Spinning the climate wheel
“If you can look into the seeds of time and say which grain will grow and which will not, speak then to me.” Perhaps Banquo in Shakespeare’s Macbeth should first have asked about the reliability of the witches’ forecasts and then pinned down the details; oracular pronouncements often leave too much wriggle room in their interpretation. In the real world, of course, uncertainty rules. Edward Lorenz (1917-2008), eminent mathematician and meteorologist, famously highlighted the difficulty in making accurate predictions of the weather: sensitive dependence on the initial state of the system eventually destroys predictive skill as the forecast evolves; in a slightly frivolous example, he cited the minute disturbance created by flapping butterfly wings in Brazil conceivably influencing the later development of tornadoes in Texas (the ‘butterfly effect’, although the original, less poetic, metaphor referred to seagulls). As we can never hope to capture such initial detail, the seeds of uncertainty are forever present and grow with the forecast; the atmosphere is a chaotic system. Lorenz’s work applies not just to meteorology but to many other complex systems that we have an interest in forecasting. The global economy, for example, probably falls into this category; the pre-crash certainty that global markets were highly resilient and self-correcting echoes the dreams, shattered by Lorenz more than 45 years ago, that we might one day have accurate weather forecasts for weeks or months ahead.
On this basis any hope of accurately forecasting the climate for the coming decades would seem to be misplaced. However, progress is possible if we sacrifice detail and enquire only about average conditions. If we fire up our climate model from today and run it out to end of the century, Lorenz’s work guarantees that after about a week each daily forecast will, in isolation, be quite useless, devoid of skill. However, the average forecast or count of extreme events, taken over a period of typically 30 years, should have some skill as the model is based on fundamental laws on science. This is indeed confirmed when the model is run for the past climate where the output can be verified against the observational record. Will it rain in Limerick on 1 December 2050? We will have to wait and see. But we would hope to be able to answer now whether the average (2031-2060) winter rainfall in Munster is likely to differ appreciably from the average for our current reference period (1961-1990).
However, uncertainty is always lurking in the background. The model, no matter how complex, is only an approximation to the huge suite of interlinked processes that make up the climate system. It turns out that the accuracy of predictions is dominated by model errors in the first few decades and by uncertainties in future Greenhouse Gas (GHG) emissions in the second half of this century.
Unfortunately, the uncertainty in climate projections is not uniform across the weather elements. Warming trends in the climate are generally robust but there is large uncertainty in future rainfall patterns, particularly at a regional scale where it matters. Even with historical data it is notoriously difficult to disentangle possible climate-change induced trends in rainfall from natural climate variability; the background noise simply swamps any signal unless the rainfall record is sufficiently long. For extreme rainfall events the situation is even bleaker given our relatively short observational records. Gut feelings and anecdotal ‘worst in living memory’ records are no substitute for a careful statistical analysis of accurate climate observations maintained by Met Éireann. Is the recent (November 2009) spell of exceptionally wet weather part of a trend, a signal of climate change? We simply don’t know. Forecasting future trends in regional rainfall is even more challenging.
The comment by a statistician that all models are wrong, but some are useful, applies to climate models. Some are more complete than others in representing the relevant climate processes but all compromise to some extent in the internal description of physical processes. A huge program of international research is focussed on improving existing models or developing new ones. Given the uncertainty in climate modelling, the development of new models is to be welcomed as it lessens the likelihood that we are underestimating the uncertainty or missing out on important climate features.
The recent establishment of the EC-EARTH consortium is a good example of progress in this area. A consortium of 19 research institutions from 10 European countries, EC-EARTH was established in 2007 to develop a new global climate model - an Earth System Model - that embraces climate processes in the atmosphere, ocean, land and biosphere. Ireland is strongly represented in the consortium as Met Éireann, ICHEC (the Irish Centre for High End Computing) and UCD are research partners.
The EC-EARTH model has an interesting pedigree. The atmospheric component came from the European Centre for Medium-Range Weather Forecasts (ECMWF) which is regarded as a world leader in atmospheric modelling; if the name does not register you will certainly have come across its forecasts since they form part of the weekly outlook broadcast daily by Met Éireann (Ireland is also one of 18 Member States of ECMWF). The linkage between operational weather forecasting and climate modelling is particularly attractive as the former is constantly being developed to improve the forecasts and these improvements feed into the climate model.
The IPCC plans to produce a new Assessment Report (AR5) in 2013/14 that will bring together the latest knowledge concerning climate change and future prospects for the Earth’s climate. Many international groups involved in climate modelling are currently in the process of running, or are planning to run, global climate models to provide basic information that will underpin AR5. The focal point for this work is the Coupled Model Intercomparison Project (CMIP), established under the World Climate Research Programme (WCRP). CMIP provides the experimental protocols for the simulations.
In support of AR5, and to improve our own forecasts of regional climate change in Ireland, the EC-EARTH model is currently being ‘spun-up’ on the ICHEC supercomputer (appropriately named after the Sligo man Gabriel Stokes whose pioneering work in fluid dynamics in the 19th century made EC-EARTH possible). This pre-industrial climate simulation will run for 500-1000 years, a necessary lengthy period as the ocean is sluggish, particularly in the lower depths, and takes a long time to adjust to the driving influences imposed by the model. At the end of the period the system should be reasonably ‘settled’ and ready for validation in the modern era when GHG concentrations began to rise (1850-2005). The really interesting bit comes when we run the model out to the end of this century using different GHG scenarios for the future (low to high emissions); it will link GHG mitigation options we choose today with the future consequences.
We cannot eliminate the uncertainty in the climate predictions but to some extent we can box it in by launching our model from slightly different initial conditions; the spread in the resulting ensemble of forecasts is a measure of the uncertainty. Concentrating on the next 10-30 years an ensemble of simulations will be carried out with EC-EARTH. The full set of data will be made available, both locally and internationally, for scientific use.
Of course, our main interest is what will happen to the Irish climate. The EC-EARTH outputs will be further refined by a process known as ‘downscaling’ to tease out the local details, spawning a host of applications. It is intended that the outputs will be available to support impact studies in climate sensitive economic sectors such as Energy, Insurance, Water Resources, Forestry, Agriculture and Fisheries. In addition, the data will be used to examine the impacts of climate change on extreme weather associated with flooding events. The autonomy in running our own simulations, and in having access to simulations produced by our international partners, will greatly facilitate this process.
Because the EC-EARTH model is global in extent the data will feed into international programmes such as the WCRP-sponsored CORDEX program to support regional climate studies; CORDEX, for example, will initially focus on Africa, providing much needed information in support of local adaptation policies to tackle climate change.
In spite of our efforts, global climate models such as EC-EARTH are not exploited to their full potential. The atmospheric component, for example, resolves details at a scale of about 16 km when used for daily weather forecasts that extend typically 10 days ahead. It is currently impossible to run EC-EARTH at this resolution as the computer requirements would be enormous for simulations stretching to the equivalent of hundreds of years; instead, the resolution is throttled back, providing a more blurred picture, to match the available resources. It raises an interesting question: should we run many simulations at coarse resolution to achieve a large ensemble, or should we focus on fewer runs at high resolution? There are arguments in favour of both approaches. Either way, it emphasises the central role and importance of facilities such as ICHEC in supporting national and international efforts to tackle climate change.
Can we ‘look into the seeds of time and say which grain will grow and which will not’? At least for the climate we should soon have a better picture thanks to the EC-EARTH model.
© Met Éireann.
Autumn Floods, by Bernie Delaney: - A sign of things to come?