Thesis title: Advancing GNSS Ionospheric Seismology for Tsunami Early Warning Systems: Small-Scale Event Detection and Data-Driven Enhancements
Global Navigation Satellite Systems (GNSS), originally designed for positioning and navigation, are now routinely employed for Ionospheric sounding as well. Indeed, GNSS observations can be used to estimate variations in the ionospheric total electron content (TEC), which represents the overall electron content in the ionosphere along the line-of-sight path between a satellite and a receiver.
Particularly, GNSS-TEC data have proven valuable in detecting ionospheric disturbances, also known as Traveling Ionospheric Disturbances (TIDs), generated by earthquakes and tsunamis, defining the so-called GNSS Ionospheric seismology. Indeed, earthquakes and tsunamis can produce acoustic and gravity waves (AGWs) which, due to the decrease in atmospheric density, can reach ionospheric heights, causing disturbances in the electron content. Specifically, TEC variations can provide timely information about the uplift at the source (acoustic and gravity waves epicenter - AGWepi), offering insights into tsunami genesis within the first 10 minutes after the shock. Additionally, TEC variations associated with the propagation of internal gravity waves (IGWs) allow for the tracking of offshore tsunami waves, where traditional data are often lacking. Therefore, TIDs analysis through GNSS-TEC can provide real-time/near real-time, continuous, wide-area coverage, that could significantly enhance the performance of tsunami early warning systems (TEWS). Indeed, TEWS currently rely primarily on seismic and tide gauge data, which can potentially result in delayed or inaccurate tsunami alerts.
While the ionospheric response to large-scale seismic events is well established, there is a gap in understanding the detection of seismic events classified as ”small” from an ionospheric perspective. This specifically refers to earthquakes with Mw∼7 (which are near to the Mw 6.5 threshold where pronounced TEC variations typically do not occur), and minor tsunamis with wave heights of just a few meters. Despite their small wave heights, these tsunamis can still have significant impacts, highlighting the need for improved ionospheric detection capabilities. This work aims to fill this gap and investigates if the ionospheric signature coming from these small events can also be used for augmenting TEWS. In detail, we use the VARION (Variometric Approach for Real-Time Ionosphere Observation) algorithm, able to estimate real-time TEC variations, to detect ionospheric perturbations caused by these small-scale events. The 2020 Samos Mw 7.0 earthquake in the Mediterranean Sea is analyzed as a case study. The shock also produced tsunami waves peaking around 3 meters on Samos Island, while reaching 1.38 m and 1.18 m respectively on the islands of Chios and Ikaria. TEC variations (up to 0.3 TECU), linked to the propagation of the internal gravity waves (IGWs) caused by the small tsunami, are identified. Moreover, comparison between IGWs arrival times in the ionosphere and the tsunami wave arrival at tide gauges reveals that the optimal observation geometries of ionospheric TEC detects the tsunami’s arrival before it reached Kos and Hraklreio coastlines. This highlights that TEC variations, though slight, can be used to complement existing tsunami early warning systems, particularly in the Mediterranean region where such phenomena are not deeply investigated.
In this context, implementing a reliable and automated detection system for tsunami-generated ionospheric disturbances within GNSS-TEC data is also essential for enhancing TEWS. Traditional methods like ionospheric power index analysis, wavelets, and 2D principal component analysis depend heavily on human expertise, making real-time, scalable detection a challenge. Furthermore, the vast volume, variety, and velocity of GNSS-TEC data present an opportunity to overcome this limitation by leveraging machine learning and deep learning techniques. To this point, this work also investigates machine learning (ML) techniques to automatically identify ionospheric perturbations caused by seismic events. To ensure the definition of a reliable model, the investigation begins with large-scale events that produce clear and distinct signatures in the ionosphere. In particular, a ML algorithm is developed using TEC data from the 2015 Illapel earthquake and tsunami and validated on the 2011 Tohoku and 2023 Turkish events. In detail, VARION-generated observations provided by 115 GNSS stations are used as input features for the machine learning algorithms, namely, Random Forest and XGBoost. The problem is approached as a supervised binary classification task, where the time frames of TEC perturbations are manually labelled as the target variable. XGBoost with a 15° elevation cut-off dsTEC/dt time series reaches the best performance, achieving an F1 score of 0.77, recall of 0.74, and precision of 0.80 on the test data. Furthermore, XGBoost presents an average difference between the detected and labeled TEC perturbation time frame (middle epochs) of 75 seconds. This work demonstrates high-probability TEC signature detection by machine learning for earthquakes and tsunamis, that can be used to enhance tsunami early warning systems. Therefore, acknowledging the importance of external validation on entirely different data sets to assess the model’s performance in different scenarios and the ability to generalization, the selected model is tested on the 2011 Tohoku and 2023 Turkish events. Additionally, an improved ML model is proposed, trained on data from multiple seismic events (2015 Illapel earthquake and tsunami, 2011 Tohoku earthquake and tsunami, and 2023 Turkey seismic sequence), which enhances the generalization of the algorithm across different scenarios. Thus, applying data detrending as a pre-processing technique significantly reduces the model’s failures in detecting TEC variations (False Negatives-FNs) and improves the alignment of labeled and predicted middle epochs of TEC perturbation. This enhances the ability of the model to generalize and detect TEC variations related to different seismic events. In conclusion, this work underscores the critical role of GNSS ionospheric real-time TEC data in strengthening tsunami early warning systems. Detecting ionospheric disturbances linked to smaller yet impactful tsunamis, alongside the integration of AI for the automatic and reliable identification of TIDs in TEC data, are essential steps toward ensuring a more timely and reliable TEWS.
Future research should focus on identifying common ionospheric patterns from smaller tsunamis, fine-tuning machine learning algorithms for real-time applications, extending them to various types of TIDs and data sets and implementing operatively the framework. Additionally, machine learning algorithms should be tested on smaller-scale seismic events to evaluate their performance and ability to detect subtle ionospheric disturbances. Such an approach will provide insights into the limitations and adaptability of these algorithms when applied to less pronounced events. Key improvements include incorporating additional features, optimizing computational efficiency, and validating models across different data sets. A database of seismic and tsunami-related ionospheric disturbances would further enhance continuous learning and performance, contributing to more robust, integrated early warning systems.
In conclusion, this research emphasizes the crucial role of ionospheric TEC data monitoring in improving TEWS, advancing our understanding of small-scale event-induced ionospheric perturbations and supporting AI-driven methods for automatic detection.