If a report on election polling indicated one candidate was leading 44% to 40%, plus or minus 40%, you’d see this information as statistically useless. This gives some context to what the National Oceanic and Atmospheric Administration (NOAA) is contending with today.

NOAA fish stock assessments were once the envy of the world. The agency’s data stream included landings, aging data, skipper interviews, natural mortality estimates and biannual trawl surveys. These inputs supplied a statistical model called the virtual population analysis, VPA for short. The time series — critical for population analysis — stretched back decades.

The New England fleet once consisted of 2,500 active fishing boats. The fleet size was intentionally grown after we closed the Gulf of Maine and Georges Bank to foreign boats. The U.S. made fishing boats tax shelters. The fleet grew drastically.

Fish stocks, rebounding from the exclusion of foreign fleets, began to crash again. NOAA regulators addressed overfishing with closed areas, gear restrictions, vessel limits, catch limits and trip limits. All these affected the data stream. The tools used to check accuracy began to go haywire.

Fisherfolk who once fished anywhere, anytime, with any gear, faced limits on everything. The VPA “saw” landings as directly proportional to stock size but couldn’t “see” the restrictions impact — the models “saw” less fish.

Seasoned NOAA employees who built relationships with fishing captains and were able to create a picture of the fish stocks in the ecosystem were replaced by private contractors who stopped gathering that kind of information. The contractors also missed a lot of aging samples critical to identifying year classes. Logbook information often languished unused in a clunky NMFS database. The model “saw” less fish.

The trawl survey was affected by incorrect deployment of the fishing gear and a net mismatched to the survey vessel. Scientists tried to “fix” the data generated with statistical “treatments.” While these fixes were declared successful, it felt a lot like a science project gone awry that needed to be fudged. The model “saw” less clearly.

While most fisherfolk worked to survive in the new regulatory climate, some misreported landings, some of those on a large scale. (Remember the “Codfather”?)  The model became less reliable.

Some stocks were assessed at very low levels, resulting in low catch limits. Despite the poor assessments, these fish often showed up in large numbers. Fishing boats either tied up or fished in other fishing grounds. The model didn’t recognize this behavior; it only saw fewer fish.

The uncertainty grew to plus or minus 40%. Statistically useless? Nope. In fact, because the Sustainable Fisheries Act requires NOAA to use “the best available science,” regulators and scientists used the shoddy science because they were not required to meet any precision standards.

There are now fewer than 200 active fishing boats left in New England and the average length is now down to 50 feet.

NOAA Fisheries takes comfort in stating that observers now cover 90% of all fishing trips and port sampling coverage percentages have gone way up. Yet uncertainty is still very high and has even gone up. Because of all this, New England fishermen leave behind as much as 65% of the legally catchable fish every year — more than 60 million pounds in 2024. How do you think the models “see” that? Short answer: Poorly, if at all.

There is some good news here. The aggregate biomass in New England waters is very high. (The species mix is different and may never be the same unless the water stops warming.) Fishing pressure is historically low.

Given advancements in digital imaging and AI, stock assessments can also be changed. An advanced digitally imaged, time stamped and geo-located survey, already under development, can and should be fully developed and used.

Imagine a “net” that catches information. Gear that can be towed for many hours at a time, recording digital images that can be decoded using AI to identify size, species and numbers of fish. Add time and location and you have a statistician’s dream database.

I’ve heard reasons from folks in Woods Hole about why this can’t work — silt clouds occasionally obscuring the images, lights possibly scaring the fish.

My questions to those folks are simple: Since when do scientists look for reasons not to solve problems? How about looking at obstacles as opportunities to create solutions? Since when do scientists accept inaccurate results and use the routinely for decades? It’s time to change. It’s time for the U.S. to produce accurate stock assessments again.