In addition, this release restructures the documentation and brings many smaller feature additions and bugfixes.
30.01.2026 16:17
👍 0
🔁 0
💬 0
📌 0
In addition, this release restructures the documentation and brings many smaller feature additions and bugfixes.
With two new datasets, one for induced seismicity and one for swarms in the Bohemia region (Germany/Czech Republic), SeisBench expands its benchmark datasets towards smaller magnitude events and near-field records.
The EQTP model allows for jointly picking phase arrival times and polarities.
The annotate/classify interface for processing seismic waveforms has been refactored and optimized, implementing key functions in C. This makes model application faster and reduces the memory footprint.
An DAS record with P and S phase picks. Both pick curves are overall good, even though some P picks that can be visually identified are missing from the results.
SeisBench now has support for DAS benchmark datasets and DAS models (built on Xdas). While native DAS models are yet to come, we already provide an efficient wrapper to apply classical 3C models to DAS. Below an examples of using PhaseNet trained on STEAD on DAS data.
SeisBench v0.11 is out now, the largest feature and performance update to SeisBench so far! In this thread, I'll highlight some of the key contributions. Thanks to everyone who helped putting together this release!
github.com/seisbench/se...
A new preprint, led by Bar Oryan, introduces the Coupling Cloud: a community database that curates, standardizes, documents, and disseminates published kinematic coupling models inverted from geodetic data. As of today it brings together 96 coupling datasets drawn from 55 publications spanning
Map of the central Apennines area indicated by the blue rectangle in the inset. Seismic stations used in the analysis are shown as triangles coloured according to network codes. Purple dots are the 84 991 earthquakes detected and located by BSI in the time period 2016 August 15–2017 August 31. Yellow stars are the three main shocks of the central Apennines seismic sequence.
Histograms of P (a, c) and S (b, d) pick difference for PhaseNet Original (PN-OR) and PhaseNet INSTANCE (PN-IN) with respect to INGV manual phase picks (BSI). Panels (e) and (f) show the histograms of P and S pick time differences for PN-OR with respect to PN-IN. Medians and Means are represented as red dashed and blue solid lines, respectively; ±1 and ±2 standard deviations are in dashed green and purple lines, respectively.
New paper! We compare PhaseNet (PN) #deep-learning (DL) with manual (BSI) event picking for routine seismic monitoring by analyzing high-precision earthquake locations and tomographic inversion for the 2016 Central Italy #earthquake sequence. doi.org/10.1093/gji/... @carlo-giunchi.bsky.social
📈 Which seismic phase associator works best?
🧠Puente Huerta et al., benchmarked 5 algorithms testing their performance across diverse settings and under noisy-complex conditions
Read more: seismica.library.mcgill.ca/article/view...
------
@jannesmunch.bsky.social
Our latest study in Nature, together with colleagues from @geomarkiel.bsky.social and universities from Greece, shows what happened beneath Santorini earlier this year when tens of thousands earthquakes shook the islands and the seafloor. It was magma. We also have 1/2
www.gfz.de/en/press/new...
This is one of the most interesting sequences I've worked on so far! It's great to see how machine learning can provide us with these detailed images of seismicity in near real time. I'm glad to have been able to contribute to this project.
Introducing Xdas — a Python library designed to manipulate distributed acoustic sensing (DAS) data. #SRL ⚒️
pubs.geoscienceworld.org/ssa/srl/arti...
Congratulations on defending! That cake is truly outstanding! I never tried phase picking on my switch so far 😅
It should clearly read v0.10...
SeisBench v.10 is out and it's time for some new models:
- SkyNet allows picking regional phases and can even distinguish Pn, Pg, Sn, Sg (beware of the Terminator though)
- SeisDAE brings easily retrainable seismic waveform denoising
Check out all changes here:
github.com/seisbench/se...
In addition, there are many smaller tweaks (pick export to Pandas, PhaseNetWC, ...). Thanks to all contributors of this release.
You're working on machine learning for seismology as well? SeisBench is always looking for new contributions!
SeisBench v0.9 is out now. Some highlights of the new version:
- The CWA dataset with broadband and strong motion data from Taiwan.
- The CEED dataset covering more than 4 million waveforms from California.
- A 20% speed-up when picking continuous datasets.
github.com/seisbench/se...
The migrations hint at the processes causing SSE initiation on the plate interface. Looking back in time, we identify numerous seismic swarms exactly colocated with the 2023 swarm, suggesting the existence of recurring SSEs in the area. (2/2)
Overview figure of the 2023 SSE with three subplots. Left: The SSE slip model with seismicity overlain. Top right: zoom in on the onset sequence in map view. Bottom right: Cross-section of the onset sequence. In all panels, events are colored by time, revealing clear migtations.
Have you ever wondered how seismic swarms around slow slip events (SSEs) might work? In our recent work, we study the complex seismic migrations around the onset of a shallow SSEs in the Copiapó ridge in 2023. (1/2)
agupubs.onlinelibrary.wiley.com/doi/10.1029/...
A green and sunny landscape in southern Germany near the Austrian border photographed from a train.
Looking forward to a week full of science and networking at #EGU25! Join my talk on Thursday afternoon if you want to learn about the seismicity in Northern Chile and how swarms reveal the mechanisms of slow slip.
#Train2EGU
A flyer for the AI 4 Seismology International Training School, listing the data (05. to 08. May 2025) and location (Leipzig).
The registration for our International Training School for AI in Seismology is open now! The school combines lectures, poster sessions, practical tutorials, and networking opportunities.
We're looking forward to welcoming you in Leipzig in May!
This animation shows an example of how QSeek helps understanding the ongoing Santorini sequence in real-time: bsky.app/profile/gfz....
Our paper on QSeek is out in Seismica now. QSeek detects earthquakes and locates them precisely by backprojecting deep learning phase probability curves in time and space. And that at a blazingly fast speed! Thanks Marius Isken for this great work!
Just to clarify, I didn't make this animation. I made a similar one a few days ago: bsky.app/profile/jann...
Nonetheless, I fully agree that we need solid science funding to enable understanding hazards like this!
The main DEI page from @agu.org has been modified to take diversity out of the title. Glad to see it's not deleted entirely, but hey AGU, we're watching what you do and remembering the big game you once talked about supporting your diverse membership.
Today vs. January
Thanks! I've personally left Twitter, but you can share the material there. Please include the link back to the original Bluesky post.
Earthquake activity over distance along the transect. Both the large-scale migration and fine-scale burst migrations in both directions are visible.
Here you go! I picked a few of the burst migration speeds by hand and they are in the range of 2 km/h.
Hi Seismo-Bluesky! I decided to test a pipeline with SeisBench, PyOcto, NonLinLoc, and HypoDD on the Santorini sequence. The level of activity (>4,500 events) is truly stunning! There are repeated bursts migrating backwards and forwards, spreading outwards from a narrow, almost linear channel.
Advertisement for the AI 4 Seismology training school, showing a seismogram, the list of organisers, and the two confirmed speakers Tarje Nissen-Meyer and René Steinmann
🤖 Ready to use artificial intelligence to revolutionize seismology? Then join us at the International Training School "AI 4 Seismology" on May 5-8, 2025 at @scadsai.bsky.social in Leipzig.
So far confirmed trainers:
- Tarje Nissen-Meyer (@exeter.ac.uk )
- René Steinmann (@gfz.bsky.social )
OTD twenty years ago the 3rd strongest earthquake ever recorded by seismometers struck offshore the island of Sumatra. A tsunami followed killing more than 200,000 people. In the aftermath we helped develop a tsunami warning system for the region. Read more here: www.gfz-potsdam.de/en/press/new...
