Prediction is the goal and test of all science, one might argue. Meteorologists have been curious about the inherent limits of weather forecasting ever since the dawn of numerical weather prediction in the late 1950s. Now, a new study published in Advances in Atmospheric Sciences offers a bold answer: even under perfect conditions, there is a fundamental limit—and it is about 129 days .
Previous attempts to determine this predictability limit have largely focused on analyzing how tiny errors in today's forecasts grow over time. But as lead author Dr. Wei Zhang, a climate scientist at the University of Miami and the NOAA Cooperative Institute for Marine and Atmospheric Studies (CIMAS), points out, that approach has fundamental blind spots.
"How could we say we can accomplish something like make skillful very long-range forecasts without actually being able to demonstrate it?" Dr. Zhang asks. "Unfortunately, there is no observational, theoretical, or modeling experience as to how errors smaller than in today's forecasts may behave," Without that knowledge, it is hardly surprising that no solid answers have emerged—until now.
The research team, including co-author Dr. Zoltan Toth (recently retired from NOAA), decided to ask whether there might be a dramatically different perspective on this perennial question. "Our question was whether there is a dramatically different, new perspective to approach the perennial questions of predictability," Toth recalls.
First, they refined the key question with unprecedented precision: Assuming ideal conditions where we exactly know the initial state of the atmosphere, the governing dynamics, and all future macro-scale boundary conditions—what, if any, would be the ultimate limit of weather prediction?
In their search, instead of focusing on prediction errors, they went back to basics—the energetics of the atmosphere. They reasoned that if the initial state were exactly known, that knowledge would be preserved by the exactly known dynamics of the atmosphere, yielding perfect forecasts of the true state forever. The one exception, however, is the quantum-scale uncertainty injected into the atmosphere in terms of the phase of photons in the incessant incoming solar radiation, which they assumed to be unknown.
From there, the story unfolded with ease. Looking at the atmospheric energy cycle, solar radiation ultimately fuels all motions, sooner or later reaching every molecule. The team figured that by the time solar energy reaches all parts of the atmosphere, the uncertainty associated with the unknown phase of the incoming photons must have erased all memory of the initial state—which would otherwise have been preserved forever. In other words, beyond a certain time—which they call the "energy turnover point" —prediction becomes impossible.
Considering the atmosphere's total energy, the incoming flux of solar radiation, and their observational uncertainties, the authors established 129 ± 7 days as the likely limit of internal (as opposed to externally modulated) predictability of the weather. Forecast skill today is limited to about 14 days, which leaves a great deal of room for improvement.
Interestingly, the extra time of predictability beyond today's 14-day limit is divided roughly equally between what one might call a genuine extension of skill and a similarly long period of marginal skill added at the end of a forecast. In other words, under ideal conditions, the skill of a 5-day forecast today could theoretically be extended all the way to about 62 days. The remaining period would offer only low-confidence guidance.
Dr. Zhang and his colleagues are now working on other, independent estimates to confirm these results. If validated, the finding not only sets a theoretical ceiling but also gives forecasters a tangible target for how far the science of weather prediction might reasonably advance in the future.
Advances in Atmospheric Sciences
A New Approach to Estimating the Limit of Predictability
29-Jul-2026