With the rapid growth of seismic data volumes, traditional automated processing methods, which have been in use for decades, face increasing challenges in handling these data, especially in noisy environments. Deep learning (DL) methods, due to their ability to handle large datasets and perform well in complex scenarios, offer promising solutions to these challenges. When I started my Ph.D. degree, although a sizeable number of researchers were beginning to explore the application of deep learning in seismology, almost no one was involved in the development of much-needed automated data annotation tools and deep learning training platforms for this field. In other rapidly evolving fields of artificial intelligence, such automated tools and platforms are often a prerequisite and critical to advancing the development of deep learning. Motivated by this gap, my Ph.D. research focuses on creating these essential tools and conducting critical investigations in the field of earthquake detection and phase picking using DL methods. The first research chapter introduces QuakeLabeler, an open-source Python toolbox that facilitates the efficient creation and management of seismic training datasets. This tool aims to address the laborious process of producing training labels in the vast amount of seismic data available today. Building on this foundational tool, the second research chapter presents Blockly Earthquake Transformer (BET), a deep learning platform that provides an interactive dashboard for efficient customization of deep learning phase pickers. BET aims to optimize the performance of seismic event detection and phase picking by allowing easy customization of model parameters and providing extensions for transfer learning and fine-tuning. The third and final research chapter investigates the performance of DL pickers by examining the effect of training data size and deployment settings on phase picking accuracy. This investigation provides insight into the optimal size of training datasets, the suitability of DL pickers for new target regions, and the impact of various factors on training and on model performance. Through the development of these tools and investigations, this thesis contributes to the application of DL in seismology, paving the way for more efficient seismic data processing, customizable model creation, and a better understanding of DL model performance in earthquake detection and phase-picking tasks.

Earthquake detection is an important task, focusing on detecting seismic events in past data or in real time from seismic time series. In the past few decades, due to the increasing amount of available seismic data, research in seismic event detection shows remarkable success using neural networks and other machine learning techniques. However, creating high quality labeled data sets is still a manual process that demands tremendous amount of time and expert knowledge, and is stifling big data innovation. When compiling a data set, it is unclear how many earthquakes and noise are mislabeled. Another challenge is how to promote the general applicability of the machine learning based models to different geographical regions. The models trained by data sets from one location should be applicable to the detection at other locations. This thesis explores the most popular deep learning model, convolutional neural networks (CNN), to build a single location detection model. In addition, we build more robust generalized earthquake detection models using transfer learning and meta learning. We also introduce a process for generating high quality labeled datasets. Our technique achieves high detection accuracy even on low signal to noise ratio events. The AI techniques explored in this research have potential to be transferred to other domains that utilize signal processing. There are a myriad of potential applications, with audio processing probably being one of the most directly relevant. Any field that deals with waveforms (e.g. seismic, audio, light) can utilize the developed techniques.

In this study earthquake data is analyzed using a newly developed Migration Based Detection (MBD) algorithm (Wagner et al. 2017). A software environment suitable for manual analysis of large quantities of earthquakes (events) detected by the MBD algorithm is set up, and the MBD algorithm is applied to 13 days of seismic data from a network of 26 seismic stations in the geologically complex Hengill-area in southwest Iceland. A total of 859 event detections are produced and manually inspected. Out of these, 483 are considered true and/or uncertain, making the assessed number of false detections about 44%. A subset of 53 well defined true events are selected for event relocation using manual picking of first arrival P-waves. The relocation resulted in a mean difference of roughly 0.6 km for each coordinate in the horizontal plane and about 1.4 km in depth. Results of the study provide reference data that may aid further development of the MBD algorithm, as well as provide some insight into the performance of the MBD algorithm. The software environment tailored for analyzing events detected by the MBD algorithm may be used as a foundation for continued analysis of detected events. / I denna studie analyserades jordskalvsdata med hjälp av en nyligen utvecklad 'migration based detection'-algoritm (Wagner et al. 2017). En mjukvarumiljö skräddarsydd för manuell analys av stora kvantiteter av jordskalv detekterade av MBD-algoritmen iordningställdes, varpå MBD-algoritmen sedan applicerades på 13 dagar av seismisk data från ett nätverk av 26 seismiska stationer i det geologiskt sett komplexa Hengill-området i sydvästra Island. Totalt detekterades 859 jordskalv som genomgick manuell analys. Av dessa klassificerades 483 stycken som bekräftade eller troliga jordskalv, vilket resulterar i en uppskattad felmarginal om ca. 44 %. En delmängd om 53 väldefinierade jordskalv valdes ut för noggrannare analys av ursprungsplats och tidpunkt genom manuell plockning av P-fasankomst. Omlokaliseringen resulterade i en genomsnittlig differens om ca. 0.6 km i vardera koordinat i horisontalplanet och ca. 1.4 km i höjdled. Resultat från projektet kan användas som referensdata vid vidareutveckling av MBD-algoritmen samt ger viss insyn i prestandan hos MBD-algoritmen. Den iordningställda datormiljön kan användas som bas för vidare analys av jordskalv detekterade av MBD-algoritmen.

Earthquakes pose a significant threat as one of the most destructive natural disasters globally. Despite ongoing efforts to predict earthquakes, the success of such research remains a challenge, necessitating interdisciplinary research and collaboration. The EU-funded Project artEmis seeks to address this challenge in earthquake-prone regions of Europe by establishing a multi-sensor Internet of Things (IoT) network to monitor radon gas concentration in groundwater near fault lines. This thesis focuses on supporting the project by developing the software interface for the transmission of data from a gateway controller to the cloud interface, addressing key functionalities. The goal is to establish two-way communication between the gateway controller and the cloud using the MQTT-SN protocol. Additionally, other functionalities such as data storage and sensor data anomaly detection are also explored and implemented. This research employs an applied research approach, consisting of literature reviews, development, and evaluation phases. The development phase consists of the design and implementation of memory storage, data generation, and bi-directional communication features which align with the project goals. The evaluation of the software development process is achieved through a comprehensive set of functional and end-to-end tests. These tests examine the intricacies of the different software components, with rigorous evaluation against all test criteria and project requirements. The evaluation process concluded with a favourable outcome, indicating that all tests were successful. Additionally, a detailed evaluation of memory capacity was conducted to understand the system’s data retention capability, alongside an analysis of throughput and latency. The memory evaluation demonstrated efficient allocation within the processor’s memory, offering 2.7 days of data storage with specific intervals. Throughput analysis revealed a positive correlation between larger data packets and increased transfer rates, and latency increased with larger packets, possibly due to network congestion and processing delays. However, it is important to acknowledge several inherent limitations in this work, including constrained bi-directional communication capabilities, the absence of a serial interface with sensors, limitations in report size, and constraints on storage capacity. These factors serve as essential contextual considerations for the scope and capabilities of our project. In summary, this research supports Project artEmis by developing a vital software interface for the IoT network. Successful evaluation of the software through comprehensive testing signifies a significant step forward in earthquake monitoring. Despite certain limitations, this work contributes to enhancing our understanding and response to seismic threats. / Jordbävningar utgör ett betydande hot som en av de mest förödande naturliga katastroferna globalt sett. Trots pågående ansträngningar att förutsäga jordbävningar förblir framgången för sådan forskning en utmaning, vilket kräver tvärvetenskaplig forskning och samarbete. Det EU-finansierade projektet artEmis syftar till att belysa detta problem i jordbävningsbenägna regioner i Europa genom att etablera ett multisensor Internet of Things (IoT)-nätverk för att övervaka radongaskoncentrationen i grundvatten nära förkastningslinjer. Denna uppsats fokuserar på att stödja projektet genom att utveckla programvarugränssnittet för överföring av data från en gateway-controller till molngränssnittet samt upplyser viktiga funktioner. Målet är att etablera tvåvägskommunikation mellan gateway-controllern och molnet med hjälp av MQTT-SN-protokollet. Dessutom utforskas och implementeras även andra funktioner, såsom datalagring och detektering av avvikelser i sensordata. Denna forskning använder en tillämpad forskningsmetod som består av litteraturstudier, utveckling och utvärderingsfaser. Utvecklingsfasen innefattar design och implementering av funktioner för minneslagring, datagenerering och tvåvägskommunikation som överensstämmer med projektets mål. Utvärderingen av programvaruutvecklingsprocessen uppnås genom omfattande funktionella och slutanvändartester. Dessa tester granskar de olika programvarukomponenternas komplexiteter och utvärderas noggrant mot alla testkriterier och projektets krav. Utvärderingsprocessen avslutades med ett gynnsamt resultat, vilket indikerar att alla tester var framgångsrika. Dessutom gjordes en detaljerad utvärdering av minneskapaciteten för att förstå systemets datalagringsförmåga, tillsammans med en analys av genomströmning och latens. Minnesutvärderingen visade på effektiv allokering i processorns minne och erbjöd 2,7 dagars datalagring med specifika intervall. Genomströmningsanalys avslöjade en positiv korrelation mellan större datapaket och ökade överföringshastigheter, och latensen ökade med större paket, möjligen på grund av nätverksstockning och bearbetningsförseningar. Det är dock viktigt att erkänna att det finns flera begränsningar i forskningen, inklusive begränsade tvåvägskommunikationsmöjligheter, begränsningar i rapportstorlek och lagringskapacitet, och ett saknande av seriellt gränssnitt med sensorer. Dessa faktorer är viktiga för förståelsen av omfattningen och förmågorna hos vårt projekt. Sammanfattningsvis stöder denna forskning Project artEmis genom att utveckla en avgörande programvarugränssnitt för IoT-nätverket. Den framgångsrika utvärderingen av programvaran genom omfattande tester har lett till ett betydande steg framåt inom jordbävningsövervakning. Trots vissa begränsningar bidrar detta arbete till att förbättra vår förståelse och förmåga att reagera på seismiska händelser.