Detecting tsunami-generated magnetic fields using machine learning

Detecting tsunami-generated magnetic fields using machine learning | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Detecting tsunami-generated magnetic fields using machine learning CHIAKI MITA, Takuto Minami, Jan Saynisch Wagner, Aaron Hornschild, and 2 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6441319/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 5 You are reading this latest preprint version Abstract Conductive seawater moving through the Earth’s magnetic field generates electromagnetic fields. When this motion is driven by a tsunami, the phenomenon is referred to as a tsunami-generated electromagnetic field. Previous studies have explored its characteristics and potential applications for tsunami early warning systems. In this study, we present a novel approach utilizing machine learning to automatically detect tsunami-generated magnetic (TGM) fields at the seafloor. To train our model, we prepared a large dataset by combining simulated TGM signals with non-tsunami magnetic data observed at the seafloor of the Northwest Pacific Ocean and the Philippine Sea. A convolutional neural network was selected for the architecture of our model, successfully detecting TGM signals from the 2006 and 2007 Kuril earthquakes observed at the seafloor of the Northwest Pacific Ocean. Notably, the model also identified TGM signals from the 2009 Samoa earthquake at seafloor sites in French Polynesia—locations that were not included in the training data. These results demonstrate the effectiveness of applying machine learning to TGM signal detection. Furthermore, we evaluated our model’s performance based on the signal-to-noise (S/N) ratio and the signal duration in the input data, establishing quantitative criteria for detection. Our analysis showed that successful detection requires an S/N ratio greater than 2 for horizontal components and greater than 5 for the vertical component. As for the TGM signal points, our model requires a minimum TGM signal duration of 10 minutes for reliable detection. These criteria, which have not been proposed previously, provide valuable guidelines for detection of TGM signals at the seafloor. Tsunami Electromagnetic fields Machine learning Convolutional Neural Network Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Full Text Supplementary Files Graphicalabstract.jpg Supplementarymaterial.pdf Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Major Revision 21 Sep, 2025 Reviewers agreed at journal 25 Aug, 2025 Reviewers invited by journal 14 Jul, 2025 Editor assigned by journal 02 Jul, 2025 First submitted to journal 01 Jul, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6441319","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":485323974,"identity":"0bfe2f32-fc7a-4939-8454-c119c4a9950b","order_by":0,"name":"CHIAKI MITA","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAz0lEQVRIiWNgGAWjYFACxgYGBh6GBH4QO6GAOC2NDSAtkg0gLQbEWwNUfADEJkYL/7TD7Q9+yNjkGZ9fnfjhgQGDPL/YAfxaJG4nNjb28KQVm914u1kC6DDDmbMT8GsxkE5sbODhOZy47cbZDSAtCQa3idDS+AeoZfOMs5t/EK2lGWTLBv7ebcTZAvLLbBmgXyRu8G6zSDCQIOwX/tnpDz6+7bHJ4+8/u/nmjwobeX5pAlrAgLEHZB9YpQQRysHgB8i+A8SqHgWjYBSMgpEGADOMR6Ru4lDlAAAAAElFTkSuQmCC","orcid":"https://orcid.org/0009-0003-9398-7405","institution":"Kobe University Graduate School of Science Faculty of Science: Kobe Daigaku Daigakuin Rigaku Kenkyuka Rigakubu","correspondingAuthor":true,"prefix":"","firstName":"CHIAKI","middleName":"","lastName":"MITA","suffix":""},{"id":485323975,"identity":"a579980c-2469-4080-a0a3-04cb454389a4","order_by":1,"name":"Takuto Minami","email":"","orcid":"","institution":"Kobe University: Kobe Daigaku","correspondingAuthor":false,"prefix":"","firstName":"Takuto","middleName":"","lastName":"Minami","suffix":""},{"id":485323976,"identity":"07f2e69f-e9ce-4997-86eb-2fba77bab3de","order_by":2,"name":"Jan Saynisch Wagner","email":"","orcid":"","institution":"GFZ: Deutsches Geoforschungszentrum Potsdam","correspondingAuthor":false,"prefix":"","firstName":"Jan","middleName":"Saynisch","lastName":"Wagner","suffix":""},{"id":485323977,"identity":"3b4d89fd-5cdc-429e-a4f3-e0ee7961aabe","order_by":3,"name":"Aaron Hornschild","email":"","orcid":"","institution":"GFZ: Deutsches Geoforschungszentrum Potsdam","correspondingAuthor":false,"prefix":"","firstName":"Aaron","middleName":"","lastName":"Hornschild","suffix":""},{"id":485323978,"identity":"80c7ef89-44b6-42ea-8041-e43ed741a36b","order_by":4,"name":"Hiroko Sugioka","email":"","orcid":"","institution":"Kobe University: Kobe Daigaku","correspondingAuthor":false,"prefix":"","firstName":"Hiroko","middleName":"","lastName":"Sugioka","suffix":""},{"id":485323979,"identity":"a46f337b-bbc4-4744-9319-47a5dc612708","order_by":5,"name":"Hiroaki Toh","email":"","orcid":"","institution":"Kyoto University: Kyoto Daigaku","correspondingAuthor":false,"prefix":"","firstName":"Hiroaki","middleName":"","lastName":"Toh","suffix":""}],"badges":[],"createdAt":"2025-04-13 23:17:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6441319/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6441319/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":86929970,"identity":"3b2ed447-e1c6-4fdc-bee4-fc101e3b1466","added_by":"auto","created_at":"2025-07-17 09:25:04","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":64903,"visible":true,"origin":"","legend":"\u003cp\u003eArchitecture of our model. The shape of the data through each layer is shown below. For the initial shape of the input data, refer to section 2.3.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/06fdf05485062713c40a7f5a.png"},{"id":86930279,"identity":"5afa3445-6f88-49d8-9ccb-11c8167b9c31","added_by":"auto","created_at":"2025-07-17 09:33:04","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":205658,"visible":true,"origin":"","legend":"\u003cp\u003eMap of observation points and tsunami wave sources. Red circles indicate the observation points, while orange stars indicate wave sources. Alphabets of indexes beside the orange stars correspond to the areas of the ten source scenarios: K indicates the Kuril, T indicates Tohoku, and N indicates Nankai, respectively.\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/77be727abff2c9086b442aba.png"},{"id":86929977,"identity":"dc8b5af0-d5c8-48dd-a002-a03a56e9d51a","added_by":"auto","created_at":"2025-07-17 09:25:04","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":252901,"visible":true,"origin":"","legend":"\u003cp\u003eProcess of model training and performance on validation and real magnetic datasets. (a)\u003cstrong\u003e \u003c/strong\u003eLoss values for training and validation datasets across epochs. (b) Training and validation accuracy across epochs. (c)\u003cstrong\u003e \u003c/strong\u003eDistribution of the model outputs categorized into four groups. (d) Classification matrix of model outputs, where the horizontal axis represents the answer label, while the vertical axis represents the output value. (e) and (f) correspond to the Kuril 2006 earthquake events and the Kuril 2007 earthquake events, respectively, while (g) represents periods of no earthquake events before the 2006 event. The top and middle panels indicate observed magnetic data filtered using a high-pass filter with the cut-off of 60 minutes, while the bottom panel shows the model output values. The red and blue vertical solid lines in (e) and (f) indicate the ETA of seismic and tsunami waves, respectively. The horizontal magenta dashed lines in (e) to (g) indicate 1.0 of output values. Note that the start and end times of the predictions differ from those of the magnetic data because the prediction value is located at the end of 30-minute segment of the magnetic data.\u003c/p\u003e","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/c982cb39ed36fac2f432d555.png"},{"id":86929971,"identity":"01a5f850-e4ab-48be-a2f5-93779ac2fbe6","added_by":"auto","created_at":"2025-07-17 09:25:04","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":520852,"visible":true,"origin":"","legend":"\u003cp\u003eThe prediction results for the 2009 Samoa earthquake event. (a) Map showing the origin of the 2009 Samoa earthquake and the TIARES sites. (b) Enlarged view highlighting the TIARES sites. Red triangles indicate locations with OBEMs, while the yellow triangle indicates a site equipped with both an OBEM and DPG. (c) Results for SOC8. The top panel displays the observed sea surface displacement, while the middle two panels show the observed magnetic fields: green, red and blue represent the downward, northward and eastward components, respectively. The bottom panel shows the model output. Note that the outputs are plotted at the time corresponding to the end of the 30-minute time window for input data. (d) Results for all TIARES sites, with three panels per site corresponding to the bottom three panels in (c). In (c) and (d), the vertical light blue lines indicate the ETAs of the tsunami.\u003c/p\u003e","description":"","filename":"Figure4r.png","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/ea453d1740b04be709e2a44a.png"},{"id":86930277,"identity":"7327240c-ec4b-45f7-bca2-7dfc5c143b93","added_by":"auto","created_at":"2025-07-17 09:33:04","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":255130,"visible":true,"origin":"","legend":"\u003cp\u003eAccuracy of the model in terms of the S/N and TGM signal points. (a) Histograms illustrating the number of model outputs with respect to the S/N for bx, by, bz and bH. The color of each bar indicates whether the model output is correct (red) or not (light blue). Additionally, the accuracy for each S/N bin is plotted on the right axis. (b) An example illustrating how we counted the number of TGM signal points. Each of the three panels shows the TGM simulation results. The horizontal axis represents the time taken from simulation, and the vertical axis represents the amplitude of the magnetic fields. In each panel, orange dots indicate the TGM signal points we defined. We set the first time window to contain at least one signal point from the vertical component. After counting the TGM signal points in each component, we employ the largest number of TGM signal points. (c) Accuracy for various min S/N plotted against the number of TGM signal points. See main text for the definition of min S/N. (d) Heatmaps showing the relationship between the number of TGM signal points and min S/N. The color bar indicates the number of input TGM data in each bin.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/c63d9ac6a01998e9f9d2b16f.png"},{"id":86931468,"identity":"43435e49-9b9d-480a-aa4b-f5062e12028d","added_by":"auto","created_at":"2025-07-17 09:49:07","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1739769,"visible":true,"origin":"","legend":"","description":"","filename":"EPSdraftrevisionr.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1_covered_53365abd-2a14-4ada-908a-9cc06c416172.pdf"},{"id":86931328,"identity":"bededd8d-dc40-4fc1-86e6-0708ae375fe6","added_by":"auto","created_at":"2025-07-17 09:41:05","extension":"jpg","order_by":9,"title":"","display":"","copyAsset":false,"role":"supplement","size":456835,"visible":true,"origin":"","legend":"","description":"","filename":"Graphicalabstract.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/8061a82934784abfe22f0d9d.jpg"},{"id":86929980,"identity":"c32462c9-fd80-419d-b502-df9bd18090f1","added_by":"auto","created_at":"2025-07-17 09:25:04","extension":"pdf","order_by":10,"title":"","display":"","copyAsset":false,"role":"supplement","size":544084,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6441319/v1/eaf26f2d3b640cf57592955c.pdf"}],"financialInterests":"","formattedTitle":"Detecting tsunami-generated magnetic fields using machine learning","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":true,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"earth-planets-and-space","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"epsp","sideBox":"Learn more about [Earth, Planets and Space](http://earth-planets-space.springeropen.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/epsp/default.aspx","title":"Earth, Planets and Space","twitterHandle":"@SpringerOpen","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Tsunami, Electromagnetic fields, Machine learning, Convolutional Neural Network","lastPublishedDoi":"10.21203/rs.3.rs-6441319/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6441319/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eConductive seawater moving through the Earth’s magnetic field generates electromagnetic fields. When this motion is driven by a tsunami, the phenomenon is referred to as a tsunami-generated electromagnetic field. Previous studies have explored its characteristics and potential applications for tsunami early warning systems. In this study, we present a novel approach utilizing machine learning to automatically detect tsunami-generated magnetic (TGM) fields at the seafloor. To train our model, we prepared a large dataset by combining simulated TGM signals with non-tsunami magnetic data observed at the seafloor of the Northwest Pacific Ocean and the Philippine Sea. A convolutional neural network was selected for the architecture of our model, successfully detecting TGM signals from the 2006 and 2007 Kuril earthquakes observed at the seafloor of the Northwest Pacific Ocean. Notably, the model also identified TGM signals from the 2009 Samoa earthquake at seafloor sites in French Polynesia—locations that were not included in the training data. These results demonstrate the effectiveness of applying machine learning to TGM signal detection. Furthermore, we evaluated our model’s performance based on the signal-to-noise (S/N) ratio and the signal duration in the input data, establishing quantitative criteria for detection. Our analysis showed that successful detection requires an S/N ratio greater than 2 for horizontal components and greater than 5 for the vertical component. As for the TGM signal points, our model requires a minimum TGM signal duration of 10 minutes for reliable detection. These criteria, which have not been proposed previously, provide valuable guidelines for detection of TGM signals at the seafloor.\u003c/p\u003e","manuscriptTitle":"Detecting tsunami-generated magnetic fields using machine learning","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-17 09:24:59","doi":"10.21203/rs.3.rs-6441319/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Major Revision","date":"2025-09-21T08:14:38+00:00","index":"","fulltext":""},{"type":"reviewerAgreed","content":"","date":"2025-08-25T14:53:45+00:00","index":0,"fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-07-14T17:43:48+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-07-02T18:30:42+00:00","index":"","fulltext":""},{"type":"submitted","content":"Earth, Planets and Space","date":"2025-07-02T00:46:24+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"earth-planets-and-space","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"epsp","sideBox":"Learn more about [Earth, Planets and Space](http://earth-planets-space.springeropen.com)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/epsp/default.aspx","title":"Earth, Planets and Space","twitterHandle":"@SpringerOpen","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"BMC/SO AJ","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"1d8adf28-0c13-4fd6-8fb1-351810e4c595","owner":[],"postedDate":"July 17th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-03-10T10:07:56+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-17 09:24:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6441319","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6441319","identity":"rs-6441319","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}