Climatic Research Unit

Methods and issues

This page provides information about methods and issues related to the construction of probabilistic climate information. The relevance of each issue is outlined briefly and links provided to ENSEMBLES reports which deal with it in more detail. Most of these reports are publicly available - a few are restricted to ENSEMBLES partners.

Weighting

The simplest way of presenting ensemble means is as un-weighted estimates. However, this does not take any account of differences in GCM or RCM model performance for the present day. Ensemble weights are calculated by comparing model output for the present day with observed data. Better performing models are assigned higher weights and hence given more emphasis. Weights differ depending on the variable and season considered, so a consistent way of producing them is needed. How to calculate and use weights is one of the major research issues for ENSEMBLES.

D2B.6	Refinement of the Reliability Ensemble Averaging (REA) framework	Outlines a refined REA scheme in which a model weight depends on a combined functional form that includes model mean bias, internnual variability and multidecadal trend performance for both temperature and precipitation.
D2B.8	Working paper on model weighting for the construction of probabilistic scenarios in ENSEMBLES	Deals with weighting model output on the basis of performance, typically evaluated by comparing observed and simulated data. Summarises some of the main issues for ENSEMBLES and makes a number of recommendations for constructing weights.
D3.2.1	Definition of measures of reliability based on ability to simulate observed climate in hindcast mode	Preliminary discussion of how to calculate model weights and how to produce regional PDFs based on the ENSEMBLES multi-model set.
D3.2.2	RCM-specific weights based on their ability to simulate the present climate, calibrated for the ERA-40 based simulations	Involves the development of techniques for the generation of probabilistic predictions by statistical processing of regional ensemble integrations, in particular, means for weighting individual RCM members of an ensemble.
D3.3.1	Evaluated RCM-system for use in RT2B (choice of RCM-GCM combinations and preliminary RCM weights)	Describes how a matrix of GCM-RCM combinations has been constructed for ENSEMBLES scenario runs, and how the preliminary weights (D3.2.2) can be used to produce probabilistic information.
D3.3.2	Final version RCM-system for use in RT2B based on RCM-specific weights from WP3.2	Describes how the GCM-RCM experiment matrix can be filled by a statistical approach. The deliverable also presents and describes how the RCM weights from WP3.2 are applied so as to produce probabilistic scenarios. (The production of the specific RCM weights is described in D3.2.2).

Pattern scaling

This technique is used to estimate regional changes from larger-scale model output. It was used in UKCIP02, for example, to construct scenarios for time periods and emissions scenarios for which RCM output was not available. Its potential advantage is that it allows larger ensembles to be built at the regional scale. It is, however, based on an assumption of linearity which may not always be applicable, particularly with respect to rainfall and extreme events.

D2B.7	Methodologies for pattern scaling across the full range of RT2A GCM ensemble members	Examines methodologies of pattern scaling, appropriate for providing climate projections at the RCM-scale across the full range of ensemble members for a given GCM. Methods for scaling mean and variance and also extremes are examined, focusing on temperature and precipitation.
D2B.25	Applicability of pattern scaling for filling the ENSEMBLES GCM-RCM matrix	The paper assesses the applicability of pattern scaling methods to provide climate projections at the RCM scale for a full range of driving GCMs.

Model emulation

In model emulation, statistical approaches are developed to draw inferences about the values of outputs from computationally expensive models for parameter values at which the models have not yet been evaluated. Thus it has the potential to considerably increase the ensemble size available. In ENSEMBLES, it is used by the Hadley Centre to produce probabilistic information from a GCM perturbed physics ensemble. Towards the end of ENSEMBLES, the possibility of using statistical downscaling to emulate RCM output is being explored in RT2B.

D2B.27

Assessment of the robustness of the statistical downcaling techniques using GCM and RCM outputs

Uses RCM output as 'pseudo-observations' to explore the stationarity of statistical downscaling

Ensemble averaging techniques

Most previous work on techniques for ensemble averaging (including Bayesian and Monte Carlo resampling techniques) has focused on the global or sub-continental scale, as well as on seasonal-to-decadal timescales. Thus the development of appropriate methods for regional spatial scales at climate change timescales is an important aspect of RT2B work.

D2B.6	Refinement of the Reliability Ensemble Averaging (REA) framework	Describes how the REA method for combining output from multiple-models with model weights in order to produce regional ensemble averages is being further developed in ENSEMBLES by the ICTP team (led by Filippo Giorgi).
D1.2	Systematic documentation and intercomparison of ensemble perturbation and weighting methods	Provides a short and fairly technical outline of new developments in the methodology of ensemble climate change projections and seasonal-to-decadal forecasting using global models, including the Bayesian methodology developed by the Hadley Centre. Also provides a useful list of references.
ETR 5	Temperature and precipitation probablilistic density functions in ENSEMBLES regional scenarios	The Gaussian kernel method is used to produce PDFs and CDFs for European cities from ENSEMBLES RCM outputs

Extreme events

A number of ENSEMBLES research themes (RTs) have undertaken work on extreme weather and climate events, focusing on heat waves, drought, intense rainfall and wind storms. RT4 gives particular attention to understanding linear and non-linear feedbacks in the climate system that may lead to climate "surprises" and to understanding the factors that govern the probability of occurrence of extreme events. RT5 interacts with RT4 in the study and understanding of processes - providing the evaluation of model errors and biases that serve as a basis for RT4 investigations. For many impacts applications, changes in the frequency, persistence and magnitude of extreme events are of greater concern than mean changes and are therefore one specific focus of RT6 work, with RT2B providing the regional information that is needed. All this ENSEMBLES work builds on earlier work undertaken in the MICE, PRUDENCE and STARDEX projects.

D4.3.1	Software for exploring extreme events in gridded data sets.	Reviews and introduces new tools for the analysis of extreme climate and weather events in gridded datasets. Software for these methods has been coded up in the R statistical language and can be freely obtained from the R software for CLIMateanalysis RCLIM website.
D4.3.2	Report on the relationship between extreme events in surface temperature and the large scale circulation over both the S Europe/Mediterranean and N Europe/Arctic regions both in reanalysis and model simulation	Presents different aspects related with extreme events. In particular, four research lines have been developed: a methodology to study the interannual variability associated with summertime months in which extremely hot temperatures are frequent; extreme temperature events over the Eastern Mediterranean region; the Artic melt seasons and identifying large-scale circulation structures that favour warm days in the Netherlands.
D5.2	Assessment of the decadal-scale variations of precipitation extremes in ERA40 by comparison to observations in the Alpine region.	A bias-correction scheme is developed and applied to ERA40 reanalysis for the Alps. The bias-corrected data shows good skill in capturing the variations of precipitation extremes on interannual to decadal time scales.
D5.3	Scientific report and R software on optimal statistical methods for combining multi-model forecasts to make probabilistic forecasts of rare extreme events.	Addresses the problem of formulating predictions of extreme events at seasonal and interannual time scales using ensembles of dynamical simulations from multiple ocean-atmosphere coupled models.
D5.4	Verification methods for forecasts of extreme events.	This report attempts to establish a simple framework for verification of certain types of extreme event, focusing on exceedences, and then discusses some of the possible approaches in the hope that this will lead to more developments in this important area of verification.
D5.21	Preliminary evaluation of precipitation extremes in RCM data for the Rhine basin	Presents an extreme-value model for the evaluation of precipitation extremes in transient RCM simulations .
D5.22	A non-stationary index-flood model for precipitation extremes in transient RCM runs	The non-stationary GEV model is applied to the 1-day summer and 5-day winter precipitation extremes in the river Rhine basin in a simulation of the RACMO regional climate model for the period 1950–2100. Paper submitted to Journal of Geophysical Research - Atmospheres.
D5.23	Detailed comparison of extremes in daily station observations and in gridded daily data	This report includes: (1) the evaluation of the gridded observed dataset with respect to the effects of gridding on the extreme values of temperature and precipitation and (2) the assessment of the extreme values and their trends in temperature and precipitation seen in the gridded observed data.
D6.2	First phase impact models to predict damage to human activities, the environment and tropical annual crops from climate extremes.	Outlines some of the models that will be used in WP6.2 to explore climate extremes: effects on health, heat stress, wind storm, flood and return levels; GLAM tropical crop model; a heat index for heat stress studies; and, a storm damage regression model.
D6.8	Preliminary report on changes in climate extremes and their relation to health, flood risk, agriculture, forest and property damage	Presents a more sophisticated assessment of the uncertainty in projected impacts than was possible in the earlier studies. The responses to climate extremes are assessed at high spatial resolution for a continuous period to 2100
D8.8c	Workshop on Extreme Climatic Events in Switzerland	This workshop held in March 2006 aimed to identify ways to answer a number of scientific questions about extreme events identified by ENSEMBLES. A follow-up workshop was held in January 2007.

Statistical versus dynamical downscaling

This issue was considered, in part, by the STARDEX project (c.f. Haylock et al., 2006, International Journal of Climatology, 26, 1397-1415; Schmidli et al., 2007: Journal of Geophysical Research, 112, D04105, doi:10.1029/2005JD007026). In ENSEMBLES, both methods are seen as complementary with their own particular advantages and disadvantages. One aim of the RT2B work is to explore how the two approaches can be used synergistically.

D2B.10

Preliminary report on a comparison of statistical downscaling and very-high resolution dynamical downscaling with the WRF model

Investigates whether there is any merit in extending dynamical downscaling to the meso-scale (~5km or less, horizontal resolution). Does it provide a more realistic description of local weather, particularly in extreme weather conditions, compared with statistical downscaling based on the outputs from a coarser (e.g. 25km horizontal resolution) dynamical downscaling approach?

Emissions scenarios

The regional climate change scenario runs undertaken in ENSEMBLES focus on the IPCC SRES A1B emissions scenario. Thus the regional probablistic information produced in RT2B are conditional on this emissions scenario. ENSEMBLES RT7 has, however, undertaken work on reviewing and developing new emissions scenarios which have been used in the stream 2 GCM simulations.

D7.1b	Critical assessment of the IPCC SRES scenarios	Outlines the SRES scenarios, and provides a critique focusing on the demographic and economic details. It is concluded that it is appropriate for the ENSEMBLES GCMs to run the SRES scenarios.
D7.1c	A review of the baseline scenarios that have emerged since the SRES, and the selection of one scenario, based on that which gives the most new scientific information over and above the SRES scenarios, to be made available for the ENSEMBLES stream 2 simulations.	A new baseline emission scenario is presented, dubbed E1. It is calibrated with purchasing power exchange rates and uses a new growth model as the engine. Despite different patterns of economic growth, global carbon dioxide concentrations are well within the SRES range - closely tracking the B1 scenario.
D7.2	Analysis of the impact of disasters on economic growth	Investigates the possibility that disasters have positive economic consequences, through the accelerated replacement of capital, using a model with embodied technical change.
D7.7	Estimates of the effect of climate change on the emissions of carbon dioxide and sulphur aerosols	Analyses the impact of climate change on CO2 and SO2 emissions by using a multicountry computable general equilibrium model
D8.6	Policy paper on greenhouse-gas emission scenarios

Regional impacts

ENSEMBLES has taken an end-to-end approach in which regional projections are linked to an extensive range of applications including agriculture, forestry, human health, water resources and energy use and demand. In order to best tailor the regional climate information a survey of user needs was undertaken.

Application studies have been undertaken on both seasonal-to-decadal and climate change timescales

Seasonal-to-decadal application studies

ENSEMBLES deliverables	Details	Access
D5.10	Workshop report on "Lessons learned from seasonal forecasting: health protection"	Public
D6.2	First phase impact models to predict damage to human activities, the environment and tropical annual crops from climate extremes, e.g. wind storm, drought, flood, and heat stres	Public
D6.4	Seasonal-to-decadal application models running as part of an integrated probabilistic ESM based on DEMETER hindcasts	ENSEMBLES partners

Climate Change impact studies

ENSEMBLES deliverables	Details	Access
D2B.20	Preliminary assessment of changes in regional weather and climate over Europe using ENSEMBLES global climate model outputs as they become available	Public
D2B.28/D6.19	Journal paper on the assessment of changes in regional weather and climate in the Eastern Mediterranean	Public
D2B.20	Preliminary assessment of changes in regional weather and climate over Europe using ENSEMBLES global climate model outputs as they become available	Public
D6.3	Calibrated and tested crop, forest, hydrology and energy impact models; baseline data and scenarios for constructing impact response surfaces	ENSEMBLES partners
D6.7	Preliminary report on a comparative study of response surface and multiple scenario approaches to assessing risks of impacts using selected impact models	ENSEMBLES partners
D6.8	Preliminary report on changes in climate extremes and their relation to health, flood risk, agriculture, forest and property damage	Public
D6.9	Report on an intercomparison study of modelled, Europe-wide forest fire risk for present day conditions	Public
D8.8d	Workshop and report on Climate, Climatic Change and Impacts: Applications to Eastern Europe in Romania	ENSEMBLES partners
D8.8e	Workshop and report on Adaptation to the impacts of climate change in the European Alps	ENSEMBLES partners
D8.8f	Workshop + report on Climate Change, Impacts and Adaptation in the Mediterranean (Greece)	ENSEMBLES partners

Also see the ENSEMBLES stakeholder newsletters: Issue 1 and Issue 2

return to top