As the Summer of COVID-19 draws to a close, many experts fear an even bleaker fall and suggest that American families should start planning for Thanksgiving by Zoom.

Because of the many uncertainties, public health scientists say it’s easier to forecast the weather on Thanksgiving Day than to predict how the U.S. coronavirus crisis will play out this autumn. But school reopenings, holiday travel and more indoor activity because of colder weather could all separately increase transmission of the virus and combine in ways that could multiply the threat, they say.

Here’s one way it could go: As more schools open for in-person instruction and more college students return to campuses, small clusters of cases could widen into outbreaks in late September. Public fatigue over mask rules and other restrictions could stymie efforts to slow these infections.

A few weeks later, widening outbreaks could start to strain hospitals. If a bad flu season peaks in October, as happened in 2009, the pressure on the health care system could result in higher daily death tolls from the coronavirus. Dr. Robert Redfield, director of the Centers for Disease Control and Prevention, has said that scenario is his biggest fear.

One certainty is that the virus will still be around, said Jarad Niemi, a disease-modeling expert at Iowa State University.

“We will not have a vaccine yet and we will not have enough infected individuals for herd immunity to be helpful,” Niemi said.

Fall may feel like a roller coaster of stop-and-start restrictions, as communities react to climbing hospital cases, said University of Texas disease modeler Lauren Ancel Meyers. Everyone should get a flu shot, she said, because if flu spreads widely, hospitals will begin to buckle and “that will compound the threat of COVID.”

“The decisions we make today will fundamentally impact the safety and feasibility of what we can do next month and by Thanksgiving,” Meyers said.

The virus is blamed for over 183,000 deaths and 6 million confirmed infections in the U.S. Worldwide, the death toll is put at almost 850,000, with over 25 million cases.

The U.S. is recording on average about 900 deaths a day from COVID-19, and newly confirmed infections per day are running at about 42,000, down from their peak in mid-July, when cases were topping out at over 70,000.

Several vaccines are in advanced testing, and researchers hope to have results later this year. But even if a vaccine is declared safe and effective by year’s end, as some expect, there won’t be enough for everyone who wants it right away.

Several companies are developing rapid, at-home tests, which conceivably could be used by families before Thanksgiving, but none has yet won approval.

More than 90 million adults are over 65 or have health problems, putting them in higher danger of severe consequences if they get sick with the coronavirus. Many of them and their families are starting to decide whether to book holiday flights.

Cassie Docking, 44, an urgent care nurse in Seattle, is telling her parents — both cancer survivors — that Thanksgiving will be by FaceTime only.

“We all want to get to 2021,” she said, “and if that’s what it takes, that’s what we’ll do.”

Caitlin Joyce’s family is forging ahead with a holiday feast. They plan to set up plywood tables on sawhorses in a large garage so they can sit 6 feet apart.

“We’ll be in our coats and our sweaters,” said Joyce, 30, of Edmonds, Washington, who plans to travel to her grandparents’ home in Virginia. “It will be almost like camping.”

One widely cited disease model projects 2,086 U.S. deaths per day by Thanksgiving — more than double compared with today.

“In our family we will not have our extended family get-together. We will stick to the nuclear family,” said Dr. Christopher Murray of the Institute for Health Metrics and Evaluation, one of the few models making a prediction for November.

Uncertainty is huge in Murray’s model: Daily deaths could be as low as 1,500 by Thanksgiving or as high as 3,100. In a more optimistic scenario, daily deaths could range from 510 to 1,200 if nearly everyone wears masks.

A more pessimistic scenario? From 2,700 to 6,500 daily deaths if social distancing rules continue to be lifted and are not reimposed.

With all the uncertainty, most disease modelers aren’t looking that far ahead — at least officially.

At Carnegie Mellon University in Pittsburgh, computer scientist Roni Rosenfeld’s team uses machine learning to project COVID-19 deaths. The team’s computer algorithm learns from patterns it finds in state and county data to improve its forecasts.

A five-time winner of a CDC competition for predicting flu season activity, Rosenfeld thinks his model’s COVID-19 projections aren’t very useful beyond four weeks because of the wild card of human behavior, including that of government officials.

“What happens very much depends on us,” he said. “People, myself included, don’t always behave rationally.” Presented with the same facts, “the same person might behave differently depending on how sick and tired they are of the situation.”