Tracking Ocean Storms Through Seismic Noise

Ocean storms are the main source of secondary microseisms. These signals result from pressure fluctuations on the sea floor caused by standing waves generated by storm systems.

Because seismic records extend far before the satellite era, they provide a powerful way to study historical patterns and intensities of ocean storms. Understanding how these storm systems have changed over time is increasingly vital in the context of ongoing climate change.

The animation on the right shows ocean wave heights alongside the power of secondary microseisms. The arrow indicates the direction of the maximum gradient of microseismic energy, with its length scaled to the gradient’s magnitude. This visualization clearly demonstrates how ocean storms influence secondary microseismic noise. To learn more, please refer to our publication.

s1.avi

Sea Ice Influence on Microseismic Activity

Arctic sea ice has been steadily declining since satellite observations began in 1979. Historical reconstructions and paleoclimate records show that the recent summer loss of Arctic sea ice is unprecedented in at least the past 1,000 years. This rapid decline threatens Arctic ecosystems and communities while also creating a powerful climate feedback loop—as diminishing ice reduces Earth’s reflectivity and further accelerates global warming.

Sea ice coverage in the Arctic Ocean changes seasonally—reaching its maximum extent around March and its minimum near September. As sea ice forms, it suppresses ocean wave activity, which in turn reduces the generation of microseisms. This effect is most pronounced in the short-period secondary microseismic (SPSM) band.

The animation on the right shows the evolution of SPSM power recorded at station A21K, along with sea ice coverage and wave height in northern Alaska. The data reveal that SPSM amplitudes closely track ocean wave amplitudes in coastal waters. To learn more, please refer to our publication.

s3.avi

Glacier Quakes and Glacier Dynamics

The Alaska Earthquake Center has recorded glacier quakes across southern Alaska for decades. These events typically cluster near the termini of major tidewater glaciers such as Columbia, Yahtse, Hubbard, La Perouse, and Bering. Among them, Columbia Glacier stands out as one of the world’s fastest retreating glaciers. Its highly dynamic terminus, long history of observation, and wealth of complementary datasets make it an ideal site for studying glacier seismicity and dynamics.

Large calving events at a glacier’s terminus create and collapse cavities in the water, releasing substantial energy that generates glacial quakes (Bartholomaus et al., 2012). The figure on the right shows annually relocated glacial quakes at Columbia Glacier using the hypoDD algorithm. The spatial migration of these events follows the retreat of the glacier terminus, supporting their origin from calving processes.

The second panel of this figure shows the fjord depth with the terminus position of Columbia Glacier. Before 2010, when the terminus was in deep water (>200 m), seismic activity was absent. As it retreated into shallower water (<100 m), quake rates rose sharply, appearing abruptly around 2010. We find that bathymetry is a strong control on calving-related seismicity: calving in shallow, grounded conditions is dominated by frequent serac failures that generate strong seismic signals through cavity collapse in the water. On the contrary, in deep water, near-floating or floating termini produce minimal seismic energy because calving ice is supported by water (panels 3 and 4).

We also find that glacier velocity, thickness, precipitation, and sea surface temperature are other factors that influence the style and rate of calving. Curious to know how all these factors shape the recorded seismicity? Stay tuned for our upcoming publication...

Resonance of microseisms due to Bathymetry

The bathymetry of the ocean plays a vital role in resonating secondary microseisms. The animation on the right shows the evolution of a noticeable storm in the Gulf of Alaska and the corresponding secondary microseisms recorded in the northeast region. We can clearly see that the microseismic amplitude jumps when the storm's arm strikes the eastern coast of Alaska. Kedar et al. 2008 show that the bathymetry of the Gulf of Alaska, especially the eastern Gulf, is conducive to resonating secondary microseisms.

Additionally, the coastline of Alaska is roughly perpendicular to the incoming storms here. This could result in ocean waves reflecting towards the incoming storm. Such settings are known to excite microseisms. Whatever the cause, it is clear that the Gulf of Alaska produces secondary microseism efficiently, dominating the noise floor observed across Alaska.

John_S2.mp4

Precursory signals from Tracy Arm Landslide

On August 10, 2025, a massive landslide (estimated volume of 64 million cubic meters) occurred in the Tracy Arm Fjord. This landslide triggered a megatsunami with waves as high as 100 m, traveling at 150 miles/hour. The National Park Service reported a run-up of at least 100 feet at Sawyer Island, where vegetation was stripped from the slopes.

This event was scientifically extremely valuable, as many hour-long precursory signals preceded it. The video on the right shows the sonification of these signals by UAF researchers. Currently, I am analyzing these precursory signals to determine what they are and whether we can use them to predict destructive landslides in the future. The figure below shows the stack of detections for different clusters recorded in station S32K. The similarity among these waveforms suggests that, to the first order, events are triggered by the same source mechanism.

Analyzing the Performance of GaMMA

The field of earthquake processing is rapidly evolving with the advent of AI. Modern phase-picking algorithms, such as PhaseNet, are emerging as alternatives to traditional methods like STA/LTA, and new association algorithms, such as GaMMA, are challenging conventional workflows. However, a key question remains: do these new algorithms consistently outperform traditional approaches? To investigate this, during my internship at ISTI, I ran the PhaseNet algorithm on seismic data and passed the resulting picks to GaMMA for association. I also tested STA/LTA-generated picks with GaMMA and compared the results to the real-time association algorithm in Earthworm, called Binder. I also used HYPOINVERSE to locate these events and compared them with the reference catalog.

Testing across multiple sites revealed several important insights. PhaseNet picks generally outperformed STA/LTA picks when processed with GaMMA. Incorporating amplitude information typically improved results, although its impact decreased for smaller array sizes. GaMMA detected more real events, demonstrating its value as a complementary pipeline. At some sites, GaMMA outperformed Binder, but it often resulted in more false detections.

Source: https://github.com/AI4EPS/GaMMA

1D-Velocity modeling of Alaska

Alaska has extremely complex geology and spans a vast geographic area. The Alaska Earthquake Center is currently transitioning to SeisComP, and I have been actively involved in this effort. A key component of this transition is developing a refined 1D velocity model for Alaska. I am currently working extensively on this task using VELEST.

The figure on the right shows the distribution of earthquakes and seismic stations used to derive the velocity model. This is an ongoing project, and our initial model shows a systematic geographic trend in residuals. This could be due to Alaska's geological complexity. We aim to address this in upcoming iterations by deriving different models for different parts of Alaska.

Random Forest regression of Short-Period Microseism.

John & West (2025) demonstrated that coastal ocean wave activity is the primary driver of SPSM energy. Additionally, these signals are also heavily modulated by sea ice.

This improved understanding of SPSM source processes enables data-driven modeling of SPSM variability. In this study, we employ a Random Forest regression model using significant wave height and sea-ice concentration as predictors to estimate SPSM power. The model achieves a mean absolute error of < 2.7 dB. Feature-importance analysis reveals that only features from coastal sea state contribute to the prediction, consistent with our published findings.

The figure on the right compares the observed SPSM power with the model predictions.

Page updated

Google Sites

Report abuse