The introduction of highresolutionobservational informationandimprovedcomputeralgorithms hasmademachinelearninga potentialnewavenue for space weather modeling. These algorithms excel at identifying complex patterns in large datasets, enabling predictive capabilitiesthatextendbeyondthereachoftraditionalphysics-basedmodels[2].

This research seeks to bridge the gap between space weather observation and actionable prediction by employing machine learning methods to assess and forecast geomagnetic variations driven by extreme solar events. By integrating solar wind parameters, X-rayflareclassifications,and geomagnetic indices, weaimtoconstruct a robustmodel thatcanlearnfrom past events and improve short-term forecasts. Such an approach not only deepens our understanding of solar-geophysical interactionsbutalsoenhancestheresilienceofcriticaltechnologicalinfrastructurevulnerabletospaceweatherimpacts.

2. Literature Review

Overthepastdecades,numerousstudieshaveexploredtheinteractionsbetweensolaractivityandgeomagneticfieldbehavior, particularly focusing on the impacts of solar flares and CMEs. The Dst and Kp indices, which quantify geomagnetic storm intensityandvariability,havebeencentraltotheseinvestigations.Burtonetal.[3]establishedearlyempiricalmodelslinking

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

solar wind pressure and southward interplanetary magnetic field (IMF) components to Dst variations. Their work laid the foundationforsubsequenteffortstoquantifysolarwind–magnetospherecoupling.

With the progression of computational capabilities, researchers began incorporating more sophisticated data analysis tools into the field. Machine learning techniques have gained traction, offering improved accuracy in forecasting geomagnetic storms.Forinstance,Camporeale[4]demonstratedthatneural networkscouldeffectivelymodelDstindexfluctuationsusing real-timesolarwinddata.Similarly,Tanetal.[5]employeddeeplearningmodelstopredictgeomagneticstormsandreported performancesurpassingthatoftraditionalautoregressivemodels.

Despitetheseadvancements,moststudieshavefocusedongeneralsolarwind–geomagneticrelationshipsratherthanisolating the influence of extreme solar events. This distinction is crucial, as high-energy phenomena like X-class flares and fast CMEs often produce more abrupt and severe geomagnetic responses compared to routine solar activity. Ji et al. [6] attempted to categorize CMEs by speed and associated their severity with Dst index outcomes, suggesting a nonlinear but discernible relationshipbetweenCMEpropertiesandgeomagneticdisturbancelevels.

Recent works have also applied ensemble machine learning methods, such as Random Forest and XGBoost, to classify geomagnetic conditions based on solar parameters [7]. These models benefit from interpretability and robustness but are sometimes limited in capturing temporal dynamics. To address this, recurrent neural networks (RNNs), especially LSTM architectures,havebeenintroducedtoleveragethesequentialnatureofspaceweatherdata[8].

This body of research underscores the need for hybrid models that can both classify and predict geomagnetic outcomes triggeredspecificallybyextremesolarevents.Ourstudybuildsuponthesefoundationsbydesigningamachinelearning-based systemthatnotonly processeshigh-frequencysolarandgeomagneticdata butalso learnstemporal dependenciescrucial for short-termforecasting.

3. Methodology

3.1 Data Sources and Preprocessing

For this research, we compiled a comprehensive dataset comprising multiple sources. Solar wind parameters were obtained fromNASA’sOMNIWebdatabase,whichconsolidatesmeasurementsfromACEandDSCOVRsatellitesat1-hourresolution.Key features extracted include solar wind speed (Vsw), proton density (Np), and the IMF components (particularly Bz). Geomagnetic activity was represented using the Dst and Kp indices, sourced from the World Data Center for Geomagnetism, Kyoto.

In addition, solar flare data, including event classification (C, M, X) and peak X-ray flux, were retrieved from NOAA's GOES satellite reports. To define the occurrence of extreme events, we selected all solar flares of M5.0 and above, as well as CMEs withspeedsexceeding800km/s.Eacheventwasmappedtoacorrespondingtimewindowofgeomagneticresponse(typically ±24to48hours)basedonpropagationdelayestimates.

The data were cleaned to remove missing values and normalized to ensure comparability across features. Time series were segmentedintooverlappingwindowstocapturebothpre-eventandpost-eventconditions,facilitatingthelearningoftemporal patterns.

3.2 Machine Learning Models

We implemented two primary machine learning models: a Long Short-Term Memory (LSTM) network and a Random Forest classifier. The LSTM model was designed to handle the sequential nature of the data and predict the likelihood of a geomagneticstorm(Dst< -50nT)givenatimeseriesofsolarparameters.Thenetworkconsistedofaninputlayer,twoLSTM layerswithdropoutregularization,andadenseoutputlayerwithasigmoidactivationfunction.

The Random Forest model, in contrast, was used to classify events into geomagnetic storm categories (quiet, moderate, intense) based on aggregated solar features. The model was trained on labeled events and validated using k-fold crossvalidationtoensurerobustness.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

3.3 Evaluation Metrics

Model performance was evaluated using standard classification and regression metrics. For the classification task, accuracy, precision, recall, and F1-score were computed. For the LSTM regression output, root mean square error (RMSE) and mean absolute error (MAE) were used. Additionally, the models' outputs were compared against actual Dst index values to assess predictivealignmentduringknownstormevents.

3.4 Implementation Tools

All modeling was conducted using Python. The TensorFlow and Keras libraries were used for the deep learning implementation, while scikit-learn was used for the Random Forest model. Data visualization and correlation analysis were performed using pandas, seaborn, and matplotlib. All code and data pipelines were executed on a GPU-enabled system to ensureefficienttraining.

4. Results and Discussion

Theperformanceoftheproposedmachinelearningmodelswasevaluatedthroughvariousexperimentsusinghistoricalspace weather datasets. Both the LSTM-based regression model and the Random Forest classifier were assessed in terms of their ability to capture geomagnetic disturbances resulting from extreme solar events. Below, we present and interpret the key findingsfromouranalysis.

4.1 Correlation Between Solar Wind Speed and Dst Index

Thefirststepinunderstandingsolar-inducedgeomagneticvariabilitywastoassessthecorrelationbetweensolarwindspeed andtheDstindex.AsshowninFigure1,thereisaclearinverserelationship,whereelevatedwindspeedsoftencoincidewith dropsintheDstindex anindicationofgeomagneticstormconditions.

Figure 1. This time-series plot reveals how increases in solar wind velocity (orange) typically precede or align with significant decreases in the Dst index (blue), confirming that high-speed streams are key drivers of geomagnetic storms.

4.2 Feature Importance in Storm Classification

Toidentifywhichparametersmostinfluencegeomagneticactivity,aRandomForestclassifierwastrainedusingkeysolarwind variables and solar event metrics. The importance scores of each feature are displayed in Figure 2. The interplanetary magneticfield’ssouthward component(Bz)emergedasthemostsignificantcontributor,followedbysolarwindspeed(Vsw) andCMEspeed.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Figure 2. The model assigned the highest weight to Bz, consistent with physical understanding that southward IMF enhances magnetospheric reconnection, leading to stronger geomagnetic disturbances.

4.3 Predictive Performance of LSTM Model

TheLSTMnetwork,trainedonsequentialinputwindowsofsolarwindandflaredata,demonstratedpromisingperformancein predictingfutureDstvalues.Figure3illustratesthemodel'soutputcomparedtoactualmeasurements.Whilesomedeviations arepresent,thepredictionslargelyfollowtherealtrendoftheDstindex,especiallyaroundstormonsettimes.

Figure 3. The LSTM model successfully captures temporal dependencies and provides a reasonably close approximation to observed geomagnetic field variations, especially during the peak activity intervals.

4.4 Classification Accuracy of Geomagnetic Storm Events

For classification, the model aimed to categorize events into "Quiet", "Moderate", or "Storm" levels based on Dst thresholds. TheconfusionmatrixshowninFigure4illustratestheperformanceoftheclassifieracrossthethreeclasses.Mostpredictions alignwithactuallabels,althoughsomemisclassificationsoccurbetweenadjacentcategories,likelyduetothegradualnatureof stormintensification.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

Figure 4. The classifier achieves a high overall accuracy, particularly in identifying intense geomagnetic storm events, making it a useful tool for early warning applications.

4.5 Interpretation and Implications

The findings support the conclusionthat machinelearning models when properly trainedandvalidated canmeaningfully anticipategeomagnetic disturbances based on solarinputdata.In particular, the LSTMnetwork offers valuein capturing the temporalevolutionofstormevents,whiletheRandomForestclassifierprovidestransparencyinfeaturerelevance.

These results have practical implications. In a real-time monitoring setting, integrating such models could enhance earlywarning capabilities for power grids, satellites, and aviation systems. Furthermore, the clarity offered by feature importance analysisaidsresearchersandoperationalforecastersinunderstandingwhichparametersmostconsistentlysignaltheonsetof geomagneticactivity.

Despite the encouraging outcomes, it is essential to recognize limitations. The dataset, while extensive, is subject to observational gaps and inherent uncertainties. In future work, expanding the dataset to include more solar cycles and incorporatingensemblelearningcouldfurtherimprovegeneralization.

5. Conclusion

Thisstudyexploredthedynamicrelationshipbetweenextremesolareventsandgeomagneticfieldvariationsusingadvanced machine learning techniques. By integrating real-time solar wind parameters, solar flare classifications, and geomagnetic indices, we developed predictive models capable of capturing both the classification and forecasting dimensions of space weather impacts. Our Long Short-Term Memory (LSTM) network effectively modeled the temporal evolution of geomagnetic disturbances, while the Random Forest classifier provided a clear understanding of the relative importance of solar features contributingtostorm-levelactivity.

The findings affirm that machine learning not only complements traditional empirical models but also brings a new level of adaptability and precision to geomagnetic forecasting. Specifically, the models demonstrated substantial potential in forecastingsuddendipsintheDstindexfollowinghigh-velocitysolarwindstreamsandsouthward-directedmagneticfields.

Inaddition,themodels'capacitytoidentifypatternsbeforegeomagneticeventsprovidesreal-worldusefulness forthe earlywarningsystemsinsatelliteoperations,navigation,andpowerinfrastructuredefense.

International Research Journal of Engineering and Technology (IRJET) e-ISSN:2395-0056

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

However, like all data-driven methodologies, the accuracy of our framework relies significantly on the completeness and qualityofinputdata.Futurestudiesmaygainfromtheintegrationofotherdatasources,includingsolarradioburstsignatures and polar cap indices, as well as investigating hybrid models combining physics-based simulations with neural network predictions. On the whole, the inclusion of smart algorithms in space weather research is a promising direction toward proactiveavoidancemeasuresforgeomagneticrisks.

References

Gonzalez, W. D., Joselyn, J. A., Kamide, Y., Kroehl, H. W., Rostoker, G., Tsurutani, B. T., & Vasyliunas, V. M. (1994). What is a geomagneticstorm?JournalofGeophysicalResearch:SpacePhysics,99(A4),5771–5792.https://doi.org/10.1029/93JA02867

Burton, R. K., McPherron, R. L., & Russell, C. T. (1975). An empirical relationship between interplanetary conditions and Dst JournalofGeophysicalResearch,80(31),4204–4214.https://doi.org/10.1029/JA080i031p04204

Camporeale,E.(2019).Thechallengeofmachinelearninginspaceweather:Nowcastingandforecasting.SpaceWeather,17(8), 1166–1207.https://doi.org/10.1029/2018SW002061

Tan, B., Huang, J., Zhou, Z., & Chen, B. (2018). Deep learning in solar flare forecasting: Models and data preprocessing. Solar Physics,293(4),61.https://doi.org/10.1007/s11207-018-1261-5

Bobra,M.G.,& Couvidat,S. (2015).Solar flare predictionusingSDO/HMI vector magneticfielddata with a machine-learning algorithm.TheAstrophysicalJournal,798(2),135.https://doi.org/10.1088/0004-637X/798/2/135

Ji, E. Y., Moon, Y. J., Lee, H., Kim, R. S., & Lee, D. H. (2012). Comparison of geomagnetic storm forecasting using CME and flare parameters.JournalofGeophysicalResearch:SpacePhysics,117(A3),A03209.https://doi.org/10.1029/2011JA017477

Mishra, N., & Mishra, A. K. (2020). Prediction of geomagnetic storms using machine learning techniques. Advances in Space Research,66(6),1301–1313.https://doi.org/10.1016/j.asr.2020.05.026

Zhang, M., Moldwin, M. B., & Cutler, J. W. (2021). Recurrent neural networks for modeling solar wind–geomagnetic coupling. SpaceWeather,19(3),e2020SW002688.https://doi.org/10.1029/2020SW002688

Liu,C.,Deng,N.,Wang,J.T.,&Wang,H.(2017).Predictingsolarflaresusinganoveldeeplearningalgorithm.TheAstrophysical Journal,843(2),104.https://doi.org/10.3847/1538-4357/aa789b

Wintoft,P.,Wik,M.,&Viljanen,A.(2015).Solarwinddrivenregressionmodelingofgeomagneticdisturbances.SpaceWeather, 13(10),686–697.https://doi.org/10.1002/2015SW001206

Sharifie, K., & Eftekharnejad, S. (2021). Forecasting geomagnetic storms using machine learning techniques based on solar winddata.Earth,PlanetsandSpace,73(1),106.https://doi.org/10.1186/s40623-021-01413-1

Schrijver, C. J., & Siscoe, G. L. (2010). Heliophysics: Space Storms and Radiation: Causes and Effects. Cambridge University Press,ISBN:9780521760515.