Prediction is often described as the ultimate test of science, and nowhere is that challenge more visible than in weather forecasting. For decades, meteorologists have worked to determine how far ahead the atmosphere can be predicted, even under the most favorable imaginable conditions. A new study published in Advances in Atmospheric Sciences proposes a striking answer: the atmosphere may have an ultimate internal predictability limit of approximately 129 days.
The estimate is far longer than the roughly 14-day horizon that generally constrains today’s operational weather forecasts, but it is also presented as a fundamental ceiling rather than a promise of routine four-month forecasts. The researchers argue that even a hypothetical forecasting system with perfect knowledge of the atmosphere, flawless physical equations and complete information about future large-scale boundary conditions would eventually lose the ability to predict the detailed state of the weather.
The study, led by Wei Zhang of the University of Miami and the NOAA Cooperative Institute for Marine and Atmospheric Studies, takes a different route from most previous investigations of predictability. Traditional studies usually examine how small errors in the initial atmospheric state amplify through time. This approach is rooted in chaos theory: because the atmosphere is a nonlinear dynamical system, tiny inaccuracies in temperature, pressure, humidity or wind can grow until the forecast diverges substantially from reality.
Yet, according to Zhang and co-author Zoltan Toth, that framework cannot fully answer what would happen if the initial state were known with absolute precision. Forecasting systems today are limited by imperfect observations and incomplete models, but researchers have little direct experience with errors smaller than those already present in modern analyses. As a result, simply extrapolating the growth of today’s forecast errors may not reveal the ultimate limit imposed by nature itself.
The authors therefore began with an idealized thought experiment. Suppose the atmosphere’s complete initial condition were known exactly. Suppose, too, that the governing equations describing atmospheric motion were perfect and that all relevant future boundary conditions were available without uncertainty. Under those assumptions, the atmosphere’s state could, in principle, be carried forward indefinitely by deterministic dynamics. The initial information would not disappear merely because the system was chaotic; it would remain encoded in the exact evolution of the atmospheric flow.
The researchers identify one exception to this perfect-determinism scenario: quantum-scale uncertainty associated with incoming sunlight. Solar radiation arrives as an enormous stream of photons, and the precise phases of those photons are not known. Although such uncertainty is minuscule at the individual-particle level, the atmosphere is continuously powered by solar energy. That energy drives winds, convection, evaporation, storms and the global circulation, eventually reaching and influencing molecules throughout the atmosphere.
This led the team to formulate what they call the “energy turnover point.” Their reasoning is that solar energy does not remain confined to the place or process where it first enters the climate system. Through the atmospheric energy cycle, it is redistributed across an immense number of motions and interactions. Over time, the quantum-scale uncertainty carried by incoming radiation could become mixed into the atmosphere’s total energy budget. Once that uncertainty has circulated through the system, it would erase the distinguishable imprint of the exact initial state.
The researchers estimated this timescale by comparing the atmosphere’s total energy with the incoming solar energy flux, while accounting for observational uncertainties in the relevant quantities. Their calculation produced a likely internal predictability limit of 129 days, with an uncertainty of approximately seven days. In this framework, the result does not mean that every detail of the weather becomes completely random on day 130. Rather, it indicates that the atmosphere should eventually lose the information needed to reconstruct its precise earlier state, even under idealized conditions.
The implications for forecasting are more nuanced than the headline figure suggests. The gap between today’s approximately 14-day practical limit and the proposed 129-day theoretical ceiling would not necessarily translate into more than three months of equally reliable forecasts. The authors estimate that the additional potential is divided roughly between a meaningful extension of forecast skill and a later period in which predictions might provide only marginal, low-confidence guidance.
Under the study’s interpretation, a forecast that is considered skillful at five days today could potentially be extended to about 62 days in an ideal future system. Beyond that point, forecasts might still contain weak statistical information about the atmosphere, but their practical value would diminish sharply. Such a distinction is important because predictability is not a simple yes-or-no property. Forecasts can retain measurable skill while becoming increasingly uncertain, and the level of confidence required depends on whether the information is used for agriculture, disaster preparation, energy management or everyday planning.
The proposed limit also does not remove the familiar influence of chaos. Errors in observations, deficiencies in numerical models, unresolved small-scale processes and uncertainty in ocean and land conditions would continue to restrict real-world predictions long before the energy turnover point was reached. Weather forecasting would therefore need major advances in observing technology, computing power, data assimilation and physical modeling before the theoretical boundary became operationally relevant.
Zhang and Toth describe their calculation as a new perspective on a long-standing scientific question rather than a final verdict. The team is pursuing independent estimates that could test whether the 129-day value is robust. If future work supports the result, it would provide meteorologists with an unusual benchmark: not a forecast for a particular storm or season, but a physically motivated target defining how far the science of weather prediction might advance before nature itself begins to withhold the necessary information.
Subject of Research: The fundamental predictability limit of weather and the role of atmospheric energy turnover and quantum-scale uncertainty.
Article Title: A New Approach to Estimating the Limit of Predictability
Web References: https://doi.org/10.1007/s00376-026-5621-8
References: Advances in Atmospheric Sciences, DOI: 10.1007/s00376-026-5621-8
Keywords: Weather forecasting, atmospheric predictability, meteorology, chaos theory, numerical weather prediction, atmospheric energy cycle, quantum uncertainty, solar radiation, forecast limits, climate science
