Work Package 4: Development of models and climate projection systems


Sub project 1 - Global climate model evaluation and development

The continuous development of climate prediction models underpins the modelling capability within a climate prediction programme. This component is focused on model evaluation and development activities and aims to develop UK-China collaborations to study methodologies for evaluating and improving the physical and dynamical processes in the numerical models of atmosphere, ocean, land, and sea-ice and to evaluate the coupling between these components in the climate system.

Sub project 2 - Climate projection systems

Societies need predictions of 21st century climate on time scales ranging from a season to a century or more ahead. This sub-project will focus on projections for the 5-40 year time scale, to inform a range of sectoral adaptation choices during a period in which uncertainties associated with natural variability and climate modelling are expected to exceed those arising from future emissions pathways. A new perturbed parameter ensemble (PPE) of the HadGEM3 coupled ocean-atmosphere model will be designed to represent important process uncertainties identified by model parametrisation experts. Following a rigorous development and evaluation phase, new simulations of historical and future climate will be run. In parallel, uncertainty quantification methods based on existing PPE and multi-model ensemble (MME) techniques (e.g. CMIP) and emerging observational constraints will be extended to assess the potential to obtain more robust estimates of plausible ranges of future variability and change over East Asia.


  • Develop improved understanding of model systematic errors and their sources at the detailed process level to underpin ongoing model development activities at the Met Office and CMA/IAP. The focus is on improving East Asian climate variability (seasonal to decadal timescales), although evaluation and diagnostic techniques will potentially range across all prediction timescales from weather to climate in order to address the relevant scientific issues.
  • Develop new PPE based on HadGEM3-GC2 coupled model and explore the methodology for using this PPE to both explore modelling uncertainties at the process level (sub project 1) and better quantify uncertainties in predictions of future climate (sub project 2).
  • Combine results from PPE and multi-model ensembles (MME e.g. CMIP), rather than relying exclusively on CMIP results in isolation, to sample plausible future outcomes in a more comprehensive fashion, and to discover new emergent observational constraints which can be shown to be robust to different choices of ensemble design.

Summary of ongoing work

Sub project 1 - Global climate model evaluation and development:

The model development research activity under sub-project 1 is new within this work package and we have been defining a science plan that will be developed with Chinese partners and working towards a project proposal call for UK academic partners for years 2-4 of CSSP.

Sub project 2 - Climate projection systems:

The long term plan is to identify how to generate a practically-sized ensemble of HadGEM3 coupled simulations that explore a diversity of forcings and feedbacks. The Met Office have developed a methodology for the selection of 50 variants of HadGEM3 atmosphere-only runs designed to be credible which use errors in a small sample of 5-day forecasts and 10-year mean climate biases combined with a climate model emulation technique. The range of climate feedback for the plausible HadGEM3 variants is similar to that explored by CMIP5. Experiments for the 50 HadGEM3 perturbed parameter variants have been run and the mean square errors on 5-day time scales for the vast majority of the 50 new runs were reasonably well predicted from the original smaller set of predictions.

Validation of our ability to predict mean square errors on 10-year time scales from the original set of runs is underway and predictions for five of our six variables in the test validate reasonably well. Precipitation has not validated as well and the validation on the 10-year time scale is not as good on the 5-day time scale because the ensemble size is smaller. The lesson for the next planned ensemble with the version of HadGEM3 coming out in June 2015 (GC3), will be to increase the ensemble size relative to the number of parameters to perturb. The predictions for feedbacks and the cloud forcing components of 4xCO2 and pre-industrial aerosol forcing generally validate well. More importantly for our methodology and aim to explore a diverse range of forcings and feedbacks, the range seen in the predictions seems to be largely retained in the larger set of experiments.

Science highlights from year 1

  • Met Office - Methodology for generating a practical-sized perturbed parameter ensemble (PPE) using small samples of 5-day forecasts and 10 year mean AMIP simulations combined with a climate emulator has been evaluated and shown to be a good predictor of the full (50 variant) PPE range of forcing and feedbacks.
  • Project WP4.2 (University of Reading) - Assessments of the skill of initialised HiGEM decadal simulations have been made for East Asia, including analysis of the drivers of China and East Asian decadal variability in the HiGEM control simulation. Results to date show evidence that initialisation gives enhanced skill on projections of SSTs in many regions of the tropical and extra-tropical Pacific Ocean, at lead times beyond 5 years (e.g. the El Nino Southern Oscillation and the Pacific Decadal Oscillation). This analysis of skill has also been extended to a number of East Asia land regions. The initialised hindcasts show additional skill in predicting the PDO and ENSO indices when compared to uninitialised simulations.
  • Project WP4.3 (University of Exeter) - It has been assessed that the available river time series observations are useful for assisting projections, despite a few, possibly dam-related problems in the record. Using the classical Budyko relationship between runoff and precipitation (as a way of understanding the differences between energy- and water-limited environments), it has been determined that CMIP5 model simulations for northern China tend to show too much rainfall compared with observations, which carries through to errors in predictions of runoff. These errors dominate errors in the modelled runoff-rainfall relationship, although there are also some rather substantial errors in the runoff-rainfall relationship - even in land surface models forced by observed precipitation. The final stage is to determine whether model predictions of precipitation change can be adapted to make sensible projections of land surface hydrological change and work is underway to prepare the final projections for the Yellow and Yangtze river, incorporating observed biases.

List of year 1 academic partner projects

  • Project WP4.2 - Understanding the benefits and limitations of existing initialised decadal projections - University of Reading, Jonny Day & Ed Hawkins
  • Project WP4.3 - Investigate methods to derive credible information of regional climate change in the East Asian region from CMIP5 multi-model simulations - University of Exeter, Joe Osborne & Hugo Lambert

Last updated: 27 February 2015

    Share this: