climate modelling

Climate Models
Providing quantitative information on future climates is a massively complex and uncertain process which began, using numerical models of the earth-atmosphere-ocean system, over 40 years ago.  These global models (general circulation models or GCMs) have become increasingly sophisticated and require some of the most powerful computers in the world to run on. They now operate at a grid size of some 300km and represent all of the important physical processes in the atmosphere, oceans and on the land surface, including sub-models of radiation, cloud, thermodynamics, precipitation (snow and rainfall) and heat transport.  The parameters of these sub-models are given values which are in many cases poorly known, and so there is considerable uncertainty and error in the outputs from these GCMs. 

The performance of GCMs is assessed by how well they can reproduce the observed climate. Whilst they give impressive results in simulating the large-scale atmospheric and oceanic circulations and the average climate, there are well-known shortcomings when their estimates of variability and extremes (e.g. heavy rainfall, droughts and wind storms) are assessed.  This inadequacy is partly due to their coarse grid size, so a strategy of downscaling has been developed using regional climate models (RCMs) covering a limited area domain (e.g. Europe) nested within a GCM and using the GCM grid variables as boundary conditions to drive the smaller model. Such RCMs typically operate at 50km resolution, and so can resolve and represent the important effects of topography (mountains and land-sea effects) in, for example, determining surface temperature and rainfall patterns.   

More information on different model types can be found if you click here.

Future Climates
GCMs are validated using simulations of the present day climate, generally using the period 1961-1990 as a standard “baseline” against which future conditions will be compared. It should be recognised however, that there has been a steady warming (and possible changes in other weather variables) throughout the latter half of the 20th century, and so this period (and any other 30-year period) is subject to possible trends in climate on top of the considerable natural variability from year to year.  The recent Hadley Centre report on The Climate of the United Kingdom and Recent Trends contains further information.

Future projections are made by perturbing the GCM with a steadily increasing greenhouse gas radiative forcing, i.e. the global warming effect of increasing CO2 emissions.   There is considerable uncertainty in the estimates of how these emissions will change over the next 100 years or so, and emissions scenarios have therefore been devised, typically ranging from low (assuming that major reductions are made) through medium-low, medium-high (assuming “business as usual”) to high (assuming further economic growth and no controls).  These emissions scenarios result in GCM projections following transient trajectories of increasing temperature from present day out to 2100.  The results are generally available as averages across 30-year periods (or time-slices), so for example the 2020s covers the period 2011 to 2040, the 2050s the period 2040 to 2070, and the 2080s the period 2070 to 2100.

UKCIP02 Scenarios
In 2002, the UK Climate Impacts Programme (UKCIP) published Hadley Centre climate scenarios from a global model (HadCM3) used to drive a regional model (HadRM3). These were termed the UKCIP02 scenarios and have been used since as a standard basis for climate impact assessment.  Projections of average climate for a wide range of climate variables were made available across the UK at a 50 km resolution and were a major advance in providing consistent resources for impact studies at that time.  The UKCIP02 Scenarios Gateway provides further information on this subject.

The next suite of UK Climate Projections (known as UKCP09) is being prepared in a DEFRA project by a team including the Hadley Centre, UKCIP, British Atmospheric Data Centre, Newcastle University and University of East Anglia.  See below for a discussion of some of the differences between UKCIP02 and these new UKCP09 scenarios.

Downscaling for Impact Assessments – the Need for a Weather Generator
Although RCM outputs at 50km resolution provide a major advance over GCM outputs, the daily outputs or long-term means available from the UKCIP02 scenarios are far from ideal for regional, sub-regional and local scale impact assessments, where realistic time series of weather variables (e.g. rainfall and temperature) are required which are representative of a catchment or location and contain realistic variability and extremes (e.g. for flood and drought considerations).   Other methods of downscaling have therefore been developed including “weather generators” which are statistical computer models which are “calibrated” using observed weather data and can then reproduce time series of weather variables at hourly or daily resolution for any desired location.   Such Weather Generator (WG) outputs are “stochastic” in that it is the overall statistical properties of the generated series which are matched up to observations, and no specific calendar date should be attached to the days in the series.

The Environment Agency Rainfall and Weather Impacts Generator (EARWIG) was developed for the Environment Agency by Newcastle University and University of East Anglia.  EARWIG produces internally consistent series of meteorological variables including: rainfall, temperature, humidity, wind and sunshine.  The system produces series at a daily/hourly time resolution, using two stochastic models in series: first, for rainfall which produces an output series which is then used for a second model generating the other variables dependent on rainfall. The series are intended for single sites defined nationally across the UK at a 5 km resolution.   Future scenarios are generated by fitting the models to observations which have been perturbed by application of change factors derived from the UKCIP02 mean projected changes in that variable.

Further technical information on EARWIG can be downloaded here.

Application of EARWIG
EARWIG has been widely used for studies across the UK in the water sector (see Kilsby et al., 2007 for a full description) and has recently been used on the following impact and adaptation studies in the north east region:

 

It was also used on the North East Climate Change Adaptation Study, which is the earlier analogue to the present Yorkshire and Humber study.

EARWIG is also being further developed for contributions in generating the UKCP09 scenarios, where a number of time series of weather variables are being made available directly from the RCM outputs. 

In the present study, EARWIG has been used to generate rainfall and weather projections for a single future time-slice (2050s) and under single emissions scenario (medium high).  This choice of time-slice and emissions scenario has important consequences for both the range and magnitude of impacts and the suitability of the recommended adaptation responses.  Modelling the climate is a complex process and one that involves many uncertainties.  The outputs from models are consciously labelled as ‘projections’ (of what might plausibly happen) because they are not intended to be ‘predictions’ (of what will definitely happen).  The results from a climate model, as with any model, are dependent on the quality of the input and the assumptions used in the process.  As one moves down through the different stages of climate modelling, the results obtained are more sensitive to these assumptions.  This is often called the ‘cascade of uncertainty’ as different visions of the social and economic future of the world, different global and regional climate models, different downscaling techniques and different ways of thinking about impacts will all increase the uncertainty of results at the local scale.  For many entirely justifiable reasons, this particular study has chosen one particular route through this cascade.  However, it is important to recognise that this one route cannot, and does not try to, encompass the total range of uncertainty.     

Smaller or larger changes in climate and impacts would result from using the earlier or later time-slices (i.e. the 2020s and the 2080s, respectively), but we believe the 2050s period is most relevant to long-term planning and adaptation strategies since it coincides with the time-horizons of many buildings and infrastructure. 

Similarly, impacts with greater magnitude would result from using the high emissions scenario, but the medium high scenario provides a plausible indication of the ‘business as usual’ situation and therefore is expected to be the most useful output for this study.  It is also the most studied scenario under UKCIP02 as more RCM simulations are available, and a linear scaling approach from this scenario is generally used to estimate changes for other combinations of time-slice and emission scenarios. 

There are major advantages of using EARWIG in the present study in addition to the UKCIP02 outputs.  Whilst UKCIP02 output still provides useful information on mean annual and seasonal projections across the entire region, it does so at 50km scale of resolution.  Through use of EARWIG we can enhance this and generating projections at 100 times improved resolution, downscaling to the 5km scale.

The figures below show the differences in model scale across the region between UKCIP02 and EAWRIG.

EARWIG also offers the advantage over UKCIP02 projections of being able to assess extreme values in different parameters as well as the means.  In this study, EARWIG has been used to make projections of relevant rainfall and weather parameters at eight locations across the region. 

These locations have been selected to represent coastal, inland and upland locations in order to identify subtleties in future projections due to altitude or proximity to the coast.  These locations cover both urban and rural areas and also represent a good geographical distribution across the region. 

Confidence in Variables
Due to the nature of the processes being simulated in climate models, and the large natural variability in some weather parameters, climate modellers attach very different reliability to estimates of change in different parameters.  Broadly, there is most confidence in long term or seasonal averages of quantities (e.g. summer mean temperatures), and least in estimates of extremes or short time resolution variables (e.g. hourly extreme rainfall).  

The UKCIP02 Scientific Report outlines the reliability for different weather variables, with most confidence cited in mean temperature, followed by mean rainfall. Projections of wind, especially, are less reliable due to the highly variable nature of wind gusts, and dependence on land surface roughness. 

Looking Towards UKCP09
The climate model projections used as the basis for the present study are predicated on the national UKCIP02 data set and then downscaled to the 5km resolution using EARWIG for the purposes of regional, sub-regional and local impacts and adaptation assessments. 

The UKCIP02 data set has been the standard source since 2002.  However, the UKCIP02 scenarios will be replaced in 2009 by a new set of scenarios known as UKCP09 which will provide comprehensive information on projected climate changes and their probabilities, maps accessible through a User Interface, and a Weather Generator for providing weather time series.  These projections differ fundamentally from UKCIP02 in that they are “probabilistic”: by this we mean that they do not provide single estimates (or best guess information) of future conditions, but rather they allow for our uncertain knowledge and imperfect models. They do this by incorporating future estimates of change from a wide range (or ensemble) of models from the Met Office Hadley Centre as well as other modelling centres worldwide.  By contrast, the UKCIP02 scenarios were based entirely on one set-up of one model from the Met Office’s Hadley Centre, namely the regional climate model HadRM3, driven by the global model HadCM3. 

Despite this difference in approach, there also remains considerable continuity and similarities between UKCIP02 and UKCP09.  In particular:

• the 1961-1990 baseline is retained;

• there will be three emissions scenarios used (low, medium, high) rather than the four in UKCIP02 (low, medium-low, medium high and high) but these will span essentially the same range of emissions;

• the 3 time slices used in UKCIP02 will be available as part of an overlapping range of 30-year  periods from 2010 to 2100.

The HadCM3 and HadRM3 models also play a major role in the UKCP09 scenarios, as does EARWIG.  Comparisons between the two data sets are difficult however, because whereas UKCIP02 provided essentially one set of changes for a given emissions scenario and time slice (e.g. 2050s medium-high), UKCP09 will provide a range of changes of different probability of occurrence, as a probability distribution.  

This probability distribution is derived by running a large ensemble of climate models with different parameters. These parameters describe processes in the model which are poorly understood so that there is significant uncertainty in the values of the parameters. Therefore, models are run using ranges of the parameter values, rather than a single well-known value, and the resultant changes in e.g. temperature or rainfall also cover ranges reflecting this level of uncertainty.  The proposed strategy for using the UKCP09 scenarios is therefore not to use a single “mean” estimate of change as is essentially done using UKCIP02, but rather to carry out impact assessments using a number of estimates, e.g. 10 percentile lowest change in temperature, 20%, 50% (or most likely), and then up to say 90 percentile highest change. This has great advantages in allowing risk-based assessments to be performed accounting for model uncertainty, but introduces a major overhead in increasing the number of assessments or simulations of impact by an order of magnitude.  

A key issue is how much difference there will be between scenarios for (a) different time periods, (b) different emissions scenarios and (c) different probability distributions (i.e. arising from model parameter uncertainty).  The effects of emissions and time periods (a) and (b) are broadly the same in UKCP09 as in UKCIP02, but these may be smaller than the range of changes arising from model uncertainties (c) in many instances.

For further information on UKCP09 can be found if you click here.

top