Skip to main content
Menu
Why We Don’t Know If It Will Be Sunny Next Month But We Know It’ll Be Hot All Year

Weather forecasts more than a week ahead are notoriously unreliable. Yet climate scientists have been predicting since December that 2016 will be the warmest year since at least the 19th century. What gives?

It is true that predictions of weather — the specific cycles of rain, temperature and winds — degrade markedly beyond three to five days. While 10-day forecasts are now becoming more common, they are only now approaching the skill of five-day forecasts of 30 years ago. Useful weather forecasts beyond that don’t exist (though plenty of folks are happy to sell them to you).

But even if the weather isn’t predictable further out, some aspects of the climate system can be usefully forecast over months, years and even decades. The key to understanding why is realizing that there are many potential sources of predictability: the motions of the atmosphere in the short term (days); the motions of the ocean (months to years); and external factors such as the sun’s activity, the orbit or greenhouse gases (years to millennia).

Weather forecasts rely predominantly on the predictability of turbulent swirls of a chaotic atmosphere; we see these in the satellite data as the bands of clouds moving west to east in the mid-latitudes.

These systems can be tracked for a couple of weeks. Using a global network of observations as input and our physical understanding of the atmosphere, weather simulations do a good job at estimating how fast these systems will travel and how they interact. However, imprecision in the observations means that simulations and reality diverge over time and, after a month, the patterns are completely different. This is the phenomenon that Ed Lorenz famously first described as “chaos,” and from this we know that, not only do we lack the technology to make a skillful 30-day weather forecast, it is fundamentally impossible.

So weather forecasts rely on the predictability of atmospheric systems that persist a couple of weeks. But some ocean patterns in the climate system can persist much longer, and understanding them can help make useful predictions for regional and global averages that don’t depend so much on specific weather patterns.

The most notable ocean pattern of variability is the El Niño/Southern Oscillation (ENSO). In the tropical Pacific, the distance from Indonesia to South America and the way tropical winds push warm water west combine to allow special waves to travel along the equator and are amplified by the atmospheric wind response to produce large fluctuations in temperatures (up to 3 degrees Celsius) in the Eastern Pacific that last for months. Since the 1980s, we’ve had sufficient understanding of ENSO to be able to predict the occurrence and speed of these waves and, consequently, the variability of ocean temperatures in the Eastern Pacific about six months in advance.

These fluctuations in temperature can be so large that their influence can be felt globally. Warm El Niño phases cause excess rain in Peru and Ecuador; drought in Brazil’s Nordeste region, Indonesia and Northern Australia, and weather responses in North America and Antarctica. There is a clear impact on global temperature, too, though the mechanisms are complex: heat released from the oceans; increases in water vapor, which enhance the greenhouse effect, and redistributions of clouds. Statistically, the impact on global temperature peaks two or three months after changes in the tropical Pacific. Because these events are large, take months to play out and can be predicted months ahead of time, we can use ENSO to predict the next year’s global temperatures.

How does this work? Global temperature estimates are tracked by multiple groups, including NASA, using data from weather stations, ocean ships and buoys. The analyses are based on calculating temperature differences at one point in time relative to the average over a certain period (anomalies) and creating a time series of averaged global temperature change.

We can separate the variations into a smooth trend and the residual year-to-year fluctuations. The short-term variations are dominated by ENSO but also can be influenced by large tropical volcanic eruptions (such as occurred in 1963, 1982 and, markedly, 1991), so the years after those eruptions are anomalously cool. The figure here shows the relationship between those annual fluctuations and an ENSO index in the previous December and January. The volcano-affected years are left out. Using that relationship for 2016 predicts a warming above the long-term trend of 0.14 degrees Celsius and, when the short- and long-term predictions are combined, gives 1.16 degrees Celsius (plus or minus 0.13 degrees Celsius, with 95 percent confidence) net warming above the late-19th century baseline. This model suggests an 80 percent chance of a record high in 2016, without any global temperature data for this year being used at all.

schmidt-climate-1

We now have over half a year of data from 2016. The fact that the observations have a “memory” from month to month (because the ocean is slow to change temperature) allows us to predict the annual mean from the year-to-date average (which implicitly includes the ENSO effect). Starting in March (and every month since), this suggested that the 2016 net warming will be about 1.3 degrees Celsius above late-19th century temperatures.

Specifically, using data through July, I am predicting 1.25 degrees Celsius (plus or minus 0.09 degrees Celsius), and a better than 99 percent chance of a record.

We also have converging predictions of the ENSO state for the end of this year. The range of predicted values for December through February is 0 to -1.5 (neutral to cool conditions), and that suggests a 2017 net warming of 1.04 degrees Celsius (plus or minus 0.13 degrees Celsius). That would mean another top-ranked warm year, but not one likely to beat 2016.

schmidt-climate-2

There are additional sources of predictability for year-to-year time scales. As alluded to above, if a large tropical volcano erupts, it can have a predictable cooling effect for a year or two. Indeed, one of the first real-time predictions made by a climate model was for the cooling in 1992 and 1993 as a result of the Mount Pinatubo eruption in June 1991. Climate modeling groups have also been experimenting with ways to use the predictability of deeper ocean circulations (where internal variations can persist for up to a decade), but results have been mixed at best.

Given the dominance of ENSO on year-to-year variations, and the difficulty in predicting how ENSO will develop beyond six months, the prospects for useful year-to-year variations in global temperature beyond that is slim. It is only over the longer time scales (decades) that the additional predictability that comes from external drivers of climate change (for instance, carbon dioxide, air pollution and ozone depletion) can start to be useful — but that’s another post.

To summarize, some key climate statistics are easily predictable far beyond the scales at which weather forecasts are skillful. Those predictions clearly suggest an annual global temperature record in 2016 and a (relative) cooling in 2017, all while the long-term upward trends continue.

Gavin Schmidt is a climate scientist at the Earth Institute at Columbia
University and the director of the NASA Goddard Institute for Space
Studies.

Comments