No One Can Predict Future Climate, So Stop the Scaremongering
Two years ago, climate scientist Ed Hawkins created what he called “warming stripes” to demonstrate global warming “undeniably,” in the hopes of, “triggering a change of attitude that will lead to mass action." I created one below, using global average temperature data from 1850 – 2019 from Berkeley Earth (Land + Ocean anomaly using air temperature above sea ice:). Other major temperature datasets give similar pictures.
The idea is simple. The years go from 1850 on the left to 2019 on the right. Colder than median years are colored blue, the darker the blue, the colder the year. Warmer years are red, the darker the red, the warmer the year. Variants of these graphs are widely reported and the creator has won awards and praise for scientific communication.
But is it scientific? A fundamental scientific principle is control. Before you claim that data or a chart confirms your hypothesis, you have to see whether the same picture could arise if your hypothesis is false. Let’s assume there is no global warming, that annual temperatures are just a random walk. In that case, while the Earth is indeed warmer in 2019 than 1850, it was just the random workings that have always affected temperatures. The Earth is as likely to cool as to warm in the future.
I simulate this by starting temperature anomalies out at 0 in 1850, then picking a year at random from 1850 to 1917 (there was no net warming over this period and it is long before the sharp rise is atmospheric CO2 that began around 1950). I take whatever the temperature change was from the random year to the next, and add it to the zero in 1850 to get the simulated 1851 temperature. I repeat the process to build a temperature series up to 2019. In this way, I simulate a world without global warming, in which temperatures are drawn at random every year from the same pool of changes that actually occurred from 1850 to 1917, and in which there has been no change in climate due to human or nonhuman factors.
One such simulated warming stripe is shown below. It’s even scarier than the actual data. There is twice as much warming as actually occurred, and warming is accelerating rapidly. Global average temperatures had only mild variation before 1970, the big jumps afterwards were unprecedented in the historical record. Yet this is just the operation of a random walk. There is no real pattern, no real tendency for temperature increase, no change in behavior in the earlier or later times.
Now I admit this was one of the scarier simulations. Out of 10,000, it was the 99th percentile, meaning 9,900 of the simulations showed less concentrated warming. Below is the median simulation, the least scary of 10,000 (simulations below median show scary cooling). It does show net warming—that’s just a coincidence, the simulations near median are as likely to show cooling as warming—but it’s not steady over the period. Moreover it’s been cooling down since a moderate peak around 2005 and there’s no sign of temperature getting more volatile or trending. Even though there is no trend, there are long stretches of cool (blue) and red (warm) periods, due to the nature of random walks.
The question for a scientist is how many of the simulated series—the ones without global warming—are as scary or scarier than the actual chart. The answer is that the actual data are 89th percentile, meaning 11% of the time you get a scarier chart using simulated data without global warming than what we actually saw. Also 13% of the time the simulated data showed cooling that was scarier than the warming we have observed (that’s assuming a degree of cooling is as harmful as a degree of warming, although at least for small changes, cooling is much more harmful than warming).
Standard statistical reasoning is that if there is a high probability of observing something as extreme or more extreme than your data by random chance, then you should not trust any patterns you see. So far from being “undeniable” evidence of warming, the warming stripes are perfectly consistent with no warming at all, just a random walk that happened to turn upwards. Incidentally, we know global average temperatures are not a random walk, they have all sorts of patterns over periods of months, years and longer. But the point is if your chart cannot be distinguished from one generated by a random walk, then you have no reason to put any faith in the patterns you see in it.
Yes, the Earth is warmer in 2019 than 1850. Yes, there are many reasons to believe the warming—especially since 1980—is a trend likely to continue or accelerate in the future, and that humans are
responsible for around half of it to date (there’s a lot of uncertainty about that fraction—but it’s almost certainly a significant share and it could be the large majority, and it will likely increase in the future, also climate change is an even worse problem if humans are not responsible, because there’s less chance we can fix stuff that we didn’t cause).
But the warming stripes aren’t evidence of any of that. The warming stripes do not communicate science, they rely on a well-known human bias to see patterns in random walks. This is the same bias exploited by many investment charlatans, and it is dispiriting to see it win awards for science communication.
There are other problems with warming stripes. There is no scale. That means the stripes would look identical if the Earth had warmed 0.1 degree, 1 degree or 10 degrees. The creator boasts that all “superfluous information is removed,” but doesn’t explain why the amount of warming is superfluous to the issue of whether we should worry about global warming.
The author’s version of the chart makes some changes from mine to fit the story he wants to sell. My chart uses white for the median temperature years, and has equal amounts of red and blue. It suggests to the eye that there was reasonably steady warming throughout the period. The author chooses a different zero point which makes most of the chart blue, and suggests a “hockey stick” effect in the last 35 years. My graph has continuous variations in color, the author uses a palette of eight colors designed for maps. Maps emphasize the edges between regions, but there are no regions in temperature data. Using only eight colors makes the series look more jagged than it actually is and focuses attention on the warmest and coldest years.
A more technical problem is that the chart does not communicate the uncertainty around the numbers. It’s always misleading to show estimates without some indication of probable error. This is particularly relevant here because the precision of the estimates changes significantly over time.
Every scientist knows the importance of controls, and anyone experienced with quantitative data will look at the warming stripes and see they are consistent with a random walk. This is obvious by trained eye, you don’t need to do the simulations to demonstrate it. So it’s unlikely warming stripes were invented, publicized and praised by mistake. Presumably all the scientists who promote them know they are propaganda, not science (I have less faith that journalists and activists are aware of that).
Is that wrong? If you are a scientist convinced that the world should pay more attention to global warming is it okay to use the tools of advertisers, investment management touts and partisan activists? What is the line between artful presentation of data and deception?
I have enough trouble figuring out how I should behave, I don’t give advice to others. But I think misleading charts are dangerous even if they are consistent with scientific ethics. Using deceptive graphics to promote an oversimplified, alarmist version of global warming can backfire. The faithful you convert will not be prepared for the complexity that is sure to occur in the future, and when faith is betrayed it can turn, not to healthy skepticism, but to opposite faith.
Educating people is much harder and slower than fooling them, but it is a solid foundation for the future—and for all issues, not just global warming. Encouraging people to see patterns in random walks is as harmful as encouraging them to see patterns in the random scatter of stars in the sky, and basing their life decisions on astrology.
The danger extends well beyond the chart promoters. All scientists and statisticians concerned with global warming have to either remain silent about the lack of control and other problems with the charts, or speak out in ways that will give support to people who deny not just the presentation, but the data themselves. When scientists boast in public that a chart consistent with a random walk is “undeniable” evidence of change that should trigger “mass action,” honest scientists find themselves allied with pure deniers.
No one knows what the climate will look like in the future. Even people with a lot of trust in the models know there are model errors, important uncertainties about parameters and significant random noise. Education is our best chance of figuring out what needs to be done, and convincing people to do it. Scaring people by playing on biases today might help gain some short-term policy objectives but it’s a poor strategy for the long run.
