The leading minds in meteorology say weather forecasts aren’t reliable more than seven to 10 days out. Yet AccuWeather, is — more than ever — defying what science says is the limit of useful weather forecasting.

In recent days, AccuWeather added forecasts extending 90 days into the future on its website, doubling the amount of time into the future for which the State College, Pa.-based company provides predictions.

But valuable forecasts so far into the future simply aren’t possible, many scientists with expertise in weather prediction say. I asked three leading meteorologists, who hold doctoral degrees, what the limit of reliable weather forecasts are and they all responded with essentially the same answer — 7-10 days:

For detailed, specific forecasts that are better than “average,” the limit is generally just over a week. However, in some situations, skillful forecasts can be obtained out to 2 weeks, but this is rare. — Gary M. Lackmann, professor of atmospheric sciences, North Carolina State University.

Meteorologists generally have excellent skill the first two days, good skill days 3-4, useful skill for days 5-6. Skill fades rapidly after 7 days, but increasingly we are situations in which the model made excellent forecasts 6-8 days out… — Cliff Mass, professor of atmospheric sciences, University of Washington.

On average, predictability is just 10-20 minutes for a tornado, a few days for a winter storm, and 7-10 days for general weather pattern change (i.e., turning warmer over the eastern U.S.). Beyond this 8-10 day time frame, changes in daily weather are very difficult to forecast. — David Novak, director of the National Weather Service Weather Prediction Center

(Note: I did not mention AccuWeather’s forecasts when querying these experts.)

Yet despite the academic view that forecasting beyond seven to ten days offers no value, AccuWeather has boldly pushed well beyond that envelope, to the consternation of some scientists.

In August 2013, when AccuWeather first launched forecasts to 45 days, it caused quite the stir in the weather community. I, for one, sharply criticized the effort.

“[AccuWeather] is simply peddling a useless product to people who don’t know better,” I wrote.

University of Washington’s Cliff Mass wrote a scathing review of the forecasts on his blog calling them “a profound disservice” to meteorology.

I published AccuWeather’s defense of its long-range forecasts and it boiled down to. “We have the information, there’s no reason the public shouldn’t have it,”  in the words of AccuWeather’s chief operating officer, Evan Myers.

But AccuWeather’s critics, including myself, have long questioned the practice of disseminating long-range forecasts just because they’re available, without explaining their uncertainty and transparently discussing their track record.

And their track record is, at best, questionable.

Several people have tested AccuWeather’s forecasts beyond 10 days and have concluded they’re not valuable because they predict the weather no better than a simple average of historical conditions, or climatology.

But in a conversation we had in December 2014, Joel Myers, AccuWeather’s founder, challenged the legitimacy of these tests. “The results were not statistically significant,” Myers said. “It’s not how scientific studies are done.”

Myers, who holds a PhD in meteorology,  was emphatic that his company’s forecasts have scientific value and that the public understands their shortcomings.

“We’re very scientifically-based company,” Myers said.  “We wouldn’t issue something that didn’t have a scientific basis.  . . .  People accept the fact a forecast is not perfect, but they want best estimate the science will allow.”

Myers rejected the idea of limits to predictability. “Models have gotten better,” he said. “It’s been an exciting development.  . . . People always stand in the way of progress, limited by what they believe and what they’ve seen.”

He added that he and his company faced pushback in the 1960s when they extended forecasts beyond today, tonight and tomorrow to three to five days.  He argued that AccuWeather’s aggressive scientific approach has helped to advance the business while improving services for consumers and gaining their trust.

“It’s important that we’re all logical and working towards public good and business good. That’s got to intertwine,” Myers said. “We’ve been around 50 years when a lot of other companies have come and gone. Trust is part of our DNA and is what it’s all about.”

85 days from now, on Independence Day, July 4, AccuWeather is calling for partial sunshine, pleasant 85-degree temperatures and just a 1 percent chance of thunderstorms. Time will tell how trustworthy that is.