Why climate projections are different

Numerical prediction of climate is a different problem, even though it starts with the same equations governing atmospheric motion and continuity of matter (the amount of air and water in the atmosphere). There are two main differences:

• First, climate projections are not about predicting the exact weather at any time in the future, but rather about projecting the statistics (average behaviour and variability) of the future weather. This reduces the relevance of short-term chaotic behaviour in the atmosphere.

• Second, because climate projections are about the statistics of weather many months, years or even centuries into the future, much slower influences on the weather or climate must be taken into account.

Thus the propagation of errors in initial conditions is not important, but rather the knowledge of slower internal variations in the climate system, and so-called 'boundary conditions'. Slower internal variations include exchanges of heat, salt and chemicals such as carbon dioxide with the deep ocean, and the growth and decay of ice sheets and glaciers. Figure 11 shows some of the components that are included in the internal part of a climate model. Others, not shown, include interactive soil and vegetation, atmospheric chemistry, and cloud interactions with particles (aerosols).

External factors include changes in atmospheric and land surface properties, variations in the orbit of the Earth around the Sun, solar variability and volcanic eruptions (which put gases and particles into the atmosphere). Some of these external factors can be specified as inputs or boundary conditions to the climate models, rather than calculated. However, they can be calculated internally in an enlarged climate system, if they can be predicted and the equations can be added to the model. As computers become bigger, faster and cheaper, more and more that used to be external to the climate models can now be incorporated into the models and thus predicted rather than given as external boundary conditions.

How good are climate models?

Two or three decades ago most climate models considered the oceans to be external and used prescribed sea surface temperatures at the bottom of the atmosphere. Results were conditional on the assumed sea surface temperatures. Today, faster computers enable nearly all climate models to have an interactive ocean, and indeed deep water temperatures are calculated, as are ocean currents. Historically, the scientific literature is full of papers describing simulations of climate using models of varying complexity and detail, and it is important in reading these papers to understand just what is calculated and what is given as assumed input. It is also important not to rely on outdated assessments of the skill of climate models, which tends to occur in the critiques of climate modelling by some who question climate change. Citing out-of-date assessments of climate models ignores recent improvements in modelling and shows a lack of appreciation of how much work has gone into improving their accuracy.

A range of different types of climate models are available, and various names or abbreviations are used to indicate differences in their complexity. For example, there are very simple 'energy balance models' which calculate only the incoming and outgoing energy into the climate system to determine the Earth's global average surface temperature, or that in latitude bands. These ignore lots of internal processes and do not give outputs useful at particular locations on Earth, but they are quick and cheap to use and are often used to study a wide range of conditions.

Next there are 'atmospheric general circulation models' or AGCMs. These calculate what goes on in the atmosphere, including changes to cloud cover and properties, but have prescribed or given sea surface temperatures. Thus, they do not allow for changes in the ocean climate (currents, and temperature and salinity with location and depth).



Figure 11: Some internal components of a climate model. Some of these components can be changed by external forces (such as variations in the Earth's orbit around the Sun, or in solar radiation), or changes in the composition and radiative properties of the atmosphere (such as the addition of more greenhouse gases or particles of dust). (After John Mitchell, UK Meteorological Office, 2003.)


Figure 11: Some internal components of a climate model. Some of these components can be changed by external forces (such as variations in the Earth's orbit around the Sun, or in solar radiation), or changes in the composition and radiative properties of the atmosphere (such as the addition of more greenhouse gases or particles of dust). (After John Mitchell, UK Meteorological Office, 2003.)

Today nearly all climate models used for climate projections have fully interactive oceans. These are called 'coupled atmosphere-ocean general circulation models' or AOGCMs, and usually include calculations of sea-ice cover. Even so, there are still many external components of the climate system that are only gradually being internalised into climate models, even where these components act on, and are changed by the climate. These components include glaciers, continental ice sheets, and surface properties determined by vegetation.

Climate models have been tested and improved quite systematically over time. There are many ways of doing this. One is to closely compare simulated present climates with observations. Climate modellers often judge models by how well they do in reproducing observations, but until recently this has mainly been by testing outputs from models against observations for simple variables such as surface temperature and rainfall. However, this process can be circular in that climate models, with all their simplifications (for example in how they represent complex processes like cumulus convection, sea-ice distribution or air-sea exchange of heat and moisture) can be adjusted, or 'tuned', to give the right answers, sometimes by making compensating errors. Such errors might then lead to serious differences from reality in some other variable not included in the tests. Comparing simulated outputs for many more variables, some of which were not used to tune the models, now checks this.

Other tests used include how well the climate models simulate variations in climate over the daily cycle, for example daily maximum and minimum temperatures, or depth of the well-mixed surface layer of the atmosphere. Changes in average cloud cover and rainfall with time of day are other more sophisticated variables that are sometimes tested.

Related tests involve calculating in the models variables that can be compared with satellite observations, such as cloud cover and energy radiated back to space from the top of the atmosphere. Until recently many climate models have not done very well on some of these tests, but they are improving.

To test longer time-scale variations, tests are made of how well climate models simulate the annual cycle of the seasons. Different test locations from those the model builders may have looked at when building their models are often used. For example, how well does an Australian climate model perform over Europe, or a European model perform over Africa?

A popular test is to use a climate model with observed boundary layer conditions, for example sea-surface temperatures in an atmospheric global climate model, to simulate year-to-year variations such as a year with a strong monsoon over India versus a year with a weak monsoon. Similarly, tests are made of how well a climate model reproduces the natural variations in a complex weather pattern such as the El Nino-Southern Oscillation (ENSO), which is important in year-to-year variations in climate. ENSO is a variation in oceanic and atmospheric circulation, mainly across the tropical Pacific Ocean, but has effects in many other parts of the world. Getting ENSO right is an important test, and it is only recently that some AOGCMs have done well with this test.

At even longer time scales, tests can be made as to how well climate models can reproduce paleo-climatic variations. This is only possible where changes in external conditions can be well specified, such as changes in solar energy input, atmospheric composition, land-sea distribution and surface properties. It is also necessary to have lots of paleo-evidence for climate patterns at the time being simulated to see if the climate models reproduce it well. This is a tall order, but nevertheless paleo-modelling is useful as a test of climate models, and also helps us to understand and test theories of what cause climate fluctuations and what is possible.

In order to provide climate change projections relevant to many local and regional climate change impacts, climate models need to provide output at finer and finer spatial scales. That is, where global climate models a decade back only gave output data on climate changes at distances several hundred kilometres apart, for many purposes the need is for data at locations only a few tens of kilometres apart. The limitation was essentially computer capacity, since the number of calculations increases roughly by a factor of eight for every halving of the distance between data points.

There are three ways in which this finer spatial resolution can be achieved:

• running global climate models at finer and finer spatial scales,

• statistical downscaling,

• running local or regional climate models driven by output from global models.

Rapid improvements in computer speed and capacity have enabled global climate models to be run at finer and finer spatial resolutions. Some models now routinely produce output at spatial scales as fine as 100 or even 50 kilometres, although this still cannot be done for many different scenarios. There has also been a technical development using variable spatial resolution in global models, whereby it is possible to run a global model with coarse resolution over most of the globe, but fine resolution over an area of interest. This latter option is fine if you only want detailed information about one region, for example if you are in a national laboratory modelling for information relevant to one region, such as the UK or Japan. However, while it is important to get detailed information for your own region, many countries will also want to know what may happen in detail in other parts of the world, at least for broad policy reasons, and maybe even for telling what the impacts of climate change may be on trade partners and competitors.

The need for truly global simulations at fine spatial scales thus remains important. Japan has recognised this, and has built the Earth Simulator supercomputer, capable of modelling the climate at fine scales for the whole globe. It contains the equivalent of many hundreds of ordinary supercomputers (circa 2004), and currently is running a climate model with 100 levels and a horizontal resolution of 10 kilometres, compared to most AOGCMs that have a resolution of around 100 or more kilometres. So far they have only carried out short simulations.

The second way of getting finer spatial detail for particular locations is to use statistical relationships between the observed large-scale climate patterns and local climate to derive estimates of local changes from model-simulated large-scale changes. This is called statistical downscaling, and requires a lot of detailed climate observations for the region of interest. It is also important that the statistical relationships between the large-scale changes and local change will be valid under conditions of climate change as opposed to present observed climate variability. That is not the case with all large-scale variables and must be tested.

The third method for obtaining detailed local output is to use a local or regional fine-scale climate model driven by the output of a global coarse-resolution model. The easy way to do this is to use output from the global model at the boundaries of the regional model domain to determine the model values at the boundaries of the regional model, and force the regional model to adjust its values inside the boundaries to be consistent with this. This is termed 'one-way nesting'.

However, local changes within the region may in reality force changes at a larger scale, for example if the region includes a large lake from which the atmosphere may pick up additional moisture. To account for this possibility, ideally the output from the regional model should be fed back into the global model and thus modify the large-scale climate. This is termed 'two-way nesting'. Many climate-modelling groups have performed one-way nesting, but so far two-way nesting is less common, and has revealed sometimes-significant differences in results for the same region.

The performance of fine spatial resolution modelling has also been carefully tested by comparing different models over the same regions, and by trying to reproduce particular historical situations using regional models forced at their boundaries by observations. Results have been mixed, and in general it is conceded that regional detail is not as reliable as the large-scale output. This applies particularly to rainfall patterns, although regional detail is necessary especially for rainfall because it can vary greatly over small distances due to topography and land-sea boundaries.

Overall, model performance and verification is complex, but is being actively tested and improved. Climate models provide projections that are far more sophisticated and reliable than simple extrapolations from observed climate trends. Moreover, the IPCC and other bodies studying climate change have taken the uncertainties into account. These uncertainties are being progressively decreased to provide more reliable and policy-relevant information.

The IPCC 2001 report summarises its conclusions regarding the state of coupled atmosphere-ocean climate models as follows:

Coupled models have evolved and improved significantly since [1995], In general, they provide credible simulations of climate, at least down to sub-continental scales and over temporal scales from seasonal to decadal, The varying sets of strengths and weaknesses that models display lead us to conclude that no single model can be considered 'best' and it is important to utilise results from a range of coupled models, We consider coupled models, as a class, to be suitable tools to provide useful projections offuture climates,

0 0


  • Kalervo
    Why different climate change projections?
    15 days ago

Post a comment