Twenty-one gray whales were found dead in and around San Francisco Bay in 2025, and researchers say two-fifths of those deaths were caused by ship strikes. That number explains why a new AI-powered whale detection system is being deployed now, not as a science demo but as a live warning layer for one of California’s busiest waterways.
The system combines thermal cameras, computer vision, and human verification to spot whales as far as 7 kilometers away, then alert nearby vessels so crews can slow down or change course, according to Science News AI. A coalition including ocean scientists, the U.S. Coast Guard, whale tracking experts, and local ferry companies unveiled the San Francisco Bay deployment on May 19.
This is not chatbot AI. It is closer to the AI used in scientific imaging: a model trained to recognize patterns in sensor data. That distinction matters as AI moves from text generation into field science, a shift we have tracked in Singularity Bet Recasts Google I/O's AI-Driven Science and in coverage of faster model deployment such as Google Sparks AI Race with Gemini 3.5 Flash’s Breakthrough Speed.
Hungry gray whales are entering a crowded shipping zone
Gray whales are not just passing offshore. Researchers first observed in 2018 that more of them were making a stop in San Francisco Bay during their long migration from feeding grounds off Alaska to mating grounds near Mexico.
The reason appears to be food stress. In the Arctic, gray whales feed on amphipods in ocean sediments. Those amphipods are nourished by algae that grows beneath sea ice. As climate change melts that sea ice, the food chain is disrupted, leaving whales hungry during migration.
The population decline has been sharp. Science News reports that gray whale numbers fell from about 20,500 in 2018 to about 14,500 in 2023, with hundreds of whales stranded along the North American west coast and many showing signs of malnutrition.
That pushes whales into a dangerous trade-off. San Francisco Bay offers a place to search for food, but it is also crossed by container vessels, ferries, tankers, speed boats, fishing boats, and tour boats. Large ships cannot maneuver quickly. A whale surfacing in the wrong place can become visible too late.
“We wanted to be able to detect whales so far out that it would give mariners time to take action,” says Daniel Zitterbart, a physicist at Woods Hole Oceanographic Institution and chief scientist of WhaleSpotter.
Thermal cameras give the AI something specific to see
The technology works because whales are warmer than the water around them. Researchers say water emitted from whales’ blowholes, or the whales’ bodies themselves, is about 2 degrees Celsius warmer than ambient water.
That small temperature difference becomes the signal. Thermal cameras scan the bay around the clock. The AI model, trained on hundreds of thousands of thermal images, looks for heat patterns consistent with a whale body or blow.
This is a narrow computer vision task. The model is not “understanding” whale behavior in a human sense. It is classifying thermal imagery and flagging likely detections for review.
The human layer is central. When the system detects a possible whale, a WhaleSpotter researcher verifies the data before an alert goes out. That step is meant to reduce false positives before mariners are asked to act.
Analysis: The most useful part of this system is not that it detects whales in theory. It is that it compresses the time between a whale entering a risky area and a warning reaching people who can change vessel behavior.
From Angel Island to vessel alerts: how the warning chain works
The first camera is mounted on a radio tower on Angel Island, where it monitors busy shipping routes inside the bay. A second camera will be installed on MV Lyra, a passenger ferry operated by San Francisco Bay Ferry on a daily route between Vallejo and downtown San Francisco.
Future camera sites could include the Golden Gate Bridge and Alcatraz, according to the deployment plan described by Science News and UC Santa Barbara materials.
| Detection point | Role in the system | Status from source material |
|---|---|---|
| Angel Island radio tower | Monitors busy shipping routes from a fixed point | Deployed |
| MV Lyra ferry | Adds a vessel-mounted camera on a daily Bay route | Planned |
| Golden Gate Bridge / Alcatraz | Possible expansion sites for wider coverage | Under consideration |
Once a detection is verified, an alert goes to vessels in the area. The operational response is practical: slow down, change course where safe, or increase caution around the whale’s location.
The U.S. Coast Guard’s Vessel Traffic Service is part of the coalition. UC Santa Barbara’s Benioff Ocean Science Laboratory says whale detections can be mapped on the Whale Safe website and shared with Bay mariners and the Coast Guard, which can radio vessels about whales under imminent threat.
The real-world case study is the Angel Island node
The best concrete example is not hypothetical. It is the Angel Island deployment.
That camera points across the bay toward Treasure Island and the Bay Bridge, covering an area where whales and vessels can overlap. Its value comes from range: the system can detect whale heat signatures up to four nautical miles, or 7 kilometers, away.
Here is the actual workflow, based on the deployment description:
- Scan: A thermal camera watches the water continuously.
- Flag: The AI identifies a possible whale based on thermal imagery.
- Verify: A WhaleSpotter marine mammal specialist checks the detection.
- Alert: Nearby vessels or traffic managers receive the warning.
- Act: Mariners can reduce speed or reroute when conditions allow.
That is the whole proposition. The technology does not remove navigation judgment. It gives crews and traffic managers earlier information than visual spotting alone might provide.
“Shipping is not going to disappear. We need to have a tech that allows us to use the ocean, but also allows the whales to go about their lives,” Zitterbart says.
The system’s biggest test is trust
The sources describe a detection-and-alert network, not a proven reduction in whale deaths. That distinction matters.
For the system to work at scale, mariners need alerts that are timely and credible. Researchers also need enough coverage to reduce blind spots across a large, busy bay. The current deployment starts with one fixed camera and one planned ferry-mounted camera; broader coverage would require more sites.
There is also a coordination challenge. The coalition includes scientists, the Coast Guard, whale specialists, and ferry operators. That mix is a strength because no single group controls the entire problem. But it also means the system’s impact depends on whether detections move cleanly from research tools into vessel operations.
Analysis: The human verification step helps credibility, but it also makes the system a hybrid: fast AI detection plus expert confirmation. That is probably the right model for a safety-critical conservation tool where both missed whales and bad alerts carry costs.
Beyond San Francisco Bay, the model is real-time conservation
The San Francisco Bay deployment shows how AI can move conservation from after-the-fact counting to live intervention. Instead of documenting whale deaths after strandings, the system tries to warn vessels while there is still time to avoid a collision.
The approach could be adapted anywhere the same ingredients exist: predictable whale presence, vessel traffic, and places to mount sensors. The supplied sources specifically name future Bay sites such as the Golden Gate Bridge and Alcatraz, and researchers want as many deployments as possible to improve visibility over the ocean.
Marine ecologist Douglas McCauley, director of UC Santa Barbara’s Benioff Ocean Science Laboratory, framed the stakes bluntly:
“It is heartbreaking to see these starving whales stumbling around in the middle of the hustle and bustle of San Francisco Bay. Every day is a nail-biter.… This new system will save whales’ lives.”
The practical watch item is whether verified detections become routine enough that captains act on them without hesitation. If the network expands and alerts prove useful in day-to-day vessel traffic, AI whale spotting could become a working layer of port safety. If coverage remains patchy, it will still help in specific zones — but the bay’s whales will keep crossing blind spots.
Impact Analysis
- AI detection could help vessels avoid gray whales in one of California’s busiest waterways.
- Ship strikes are a major known threat, causing two-fifths of reported gray whale deaths in the area in 2025.
- Climate-driven food stress is pushing whales into crowded migration zones, raising collision risks.
