From Pandemics to Whales: Insights into Forecasting

The COVID-19 pandemic has rapidly transformed the world. Seemingly overnight, it flipped daily routines upside-down, bent economies, and obscured the future. Doctors, researchers, governments, businesses, and individuals continue to battle this disease on every front, every day.

One of the most powerful tools in this fight is data. Total cases, recovery rates, new transmissions — these numbers are the fuel behind mathematical models that elucidate possible versions of our unknown future and help us make progress against the virus.

“Decision-makers and society as a whole are walking a tightrope right now, trying to both save lives and figure out how to keep the economy going,” said Senior Research Scientist Nick Record, a mathematician and modeler. “To get the balance right, we need as much data and science as possible.”

Just as Record uses environmental data to forecast natural systems, epidemiologists feed data related to the novel coronavirus into predictive algorithms. The resulting forecasts shape our daily lives in unprecedented ways as countries, businesses, and individual people rely on them for long-term and even daily decision-making.

But predicting the path of this virus, or any natural system that involves human behavior, is complicated. The current situation underscores the inherent complexity, valuable opportunities, and deep importance of both data and forecasting — and it highlights broader questions of how scientists and societies can best wield these powerful tools.

After reflecting on the relationship between his forecasting research and the current situation, Record came up with four key insights about the challenges and opportunities of this rapidly evolving science that he hopes we can all learn from this pandemic.

1: People lack intuition for interpreting biological forecasts

Many people see weather forecasts every day. Over years, we have developed intuition about how to read them and react — a gut feeling for what a 10 percent chance of rain really means, or whether next week will require us to break out a warmer jacket. But before the current crisis, no one had seen daily pandemic forecasts. We often can’t readily interpret these with a casual glance, and the exponential growth that natural systems like the virus follow is unintuitive. In addition, people tend to disbelieve predictions of threats they haven't personally experienced before.

2: Human behavior is often part of the equation

When Record forecasts the constant ebb and flow of summer jellyfish populations, the way that people act has no bearing on a given day’s number. But his forecasts of encounters between jellyfish and humans can be made inaccurate by human action — people may choose to head to the lake on a high-jelly day, or take a hike instead.

Climate scientists navigate this same terrain with climate projections. Global temperatures depend in part on human actions. If a model projects extreme global warming, and humans respond by dramatically cutting carbon emissions, the projection won't match reality anymore. When projections like this serve their purpose, they cease to be accurate predictions of the future.

Faced with the current pandemic, the world is dealing with this paradox in real time. We’re making personal decisions and crafting policies based on forecasts that shift in sync with our choices. It’s helpful to think of epidemic projections not as absolutes, but as consequences given certain choices that we make.

3: Uncertainty with human systems can be trickier

Nature is a math problem. Meteorologists know the equations that describe how the atmosphere moves, how winds gain strength, and how water freezes and thaws. For the most part, they just need to collect the right data to solve those equations and produce weather forecasts.

But with biological systems, like pandemics or red tides, scientists don't always know the right equations. Researchers are accustomed to dealing with typical sources of uncertainty, like measurement error. In uncharted forecasting territory, another realm of uncertainty exists about which processes are important to the equations. That uncertainty can lead to wildly differing predictions about the future.

4: Sharing forecasts can have unintended consequences

Biological forecasts are powerful tools, and they can be used in unexpected ways once made public. For instance, Record works with scientists and resource managers to forecast the locations of North Atlantic right whales, informing strategies to protect this endangered species. Openly sharing these forecasts, however, could lead to whale watching vessels trying to view these remarkable animals, inadvertently increasing the risk of ship strikes.

The same is true of the pandemic. The way that forecasts are presented, interpreted, and used can have complex and unintended consequences. Any public forecast needs thoughtful review from multiple perspectives to understand exactly how it will be used — and how it could be misused.

Scientific momentum

The field of forecasting is expanding rapidly, fueled by new approaches in computer science like artificial intelligence and big data. As scientists generate, respond to, and learn from pandemic forecasts, this momentum can feed back into new methods and discoveries for basic research.

“As computer science becomes more and more powerful, the question is, how can we forecast other aspects of the natural world?” Record said. “Much of my research involves piecing data together into a larger picture of what the ocean and the ecosystem are doing, and that really amounts to a big math problem.”

Important questions remain in the nascent field of forecasting, and their answers will help researchers work creatively and capitalize on the trove of ocean science data that already exists. A new computational hub at Bigelow Laboratory, called the Big Data Discovery Initiative, will soon expand researchers’ capacity to mine data for discoveries and solutions to global challenges.

Some of Record’s ecosystem forecasts rely on crowd-sourced data, presenting him with a common big data challenge — a wide range of information quantity and quality. Differences in jellyfish reports from along Maine’s coastline, for example, can give Record a lot of high-quality data from one specific location, while sparsely populated areas of the state may have very few reports.

“Integrating two types of data is a really interesting math problem, and finding the solution can help researchers answer a host of new questions,” Record said. “We still know so much more about the land and the atmosphere than we do about the ocean. Continuing to build our approach to modeling can allow us to learn new things from data we already have and are collecting every day.”