A massive 8.8 magnitude earthquake, one of the strongest ever recorded, struck off Russia’s Far East on Wednesday. This powerful quake generated tsunami waves that reached Japan, Hawaii, and the U.S. West Coast.

Tsunami in Japan and Russia (latest news)

While no major damage has been reported so far, authorities urged people to stay away from shorelines, emphasizing that the danger could persist for over a day. In Russia’s Kamchatka Peninsula, near the epicenter, ports experienced flooding, and residents evacuated inland. Japan’s northern coast saw frothy waves, and in Hawaii’s capital, traffic became gridlocked as people moved away from the coast.

Many in Japan sought refuge in evacuation centers, with the devastating 2011 earthquake and tsunami still fresh in their minds, though no issues were reported at nuclear plants this time. Some injuries were reported in Russia and at least one in Japan. Tsunami wave heights varied, with Kamchatka recording 3-4 meters, Japan’s Hokkaido 60 centimeters, and Alaska’s Aleutian Islands up to 1.4 feet above tide levels.

Hours after the quake, both Hawaii and parts of Japan downgraded their tsunami warnings to advisories, though higher alerts remained in effect for some northern Japanese coastal areas.

AI in Tsunami and Weather Disaster Forecasting: Analysis and Perspectives

Need for AI in Forecasting Disasters

Traditional forecasting methods for tsunamis and weather hazards have critical limitations. Tsunami warning systems often rely on fixed seismic thresholds and simplified wave models, which can lead to delayed alerts or false alarms. Similarly, global weather models are extremely computationally intensive and use approximations (e.g. coarse grid physics) that limit accuracy and lead time. AI offers a way to leverage vast, diverse datasets to improve both speed and accuracy. For example, large AI models (“foundation” models) can analyze satellite, radar, and sensor data at high resolution with far less computation. By learning complex patterns that simple rules or traditional simulations miss, AI can extend lead times and reduce uncertainty in forecasts. These advantages are crucial as climate change is driving more frequent and severe storms, floods and tsunamis, straining existing early-warning systems.

AI Approaches and Methodologies

Researchers employ a variety of AI techniques to predict tsunamis and weather disasters. Key methods include:

• Machine Learning (ML): Algorithms such as random forests, decision trees, and support vector machines are trained on historical disaster and environmental data (e.g. past earthquakes, sea levels, weather records) to estimate risk. In tsunami research, for instance, a recent study showed a Random Forest classifier (accuracy ~90%) could distinguish tsunami-generating quakes better than traditional logistic regression. ML models excel at pattern recognition in structured datasets and can be tuned for imbalanced events (e.g. rare large quakes).

• Deep Learning (DL): Neural networks (CNNs, RNNs/LSTMs, Transformers) handle high-dimensional inputs like satellite imagery, time-series sensor streams, and 3D ocean data. Convolutional Neural Networks can detect storm structures in cloud imagery, while recurrent networks or Transformers can model temporal dynamics of atmospheric pressure or river flows. For example, DL “emulators” (purely data-driven models) have recently outperformed traditional forecasts: Google DeepMind’s GraphCast model predicted Hurricane Lee’s track 9 days in advance – three days earlier than operational models. Similarly, Microsoft’s large-scale “Aurora” model was trained on >1 million hours of weather data and fine-tuned to storm forecasting, achieving significantly better 5–10 day forecasts at much lower cost.

• Hybrid/Physics-Informed Models: These combine AI with established geophysical models. For instance, ML algorithms may be constrained by physical laws (e.g. water wave equations) or used to augment numerical simulators. Knowledge-guided ML embeds domain expertise into training, ensuring outputs respect physics. Such hybrid models can correct or speed up traditional simulations. Studies advocate Hybrid AI/Mechanistic frameworks – e.g. using AI to learn residual errors of a simulator or using model outputs as ML training features – to improve robustness and interpretability.

• Data Fusion and Foundation Models: AI excels at merging heterogeneous data sources. Tsunami forecasting efforts stress integrating seismic data, ocean-buoy measurements, geospatial maps and even weather conditions into a single model. Modern “foundation” models (large pre-trained AI networks) can ingest multi-modal inputs (satellite imagery, IoT sensors, text reports) and be fine-tuned for specific tasks. For example, IBM and NASA are developing an open climate foundation model to unify diverse Earth-system data. The WMO also highlights integrating meteorological and geospatial AI models for multi-hazard early warnings.

• Real-Time and Streaming Analytics: AI algorithms can process live data streams (e.g. real-time seismic feeds, river gauge sensors, or social media alerts) to update forecasts on the fly. Although not all studies explicitly cover this, it is an active area: e.g. ML-based early-warning systems analyze incoming ocean-sensor and buoy data to immediately assess tsunami risk. By contrast, traditional methods often require batch simulation. Streaming ML enables continuous hazard monitoring and quicker response.

Combined, these techniques enable AI systems to handle the complexity of disaster prediction far beyond static statistical models. The figure below illustrates some AI approaches in disaster forecasting.

Case Studies and Successes

AI-based systems have already shown remarkable performance in predicting major weather events (Table 1 summarizes examples). Notable case studies include:

• Hurricane and Typhoon Forecasting: Deep neural networks are improving cyclone tracking. Google DeepMind’s GraphCast accurately predicted Hurricane Lee’s landfall in Nova Scotia nine days before the event, outperforming traditional physics-based forecasts by ~3 days. More recently, Microsoft’s Aurora foundation model (trained on global weather/ocean data) correctly forecast Typhoon Doksuri’s track in the Philippines four days in advance, whereas official models mislocated it. In testing, Aurora outperformed all seven major forecasting centers in 5-day cyclone track accuracy for the 2022–23 season. Similarly, Google and NOAA have partnered on AI models: a DeepMind system’s 5-day hurricane forecasts were on average ~140 km closer to the true path than the European ensemble baseline, and it also improved rapid intensity change forecasts. These AI improvements promise earlier, more precise storm warnings and larger lead times for evacuation.

• Flood Forecasting: River flood prediction has greatly benefited from AI. Google Research demonstrated that AI (notably LSTM networks) can extend accurate flood forecasts up to 7 days ahead, even in data-poor regions. Their model is now operational in the Flood Hub, covering 460 million people in 80 countries. Flood forecasts once unavailable in much of the developing world are now possible thanks to ML-driven global models and public alerts. In pilot studies (e.g. India’s Ganges-Brahmaputra basin), AI-derived forecasts significantly improved on physics-only models, enabling timely warnings.

• Tsunami Early Warnings: Tsunamis remain very challenging to predict, but AI is beginning to contribute. For example, Western University (Canada) used machine learning to optimize tsunami early-warning timing for the town of Tofino, B.C.. In their analysis, a Random Forest model (requiring seismic and oceanographic inputs) outperformed traditional multi-linear regression (the standard approach) in minimizing false alarms and timely evacuations. While no AI system has yet predicted an actual tsunami event beforehand (data scarcity is a major hurdle), these studies show ML can refine alert thresholds and fusion of local data to improve warnings. Researchers emphasize that more ocean-bottom sensors (Japan has 150 vs. only 4 around Vancouver Island) would dramatically enhance any AI/physics forecast system.

Other AI successes include extreme weather events like air quality and storms. For instance, the Aurora model predicted a severe Baghdad dust storm one day early, and has achieved state-of-the-art results on global precipitation and wave forecasts as well. In sum, documented cases span cyclones, floods, and tsunamis – demonstrating AI’s growing role in early-warning systems.

Table 1. Summary of AI forecasting successes: notable disaster events, AI approach, and key outcomes.

Approach to tsunami forecasting

Input data

New research in Physics of Fluids shows the possibility of getting real-time earthquake information from underwater sensors to add more information to risk calculations. The researchers tested how hydrophones can pick up the unique acoustic radiation — that is, sound — from earthquakes. 

Monitoring systems on seas:

Deep-Ocean Assessment and Reporting of Tsunamis buoys are stationed all around the world, but especially in the seismically active Pacific Ocean. These sensors are real-time tsunami monitoring systems and play a critical role in tsunami forecasting. (Image courtesy of NOAA National Data Buoy Center)

Land Sensors:

Sensors can be on land (seismometers), circling the globe (global navigation satellite system), or on the ocean (Deep-Ocean Assessment and Reporting of Tsunamis buoys). The DART buoy array is a set of floating sensors placed around the world but is heavily concentrated within the seismically active Pacific Ocean region.

Moore explains that while DART gives highly accurate information, the sensors are expensive to place and maintain and as such are only placed in the most critical regions. “We’ve done our best to optimize DART buoy placement, according to not only where earthquakes occur, but where people live,” he says. But the distance of the buoys from some earthquake epicenters means there can be a lag in new information of 10-90 minutes after an earthquake.

“We’re continually trying to find new sensors to slot in between the time where we have seismic information and when we have that highly accurate DART buoy information,” says Moore. This can include GNSS information, disruptions in communication cables, or even hydrophones.

Sound Hydrophones:

Hydrophones are placed on the ocean floor and can detect sounds from all sorts of activities, from underwater eruptions to bomb testing to earthquakes. These sound waves, also called acoustic-gravity waves, travel at the speed of sound (1,500 m per second in seawater) and can often play off the ocean floor, doubling their speed.

“Sound carries information about the originating source, and its pressure field can be recorded at distant locations, even thousands of kilometers away from the source,” says Gomez. “They carry information about the earthquake’s characteristics and can be recorded by hydrophones in the far field.”

Artificial-intelligence-informed models 

“Artificial intelligence can play a prominent role in the classification of earthquake types,” Gomez says. “In combination with state-of-the-art acoustic-gravity wave technology, we can have a more reliable real-time tsunami warning system.” 

“This work is part of a larger project aiming at enhancing warning systems from natural hazards,” says Gomez. “The nature of the developed technology is complementary, and as such we look forward to collaborating and complementing experts on further enhancing the system.”

The method has the potential to be an innovative tool in tsunami forecasting, say Wei and Moore.

“We have hope that AI will lend itself to solve some of the problems we have in characterizing a tsunami generated from something like a landslide or a volcano,” says Moore. Because these natural events do not have associated earthquakes, he says it is harder to characterize the location and size of a tsunami that might be generated.

Hydrophones and AI modeling could be useful tools in the development of tsunami warning systems. “We all know that new technology needs time to evolve,” Wei says. “I think this technology has a great potential to be implemented in the future system.”

Approach to tsunami forecasting

Current tsunami forecasting relies on a network of sensors, including land-based seismometers, satellite navigation systems (GNSS), and Deep-Ocean Assessment and Reporting of Tsunamis (DART) buoys. While DART buoys provide highly accurate data, they are expensive and thus sparsely located, often leading to a 10-90 minute delay in receiving information after an earthquake.

New research suggests that underwater hydrophones could significantly enhance real-time tsunami warnings. These sensors, placed on the ocean floor, can detect the unique acoustic signals from earthquakes, as sound travels quickly through water. This acoustic data, which carries information about the earthquake’s characteristics, could help bridge the gap between initial seismic detection and the more precise DART buoy information.

Furthermore, artificial intelligence (AI) is emerging as a powerful tool in this field. AI can help classify earthquake types and, when combined with hydrophone technology, could lead to a more reliable real-time tsunami warning system. Researchers are optimistic that AI and hydrophones could also help forecast tsunamis caused by events without associated earthquakes, like landslides or volcanic eruptions, which are currently harder to characterize. This integrated approach of using hydrophones and AI models holds great promise for future tsunami warning systems.

Current Applications and Projects

The use of AI in weather and tsunami forecasting is growing rapidly among research groups and agencies. NOAA (US National Oceanic and Atmospheric Administration) has created a NOAA Center for Artificial Intelligence to coordinate ML projects across weather and climate. Notably, NOAA’s National Hurricane Center (NHC) has begun incorporating experimental AI forecasts: in mid-2025 NOAA announced a partnership with Google DeepMind to feed AI-predicted storm tracks and intensities into its workflow. NOAA is also improving hurricane models (HAFS) with ML enhancements and exploring ML in translation and communication of warnings.

NASA is likewise active: it released an open-source climate AI model (“Prithvi”) trained on massive Earth data. In tests, Prithvi (a transformer-based foundation model) outperformed specialized AI models on hurricane forecasting. NASA scientists envision future forecasts that fuse data-driven models with traditional physics (hybrid approach). NASA’s Disasters Program uses satellite data (often processed by ML) for real-time damage assessment (e.g. mapping power outages after storms).

The World Meteorological Organization (WMO) has launched an AI initiative. In 2023 WMO adopted a strategic plan to integrate AI into Earth-system science. WMO’s Early Warnings for All partnership (with UNDRR and others) is piloting AI solutions for disaster risk reduction. For example, WMO reports that companies like Google DeepMind and NVIDIA have released AI-based medium-range weather models that are now being blended into Europe’s operational ECMWF forecasts. Regional programs like MedEWSa (EU-funded) are deploying AI/IoT to enhance European forecasts of floods, droughts and storms.

In the private and research sectors, tech companies and universities are building new tools. Google offers free global flood forecasts via Flood Hub (collaborating with governments). Microsoft and IBM, in partnership with academic labs, are developing open AI models for climate data: e.g. IBM/NASA’s planned open weather foundation model and Microsoft’s open-source Aurora. Academic groups (e.g. Western University, University of Houston) continue to publish ML-based tsunami and storm forecasting models. Overall, a broad ecosystem has emerged involving national agencies, international bodies, tech companies, and universities all contributing to AI-driven disaster prediction.

Future Directions and Recommendations

To fully realize AI’s potential in disaster forecasting, technical, policy and infrastructure enhancements are needed. Key recommendations include:

• Data & Sensor Expansion: AI’s power depends on data quantity and quality. As noted, tsunami models suffer from sparse sensors (Canada has 4 buoys vs. Japan’s 150). Investing in wider sensor networks (ocean bottom pressure gauges, river gauges, weather radar, satellite constellations) will dramatically improve AI forecasts. Governments and agencies should fund more real-time monitoring (IoT) and open data sharing so models have rich training and input data.

• Advanced Hybrid Modeling: Continue developing hybrid AI-physics systems. Embedding domain knowledge into ML (known as physics-informed or knowledge-guided ML) ensures consistency with geophysical laws. For example, constraining ML outputs to follow known tsunami wave equations or conservation laws can reduce implausible results. Research should focus on causal and interpretable AI architectures (as recommended by recent reviews) so that forecasts are not just accurate but also explainable to experts.

• Foundation and Ensemble Models: Build and leverage large-scale AI models for climate and multi-hazard forecasting. The success of models like Aurora shows that a single “foundation” network can be fine-tuned to many tasks (storms, floods, air quality). Future work should explore generative and ensemble approaches – e.g., using climate model ensembles combined with generative AI to produce long-term risk forecasts. Cloud-computing platforms and shared model repositories (like WMO’s planned AI framework) can democratize access to these tools worldwide.

• Fairness, Trust, and Accessibility: AI forecasts must be equitable and trusted by end-users. Following guidelines like the WMO/UN FATES principles (Fairness, Accountability, Transparency, Ethics, Sustainability) is essential. This means being vigilant about biases in training data (most satellite data favors richer countries), and ensuring systems are validated across diverse regions. User-centered design is crucial: early-warning interfaces should be clear and localized (multilingual alerts, mobile apps). Addressing the digital divide – for example by making AI tools energy- and cost-efficient (the new AI emulators run on laptops) – will broaden their reach to underserved communities.

• Collaboration and Policy: International collaboration should be strengthened. Agencies should continue joint data sharing (e.g. through WMO) and multi-country drills. Policy makers need to update standards to allow AI forecasts in official warnings once properly validated. The successful WMO MedEWSa and Early Warnings for All projects show that blending AI with traditional systems under coordinated governance accelerates progress. Finally, legal and ethical frameworks (for AI in public safety) must be developed so that responsibility for AI-assisted alerts is clearly defined.

In summary, integrating AI into tsunami and weather disaster forecasting demands technical innovation (advanced models, sensor networks, hybrid systems) along with infrastructure and policy support (open data, agency cooperation, community engagement). As multiple case studies have shown, AI can significantly extend lead times and accuracy of forecasts. Continued research on robustness, fairness and real-time implementation will further improve its societal impact.