Butterfly effect limits climate models

National Center for Atmospheric Research’s Clara Deser. Credit: NCAR

Natural chaos in our climate system creates uncertainty in predictions that can’t be removed, no matter how good scientists’ models get. Clara Deser from the US National Center for Atmospheric Research (NCAR) in Boulder, Colorado and her colleagues have shown these effects can be as strong as human-caused warming. “Over multiple decades intrinsic climate variability on a local and regional scale can be on a par with climate change due to greenhouse gas emissions,” she told me. “You’re not going to just see the result of the greenhouse gas increases – you’re going to see both. This simple message has been missing from the climate change literature.”

Climate scientists are working hard to improve the accuracy of their models’ predictions – perhaps so hard they haven’t yet looked at what their limits are. “We’ve been focussed on identifying how greenhouse gas changes and the like can affect the climate system,” Clara said.  “The uncertainties in climate projections have all been lumped together. There hasn’t been a set of runs that were designed the way that we have done them to really address this point.”

Anyone who’s had to run outside to rescue drying clothes from a rain shower knows that weather can be variable from day to day. Climate patterns also vary from year-to-year, like El Niño or the North Atlantic Oscillation, and some chaotic climate processes work over decades. Wanting to reduce model uncertainty, Clara previously tried to answer a series of detailed questions about these kinds of natural variability. Her team’s answers showed that they accounted for at least half of the disagreement between different climate model predictions. When she told this to two fellow climate scientists, Reto Knutti, from the Swiss Federal Institute of Technology, Zurich and Massachusetts Institute of Technology’s Susan Solomon, they were surprised. “They said, ‘Something very simple and illustrative is needed to get this important message across,’” Clara recalled.

Know your limits

Together with Clara’s NCAR colleague Adam Phillips, the three scientists set out to find how wide a range of climate predictions for North America natural variability can produce. They used one ‘fully-coupled climate model’ that simulates processes happening in the ocean, atmosphere, sea, ice and land in 40 slightly different runs. Clara’s team started all the tests with the model being run for 540 years with the same conditions to ensure they’re stable, then putting them through a simulation of the 20^th century. Those simulations involve conditions exactly as measured over this period, including energy reaching the Earth from the Sun and the levels of greenhouse gases in the air.

In each model, on January 1 2000, Clara and her colleagues made a small change in what was happening in our weather systems, switching them for real conditions seen on 40 different days at the end of 1999 and beginning of 2000. They then ran the models to 2060 in a commonly used ‘A1B’ future climate scenario, where CO2 concentrations grow from 380 ppm in 2000 to 570 ppm in 2060. In a paper published in the research journal Nature Climate Change last week, they found that on average, the different tests showed warming everywhere, but the runs were obviously different from each other.

a, December–January–February (DJF) temperature trends during 2005–2060. Top panel shows the average of the 40 model runs; middle panel shows the warmest model run for the mainland US, and the bottom is the coldest. b, DJF temperature for selected cities (marked by circles on maps), the mainland United States and the globe (land areas only). Black lines show observed records from 1910 to 2008; red lines show model projections for 2005–2060 from the warmest runs and blue lines the coolest, for each location or region. Dashed lines are best straight-line fit for their temperature change.

For example, the run with most warming over the mainland US in winter sees the northeast of the country warm the most, with a 4–6°C temperature rise, while Western Canada and Alaska warmed least, with temperatures rising by 1–3°C. Meanwhile, the coolest run over the mainland in winter generally shows warming of less than 1°C, with some of the north-west US getting colder.  Because all simulations are exactly the same until 2000, the differences can’t be put down to problems in the model, and show the uncertainty that can’t be removed.

We always have this intrinsic climate variability, Clara said, laid over changes from ‘forcings’ such as CO2 in the air. “It’s the butterfly effect,” she explained. “As greenhouse gas emissions get very large, say towards the end of this century, they can start to cause a forced climate response that will be larger than just the intrinsic variability. But we demonstrated in the paper that over the next 50 years intrinsic variability can be on a par with the forced climate signal.”

Journal reference:

Deser, C., Knutti, R., Solomon, S., & Phillips, A. (2012). Communication of the role of natural variability in future North American climate Nature Climate Change, 2 (11), 775-779 DOI: 10.1038/nclimate1562