Jozinović, Dario
Preferred name
Jozinović, Dario
Email
dario.jozinovic@ingv.it
ORCID
16 results
Filters
Settings
The search returned 16 results
Now showing 1 - 10 of 16
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Masked graph neural network for rapid ground motion prediction in Italy
([Montréal]: McGill University Library, [2022]-, 2025-09-15) ;Trappolini, Danele; ; ; This study presents an updated version of TISER-GCN, a graph neural network (GCN) designed to predict maximum intensity measurements (IMs) from 10-second seismic waveforms starting at the earthquake origin time, without prior knowledge of location, distance, and magnitude. The improved model was applied to nearly 600 seismic stations from the INSTANCE benchmark dataset, significantly expanding the original TISER-GCN setup, which was limited to 39 stations in a smaller area of central Italy. Input data consist of three-component waveforms selected to ensure high quality and minimize saturation. Results show that masking stations where the P-wave arrives within the first 10 seconds , combined with the integration of additional information, reduces the mean squared error (MSE) by up to 6% for peak ground acceleration (PGA) and 5.5% for peak ground velocity (PGV), compared to the unmasked baseline. Moreover, the proposed approach yields near-zero median residuals across all IMs, mitigating the systematic underestimation observed when using a ground motion model specifically developed for Italy. These findings indicate that the model provides accurate predictions of ground motions, comparable to those obtained with the original TISER-GCN, which, however, requires a fixed seismic network geometry.8 36 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Effects on a Deep-Learning, Seismic Arrival-Time Picker of Domain-Knowledge Based Preprocessing of Input Seismograms
(2024-06-20); ; ; ; Automated seismic arrival picking on large and real-time seismological waveform datasets is fundamental for monitoring and research. Recent, high-performance arrival pickers apply deep-neural-networks to nearly raw seismogram inputs. However, there is a long history of rule-based, automated arrival detection and picking methods that efficiently exploit variations in amplitude, frequency and polarization of seismograms. Here we use this seismological domain-knowledge to transform raw seismograms as input to a deep-learning picker. We preprocess 3-component seismograms into 3-component characteristic functions of a multi-band picker, plus modulus and inclination. We use these five time-series as input instead of raw seismograms to extend the deep-neural-network picker PhaseNet. We compare the original, data-driven PhaseNet and our domain-knowledge PhaseNet (DKPN) after identical training on datasets of different sizes and application to in- and cross-domain test datasets. We find DKPN and PhaseNet show near identical picking performance for in-domain picking, while DKPN outperforms PhaseNet for some cases of cross-domain picking, particularly with smaller training datasets; additionally, DKPN trains faster than PhaseNet. These results show that while the neural-network architecture underlying PhaseNet is remarkably robust with respect to transformations of the input data (e.g. DKPN preprocessing), use of domain-knowledge input can improve picker performance.43 38 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Graph neural networks for multivariate time series regression with application to seismic data
Machine learning, with its advances in deep learning has shown great potential in analyzing time series. In many scenarios, however, additional information that can potentially improve the predictions is available. This is crucial for data that arise from e. g., sensor networks that contain information about sensor locations. Then, such spatial information can be exploited by modeling it via graph structures, along with the sequential (time series) information. Recent advances in adapting deep learning to graphs have shown potential in various tasks. However, these methods have not been adapted for time series tasks to a great extent. Most attempts have essentially consolidated around time series forecasting with small sequence lengths. Generally, these architectures are not well suited for regression or classification tasks where the value to be predicted is not strictly depending on the most recent values, but rather on the whole length of the time series. We propose TISER-GCN, a novel graph neural network architecture for processing, in particular, these long time series in a multivariate regression task. Our proposed model is tested on two seismic datasets containing earthquake waveforms, where the goal is to predict maximum intensity measurements of ground shaking at each seismic station. Our findings demonstrate promising results of our approach—with an average MSE reduction of 16.3%—compared to the best performing baselines. In addition, our approach matches the baseline scores by needing only half the input size. The results are discussed in depth with an additional ablation study.98 89 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Rapid prediction of ground shaking intensity with Graph Neural Networks
Rapid accurate prediction of strong ground shaking can be crucial for earthquake early warning. Recently, machine learning (ML), with its advances in Deep Learning (DL), has shown great potential in analysing seismic waveforms. More specifically, when using the data acquired by a seismic network, the incorporation of additional information consisting of the network station positioning into the DL model has been found beneficial to improve the accuracy of the ground motion predictions (Jozinović et al., 2022). Such spatial information can be exploited thoroughly by adopting graph structures, along with the seismic waveforms. Recent advances in adapting DL to graphs have shown promising potential in various graph- related tasks. However, these methods have not been completely adapted for seismological tasks. In this work, we advance an architecture capable of processing a set of seismic time series acquired by a network of stations using the benefits of Graph Neural Networks (GNNs) (see Fig. 1). The objective of the study is the rapid determination of the ground motion (PGA, PGV, and SA 0.3s, 1s and 3s) at farther stations that have not been yet reached by the strong ground shaking by availing of the first signals recorded at the stations close to the epicentre. The work builds upon the GNN approach proposed in Bloemheuvel et al. (2022) and incorporates transfer learning, see Jozinović et al. (2022). We apply the methodology to two datasets having very different source-receiver geometries sited in central Italy (CI, Jozinović et al., 2020, Jozinović et al., 2022) and in north-western central Italy (CW), respectively (Fig. 2). The two datasets have already been the object of similar studies using convolutional neural networks which serve as baselines for comparison. We find that the GNNs are highly suited for the analysis of seismic data from a set of stations and show improvement when compared to the previous work (Bloemheuvel et al., 2022 and Jozinović et al., 2022). We exemplify the early warning capabilities of the proposed approach.173 399 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , ShakeMap4‐Web: Web Application for ShakeMap 4 Product Visualization
(2022-05-10); ; ; ; We present a web portal for the prompt visualization of the maps of ground shaking generated using the USGS ShakeMap 4 software (Worden et al., 2020). The web interface renders the standard products provided by ShakeMap dynamically (using Leaflet) and statically (standard shakemaps). The information included in the dynamic maps can be onfigured through different overlays. The dual view rendering modality allows presenting side-by-side maps of different intensity measurements. In addition, for each earthquake, it is possible to download all the data that contributed to the calculation, together with information on the seismological models adopted. The appearance of the web portal is easily configurable by replacing the logo and banners. The software can be installed both on laptops and on server computers. The user can opt between the docker image or installation of the software after installation of a web server (e.g., NGINX or Apache).211 197 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Restricted , SeisBench—A Toolbox for Machine Learning in Seismology
(2022-03-16); ; ; ; Machine‐learning (ML) methods have seen widespread adoption in seismology in recent years. The ability of these techniques to efficiently infer the statistical properties of large datasets often provides significant improvements over traditional techniques when the number of data are large (millions of examples). With the entire spectrum of seismological tasks, for example, seismic picking and detection, magnitude and source property estimation, ground‐motion prediction, hypocenter determination, among others, now incorporating ML approaches, numerous models are emerging as these techniques are further adopted within seismology. To evaluate these algorithms, quality‐controlled benchmark datasets that contain representative class distributions are vital. In addition to this, models require implementation through a common framework to facilitate comparison. Accessing these various benchmark datasets for training and implementing the standardization of models is currently a time‐consuming process, hindering further advancement of ML techniques within seismology. These development bottlenecks also affect “practitioners” seeking to deploy the latest models on seismic data, without having to necessarily learn entirely new ML frameworks to perform this task. We present SeisBench as a software package to tackle these issues. SeisBench is an open‐source framework for deploying ML in seismology—available via GitHub. SeisBench standardizes access to both models and datasets, while also providing a range of common processing and data augmentation operations through the API. Through SeisBench, users can access several seismological ML models and benchmark datasets available in the literature via a single interface. SeisBench is built to be extensible, with community involvement encouraged to expand the package. Having such frameworks available for accessing leading ML models forms an essential tool for seismologists seeking to iterate and apply the next generation of ML techniques to seismic data.175 351 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Intra-domain and cross-domain transfer learning for time series data -- How transferable are the features?
(2022-03-05); ; ; ; In practice, it is very demanding and sometimes impossible to collect datasets of tagged data large enough to successfully train a machine learning model, and one possible solution to this problem is transfer learning. This study aims to assess how transferable are the features between different domains of time series data and under which conditions. The effects of transfer learning are observed in terms of predictive performance of the models and their convergence rate during training. In our experiment, we use reduced data sets of 1,500 and 9,000 data instances to mimic real world conditions. Using the same scaled-down datasets, we trained two sets of machine learning models: those that were trained with transfer learning and those that were trained from scratch. Four machine learning models were used for the experiment. Transfer of knowledge was performed within the same domain of application (seismology), as well as between mutually different domains of application (seismology, speech, medicine, finance). We observe the predictive performance of the models and the convergence rate during the training. In order to confirm the validity of the obtained results, we repeated the experiments seven times and applied statistical tests to confirm the significance of the results. The general conclusion of our study is that transfer learning is very likely to either increase or not negatively affect the predictive performance of the model or its convergence rate. The collected data is analysed in more details to determine which source and target domains are compatible for transfer of knowledge. We also analyse the effect of target dataset size and the selection of model and its hyperparameters on the effects of transfer learning.82 27 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Which Picker Fits My Data? A Quantitative Evaluation of Deep Learning Based Seismic Pickers
(2022-01-06); ; ; ; Seismic event detection and phase picking are the base of many seismological workflows. In recent years, several publications demonstrated that deep learning approaches significantly outperform classical approaches, achieving human-like performance under certain circumstances. However, as studies differ in the datasets and evaluation tasks, it is unclear how the different approaches compare to each other. Furthermore, there are no systematic studies about model performance in cross-domain scenarios, that is, when applied to data with different characteristics. Here, we address these questions by conducting a large-scale benchmark. We compare six previously published deep learning models on eight data sets covering local to teleseismic distances and on three tasks: event detection, phase identification and onset time picking. Furthermore, we compare the results to a classical Baer-Kradolfer picker. Overall, we observe the best performance for EQTransformer, GPD and PhaseNet, with a small advantage for EQTransformer on teleseismic data. Furthermore, we conduct a cross-domain study, analyzing model performance on data sets they were not trained on. We show that trained models can be transferred between regions with only mild performance degradation, but models trained on regional data do not transfer well to teleseismic data. As deep learning for detection and picking is a rapidly evolving field, we ensured extensibility of our benchmark by building our code on standardized frameworks and making it openly accessible. This allows model developers to easily evaluate new models or performance on new data sets. Furthermore, we make all trained models available through the SeisBench framework, giving end-users an easy way to apply these models.227 143 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , The Choice of Time–Frequency Representations of Non-Stationary Signals Affects Machine Learning Model Accuracy: A Case Study on Earthquake Detection from LEN-DB Data
(2022-01); ; ; ; Non-stationary signals are often analyzed using raw waveform data or spectrograms of those data; however, the possibility of alternative time–frequency representations being more informative than the original data or spectrograms is yet to be investigated. This paper tested whether alternative time–frequency representations could be more informative for machine learning classification of seismological data. The mentioned hypothesis was evaluated by training three well-established convolutional neural networks using nine time–frequency representations. The results were compared to the base model, which was trained on the raw waveform data. The signals that were used in the experiment are three-component seismogram instances from the Local Earthquakes and Noise DataBase (LEN-DB). The results demonstrate that Pseudo Wigner–Ville and Wigner–Ville time–frequency representations yield significantly better results than the base model, while spectrogram and Margenau–Hill perform significantly worse (p < 0.01). Interestingly, the spectrogram, which is often used in signal analysis, had inferior performance when compared to the base model. The findings presented in this research could have notable impacts in the fields of geophysics and seismology as the phenomena that were previously hidden in the seismic noise are now more easily identified. Furthermore, the results indicate that applying Pseudo Wigner–Ville or Wigner–Ville time–frequency representations could result in a large increase in earthquakes in the catalogs and lessen the need to add new stations with an overall reduction in the costs. Finally, the proposed approach of extracting valuable information through time–frequency representations could be applied in other domains as well, such as electroencephalogram and electrocardiogram signal analysis, speech recognition, gravitational waves investigation, and so on.83 68 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: Open Access , Transfer learning: Improving neural network based prediction of earthquake ground shaking for an area with insufficient training data
In a recent study (Jozinovi\'c et al, 2020) we showed that convolutional neural networks (CNNs) applied to network seismic traces can be used for rapid prediction of earthquake peak ground motion intensity measures (IMs) at distant stations using only recordings from stations near the epicenter. The predictions are made without any previous knowledge concerning the earthquake location and magnitude. This approach differs from the standard procedure adopted by earthquake early warning systems (EEWSs) that rely on location and magnitude information. In the previous study, we used 10 s, raw, multistation waveforms for the 2016 earthquake sequence in central Italy for 915 events (CI dataset). The CI dataset has a large number of spatially concentrated earthquakes and a dense station network. In this work, we applied the CNN model to an area around the VIRGO gravitational waves observatory sited near Pisa, Italy. In our initial application of the technique, we used a dataset consisting of 266 earthquakes recorded by 39 stations. We found that the CNN model trained using this smaller dataset performed worse compared to the results presented in the original study by Jozinovi\'c et al. (2020). To counter the lack of data, we adopted transfer learning (TL) using two approaches: first, by using a pre-trained model built on the CI dataset and, next, by using a pre-trained model built on a different (seismological) problem that has a larger dataset available for training. We show that the use of TL improves the results in terms of outliers, bias, and variability of the residuals between predicted and true IMs values. We also demonstrate that adding knowledge of station positions as an additional layer in the neural network improves the results. The possible use for EEW is demonstrated by the times for the warnings that would be received at the station PII.117 124
