What if the climate were predictable?
Butterflies, and what I do
Guest Post by Ve Balaji
Ve Balaji is a dear old friend. Professionally he works at the intersection of climate change and software. At a personal level, he supplements these with deep social concerns.
This post was presented as a paper at the Oxfam Online Conference on Disaster Risk Reduction and Climate Change Adaptation, February 2009.
A bus queue in the pouring rain, and only one man without an umbrella. He turns to the man in front, and says, “Yes, I work for the Weather Bureau. How did you know?” A classic cartoon of R.K Laxman, whose laconic single-frame still-lifes of modern civilisation reflected in the bemused gaze of the Common Man appeared on the front page of The Times of India for over fifty years, dates from before I entered the weather business, and probably had nothing to do with it. And yes, many people to whom I explain what I do are still amused to trot out some variant of this remark.
Well, why is it so damnably hard to predict the weather? There is in fact an answer, and it’s an illuminating episode of modern science. the clearest answer was given by E.N. Lorenz in 1963, in a classic paper that gave birth to chaos theory. First look at this beautiful image, known as a Lorenz attractor on the right.
Lorenz showed that the equations of weather constituted a strange attractor. While going round and round in circles, the fly will suddenly, and unpredictably, move and spin around a different centre of attraction. It’s as though there are two sources sending out seductive signals: the fly stays in the vicinity of one, but its trajectory every now and then sends it into the orbit of the other.
Now can you predict the trajectory of the fly, if you know its initial condition? Look at the Lorenz attractor: you can see some points very very close to each other that actually end up in different attractors. Or one might cycle around one attractor 10 times before skipping to the other, while another nearby trajectory skip after just 1 or 2.
That makes an actual forecast of the trajectory very difficult, as you may not know the initial condition to great accuracy. Lorenz calculated the limits of predictability for weather, and showed that it’s impossible to predict the atmospheric weather beyond 12 days or so, even though you perfectly understand the physics that causes it! Lorenz, and chaos theory, drove that wedge between understanding and prediction that we are still grappling with today. “It is as though,” Lorenz said, “that a flap of a butterfly’s wing in Brazil will, a few days later, cause a tornado in Texas.” This led to the popularization of the “butterfly effect”, a phrase that also recalls the elegant contours of the Lorenz attractor.
If you can’t even predict the weather, what makes you think you can predict the climate?
The phrase “butterfly effect” actually has a longer history. Lorenz himself attributed the origin of the remark to an informal conversation with Joe Smagorinsky, founding director of the Geophysical Fluid Dynamics Laboratory in Princeton. A lot of discussions about origins tend to buzz around this lab, and in fact take us back to the very dawn of digital computing. John von Neumann, who more or less invented modern computing at the Institute for Advanced Study, also in Princeton, in fact suggested that weather and climate would be two principal applications of this new gadget, and suggested that two centres be formed to start these efforts. The team that Smagorinsky assembled was responsible for many firsts: the first paper to simulate, and give a correct physical accounting of the effect of increasing CO2 concentrations on the planet’s temperature; the first coupled ocean-atmosphere model used to show how the entire system is needed to model the climate; etc. In a minute, we’ll get to the importance of the ocean in the scheme of things.
If weather cannot be predicted beyond a couple of weeks per Lorenz, why should we even think about the longer term? Climate, the saying goes, is what you expect, weather is what you get. While the day to day fluctuations of weather may indeed be unpredictable, there are slower changes that are quite predictable. The changing of the seasons, the ebb and flow of the monsoon. I may not be able to tell you what the weather will be like on the 13th of August, 6 months from today: but I can predict with some confidence that August is likely to be cooler than February, if you live in the temperate zones of Chile or Australia. That has to do with the Earth’s orbit and the tilt of its axis to the orbital plane, and not dependent at all on today’s weather. We now begin to distinguish between two ways in which things change: forced changes caused by events external to the climate system, and internally-driven change, in which the future state of the planet is dependent upon its state now, the “initial conditions”. It’s the initial-condition driven portion of the state that is chaotic, hard to predict.
We’ve seen that the system is chaotic on the scale of days, of weather; what about on longer time scales, seasons, years, centuries? Indeed it is: consider that when year follows year, summer is indeed warmer than winter, the monsoons arrive and recede. But the trajectory is never repeated exactly the same way twice, as in the Lorenz attractor: we’re back among the butterflies. What causes changes from year to year? It turns out that compared to the capricious atmosphere, the ocean is a ponderous beast, retaining a memory of its state over years and even millennia. And it too may be capable of shifting abruptly between different states. By now the phenomenon of El Nino is well-known: as George Philander documents, it was once the fashion to attribute everything from hemlines to harvests to El Nino. This phenomenon, where the equatorial Pacific abruptly (by oceanic standards) shifts from one state to another, can indeed affect the Indian monsoon, and weather patterns worldwide. It was in attempting to find an explanation for the failures of the monsoon and the resulting terrible droughts over India that El Nino was in fact first noted, a history told in depth by Mike Davis.
El Nino and the Southern Oscillation (ENSO) shift between their equilibria every few years. We are beginning now to understand other, even longer-term modes of variability of the climate system. These become progressively harder to disentangle from forced climate change caused by human activities, our warming of the planet by adding greenhouse gases, other effects of industrial pollutants, as well as natural forcing agents such as volcanoes and variations in the Sun’s energy, over which we have no control.
The Language of the dice
So we have a system that is chaotic, and thus with inherent limits to its predictability on time scales of days as well as decades. What might it mean to issue a forecast, at any time scale? The answer is that all such forecasts are by their very nature probabilistic, they assign a certain risk of a particular outcome. In the case of weather, we run our weather models, issue forecasts, assess the skill of the model (how often they issued a correct forecast) and continually refine them. The models are initialized with observational data from satellites and weather stations. We account for errors and uncertainties in these initial estimates by running an ensemble of models with a variety of initial conditions set up to mimic the uncertainty in the initial state. In some the coin flips, the butterfly flaps, and there is a tornado in Texas, and in others there is none. We can count up the outcomes and assign a risk to the outcome.
Climate change scenarios, such as those coming from the IPCC, similarly speak the language of the dice. We cannot tell you for certain with the best of our models what the outcome in any particular year will be of continually loading the atmosphere with greenhouse gases. But we can tell you that the dice have now been loaded in favour of hotter summers here, wetter winters there, villages sinking into the permafrost in Alaska while bushfires burn unchecked in Australia.
What if we could predict the climate?
Let’s now assume that our climate models have become very good indeed. (As indeed they have… in the last IPCC, many models were able to input all the external forcing agents of our recent history, the industrial emissions, the volcanoes, the solar variability, the changes to land use; and issue perfectly satisfactory accounts of the known climate record up to the end of the 20th century). We now use these models to make some bold assertions about what will happen in the 21st century. What should we expect to happen next? In particular, what do we expect to happen in marginal economies, in the most vulnerable, in the global South? I expect the North to take care of itself, one way or another.
As described above, the “bold assertions” will be in the form of statements about risk. A lot of current debate in the climate community is about finding an idiom for correctly communicating risk and probability.
Let’s assume we find it. What then? Let’s start by looking at how we’ve dealt with forecasts of risk of dire outcomes at various time scales. At short time scales, the classic example would be forecasts of tropical storms making landfall. Even very recent history shows cause for much pessimism: take Hurricane Gustav’s devastating passage through the islands of the Caribbean. Haiti is still picking up the pieces. Cuba, however, survived without the loss of a single life: while this is described as miraculous in the mainstream media, it was clearly anything but an act of God: Cuba’s civil defense against natural disasters would be the envy of many countries richer many times over. Ask any Katrina survivor, still dealing with the deadly double- barrelled weapon of official incompetence and indifference.
At time scales of a year or so, the case to look at would be forecasts of drought and monsoon failure. Again, our collective history here is not admirable. Late Victorian Holocausts, the book by Mike Davis, unsparingly documents the British record in dealing with predictions of drought in India. Davis convincingly decouples the natural phenomenon of drought from the human phenomenon of famine. The best possible science and the most skilled forecasts are nothing in the face of official callousness. More recently, there is an egregious case of a forecast actually contributing to causing a famine, documented by Philander. Based on a robust forecast of an El Nino event in 1997-1998, “expert advisers” incorrectly issued warnings of drought in Zimbabwe. This led to under-planting during the growing season, with many banks refusing loans to farmers. It turned out to be a normal rainfall year: no drought, yet there was widespread hunger due to a poor harvest.
Is there anything in this history to inspire confidence how we will deal with longer term risk reports? Let us say we assign a certain level of risk to a prolonged drought in the Sahel in the 21st century, as indeed some of us do. What shall we do with this information?
Let’s not be pessimistic. I think we collectively realize it is necessary to be Cuba and not Haiti in the path of Gustav. The will is there, especially in gatherings such as this. As is the expertise. If anyone inspires confidence that we will build the institutions necessary to help the South tackle climate change, it is you all.