Final Report Summary - EMBRACE (Earth system Model Bias Reduction and assessing Abrupt Climate change)
EMBRACE brought together 19 European research institutes specializing in Earth system science and modeling, representing 5 leading European Earth system models (ESMs). The primary objectives of the project were:
• To improve the scientific realism, and simulation quality of the 5 project ESMs, targeting a number of systematic biases and process deficiencies common across the 5 models.
• To develop and release a comprehensive and open-source Earth system model evaluation tool (ESMValTool).
• To use ESMValTool in analysing the 5 ESMs, at the outset and conclusion of the project, to assess the degree of simulation improvement arising from model developments carried out during the project.
• To develop an efficient protocol for assessing future risks of abrupt, potentially irreversible changes in regional climate phenomena.
• To apply this protocol to available ESM simulations in order to catalogue and understand the risks of abrupt changes at a range of future time horizons and global mean warming levels
In this report we summarize the main science and model development achievements of the project. We also outline some of the impacts the project will have on international efforts in Earth system modeling, in simulating and understanding future Earth system change and in the provision of reliable and actionable information on Earth system change both to policymakers and the general public.
EMBRACE started in October 2011, coincident with most ESM groups having frozen their coupled model configurations for the 5th Coupled Model Intercomparison Project (CMIP5). Model improvements developed within EMBRACE should therefore be seen as a part of a broader effort to improve European ESMs from CMIP5 towards readiness for CMIP6. At the conclusion of EMBRACE (February 2016) the 5 project-ESMs are now actively finalizing their coupled configurations for CMIP6. Many of the parameterization improvements developed within EMBRACE are central to these CMIP6 configurations and lead either to improved simulation quality, when the models are evaluated against observations, or introduce new (previously missing) process descriptions that more fully represent the full Earth system and its potential future sensitivity to anthropogenic greenhouse gas emissions. The expectation is that these improved and more process-complete models will deliver more realistic future projections of the full Earth system and therefore offer improved guidance for policy decisions addressing climate change mitigation and adaptation.
In addition to improving European ESMs, another important outcome of the project is the 1st release of the open-source ESMValTool. This community-developed tool offers the potential for a step-change in our ability to rapidly analyze future CMIP multi-model ensembles, both in terms of evaluating models against observations and earlier model configurations, and in in documenting future Earth system change and thereby delivering key findings early into policy discussions and major international assessments. ESMValTool also offers the potential for increased sharing of evaluation methods across the international community, increasing the overall efficiency of the research community with respect to analyzing ESMs and feeding these findings back into the model development process. ESMValTool will continue to be developed post-EMBRACE, with a particular focus on diagnostics and performance measures for biogeochemical phenomena and feedbacks with the Earth system.
Finally, while a number of parameterization and simulation improvements have been developed during EMBRACE, we stress that the 5 ESMs still have a number of systematic errors and still lack certain key processes. A continued European collaboration targeting further development of Earth system models therefore remains a priority.
Project Context and Objectives:
EMBRACE Objectives
The EMBRACE project brought together 5 European Earth system models (ESM) (from the model families/partner institutes: HadGEM (METUK), IPSL models (IPSL-CNRS), CNRM models (Meteo-France), MPI-ESM (MPI) and EC-Earth/IFS (SMHI, KNMI, ULUND and ECMWF)) around a collaborative effort to better understand the underlying causes of a number of systematic biases common to these models. Based on this understanding, the project then constructed a range of constrained modelling protocols through which improved process parameterizations could be developed with the aim of reducing the targeted systematic biases in some or all of the models. A key requirement was that the model improvements were physically based and justified (either by observations or explicit modelling of the target phenomena) thereby increasing the overall realism and reliability of the coupled models.
The project also targeted development of a community model evaluation tool (ESMValTool, Eyring et al. 2015) to facilitate inter-comparison of models against each other and against suitable observations. The tool emphasized both “quick-look” performance-metrics that give an integrative overview of model simulation quality, as well as a range of more in-depth, process diagnostics concentrating on the model biases/phenomena highlighted for improvement in the project. ESMValTool was used to evaluate the performance of the EMBRACE models, at the outset of the project, and to compare these models against updated versions of the 5 models at the conclusion of the project. The latter analysis was to determine whether simulation improvements had occurred as a result of the parameterization developments made during the project. ESMValTool has now been released as an open-source, community evaluation tool and is becoming widely used in the climate modelling community and will likely play a central role in documenting the overall performance of models in CMIP6.
Finally, the project constructed a set of experiment protocols to investigate the risk for future abrupt changes in a number of key climate phenomena, developing a set of automated tools to scan large multi-model data sets for the existence of such abrupt changes. Once abrupt changes were identified the underpinning cause of these changes were analysed in each model, with a view to establishing the robustness of the changes across models and the likelihood of the changes actually occurring based on understanding of the simulated processes.
The project was organised into 5 individual research work packages (WPs) and a 6th WP for project management, coordination and knowledge dissemination. The 5 research WPs were split as: WP1 Atmospheric convection and links to tropical circulation, WP2 Ocean and sea ice processes, WP3 Land and vegetation processes, WP4 Model evaluation and development of ESMValTool and WP5 Abrupt climate change. WP1, WP2 and WP3 concentrated on parameterization improvement in their three respective model domain, with improvements developed then assessed in coupled model versions in WP4. The main science achievements and model developments from the project are presented in section 3 where each WP is reported on individually.
The main model biases and climate phenomena targeted for improvement in the project include:
1. The representation of deep tropical convection. In particular: (i) the diurnal cycle of convection, (ii) the role of deep convection in the coupled (ocean-atmosphere) equatorial mean climate and (iii) the role of deep convection in monsoon circulations.
2. The simulation of ocean upwelling and stratocumulus clouds in the Eastern tropical oceans and their role in the large-scale coupled tropical climate.
3. The representation of the South Asian and West African summer monsoons.
4. The Atlantic Meridional Overturning Circulation (AMOC), as well as the near surface Gulf Stream and its North Atlantic extension.
5. Systematic biases in the Southern ocean surface energy budget resulting from cloud-radiation and ocean mixed-layer processes.
6. The representation of surface hydrology and its role in summer season continental dry biases.
7. Simulated soil and vegetation biogeochemical processes
8. Processes controlling ocean alkalinity.
9. The representation of sea-ice rheology.
In addition to these climate- processes/phenomena, a number of model developments were pursued with an aim to increase the breadth and flexibility of tools available to the European climate research community. These include:
1. The ability to perform online advection of 3D biogeochemical tracers at a reduced resolution relative to the host dynamical ocean in the NEMO-ORCA model.
2. The ability to apply coupled (atmosphere and ocean) localised model resolution enhancements, using the AGRIF zoom facility, within the NEMO, OASIS and WRF model systems.
3. A reduced-radius (small-planet) version of the ECMWF Integrated Forecast System (IFS) to enable the simulation of resolved convection and its interaction with tropical waves at reduced computational cost.
4. A flexible and open-source climate model evaluation tool (ESMValTool).
Finally EMBRACE has developed a number of simulation data sets that will prove useful beyond the life time of the project for both model evaluation and parameterization development. These include: (i) WP1 performed a number of high-resolution (convective resolving and large eddy simulation) model simulations of observed and idealized convective events. These data were used in the project to guide the improvement of convection, cloud and turbulence schemes and will be available for continued use in this manner after EMBRACE. This activity has already begun through a collaboration with the GEWEX GASS community; (ii) WP2 performed a number of high-resolution nested simulations of the North Atlantic Ocean ranging from 1/8° to 1/32° AGRIF zooms over the North Atlantic embedded within a ½° global version of NEMO-ORCA. This data can also be used in the future as a reference for global ocean model development. Finally, in WP3 a large suite of offline land-vegetation model simulations were performed using observed atmospheric forcing. These data sets will also be of future use in evaluating fully coupled model simulations.
EMBRACE context
EMBRACE started in the autumn of 2011, just as the modelling centres were finalizing their model configurations for CMIP5. Hence these models were the entry point for EMBRACE, from which systematic biases were analysed and against which model improvements were to be developed. Most ESM groups now follow a coupled model development cycle somewhat phased with CMIP cycles. This means that groups are only now (in ~early to mid 2016) finalizing model configurations for CMIP6. The model improvements carried out within EMBRACE can therefore be viewed as contributing to the overall model development from CMIP5 to CMIP6. At each of the participating centres, EMBRACE developments constitute only a sub-set (albeit relatively significant) of the full model development effort over the past ~5-6 years. One difficulty with the timing of EMBRACE was that towards the conclusion of the project (i.e. the latter half of 2015), new coupled historical simulations were required to evaluate the impact of EMBRACE parameterization improvements on overall model performance compared to the CMIP5 generation of models. This requirement was not well phased with the modelling groups final model coupling and tuning effort in advance of CMIP6, with this effort largely occurring now (mid 2016). The result of this timing mismatch is that the new coupled historical simulations (made by the 3 EMBRACE models, HadGEM3, EC-Earth and MPI-ESM, following the CMIP5 protocol) should be viewed as preliminary coupled versions only. The IPSL and CNRM models were deemed to be in too early a stage of development for new coupled simulations to be performed. Furthermore, all the improved models also include a number of developments realised outside the EMBRACE project during the same time period. This is an unavoidable aspect of coupled model development, where one set of parameterization improvements is often dependent on another parallel set of developments to realise their benefit in a fully coupled model. Furthermore, it is often difficult to isolate the contribution of one specific set of model improvements from another, with respect to full coupled model performance. This does not reduce the importance of the model improvements made in EMBRACE, rather these should be considered as an important part of the overall development effort at each of the modelling centres moving from CMIP5 to CMIP6. In addition to the new coupled historical simulations, all 5 EMBRACE models performed new AMIP simulations (atmosphere-land only models with prescribed sea surface temperatures, SSTs). These simulations can be used to evaluate atmosphere and land process improvements in the 5 models. Due to the prescribed SSTs the degree of tuning required for an AMIP-style simulation is less than for fully coupled runs. For the 3 models that also made coupled historical simulations, the role of ocean coupling in the phenomena investigated was also analysed through comparison of the coupled and AMIP simulations.
It is important to reiterate that the models at the conclusion of EMBRACE, while including a number of important improvements, are intermediate versions, on the way to fully coupled models ready for use in CMIP6. This intermediate nature of the “improved” coupled models also led the project to decide it would not be sensible to make new projections with the improved models for an updated assessment of the abrupt changes identified in the CMIP5-class models. Rather it was decided the original abrupt analysis be extended to a larger multi-model data set and more carefully analyse processes underpinning abrupt changes identified. A more detailed consideration of the likelihood of committed changes in simulated vegetation, once the greenhouse gas forcing scenario (RCP8.5) reached is maximum value in 2100, was also performed. More details on this latter point can be found in the detailed report on WP5 science results and in deliverable D5.7.
Summary of the main EMBRACE science results
The main EMBRACE science results and model developments are outlined in section 3 of this report. Here we briefly highlight a few of the main successes and document those areas still requiring significant research effort.
1. In the overall evaluation of the EMBRACE-updated AMIP and coupled historical simulations there was a clear improvement in the representation of the Southern ocean surface and top of atmosphere (TOA) radiation budget, with an accompanying improvement in the simulation of cloud cover. This improvement was seen across most of the models and is a significant improvement on CMIP5. Problems do still exist, particularly in the extreme south of the Southern Ocean (e.g. ~60-65°S).
2. Four of the five EMBRACE-updated models showed improvement in simulated tropical precipitation in the AMIP experiments, while 2 of the coupled historical runs showed significant improvement in the representation of equatorial ocean SSTs and associated coupling phenomena, such as near surface wind stress and convective precipitation.
3. Parameterizations for convectively forced gravity waves were successfully implemented into 2 of the EMBRACE models (IPSL and HadGEM3) with successful simulation of the quasi-biennial oscillation (QBO).
4. The diurnal cycle of tropical convection was significantly improved in the IFS/EC-Earth model as documented in Bechtold et al. (2014).
5. Significant improvement in the simulated Atlantic Meridional Overturning Circulation (AMOC) and associated Gulf Stream dynamics were seen when two AGRIF zooms (at 1/8° and 1/32°) were implemented into a global ½° resolution version of NEMO-ORCA. Findings from this simulation will help guide future parameterization development in NEMO-ORCA.
6. An online coarsening option for (reduced-resolution) advection of 3D marine biogeochemical tracers has been implemented in the NEMO-ORCA dynamical ocean model. Many groups hope to use this facility in CMIP6 which will allow a significant computational saving in the ocean marine biogeochemical component of the ESMs with little degradation in simulation quality.
7. An improved treatment of CaCO3 dissolution and its impact on ocean alakalinity has been developed in the PlankTOM marine bioeochemical model. A simplified form of this scheme developed and tested in HadOCC.
8. EMBRACE made a significant contribution to the implementation of a nitrogen cycle in the terrestrial components of the 5 project ESMs, with it now likely these models will all include some form of terrestrial nitrogen cycling in their CMIP6 configurations. This is a major improvement on CMIP5 where none of the 5 models included terrestrial nitrogen.
9. ESMValTool has been successfully released, accompanied by a detailed user guide, and the tool fully documented (Eyring et al. 2015)
10. Drijfhout et al. (2015) present a catalogue of identified abrupt changes in the multi-model data set underpinning IPCC AR5. For abrupt changes associated with the overturning Atlantic Ocean circulation, they found that models with a more realistic simulation of present day processes also exhibited a greater number of abrupt changes. Surprisingly, a relatively significant fraction of abrupt changes were identified even below a 2°C global warming threshold relative to pre-industrial values.
At the conclusion of EMBRACE ESM biases and phenomena still requiring improvement include:
1. Despite a number of significant improvements in the parameterization of deep convection, the representation of both the South Asian and West African monsoons remains a major challenge for all EMBRACE models and is an area requiring significant continued effort. Coupled with the West African monsoon bias all 3 EMBRACE-updated coupled models (HadGEM3, EC-Earth and MPI-ESM) continue to have warm SST biases off the Angola-Namibia coast. Some improvement to higher time frequency precipitation variability was seen for both monsoon systems when EC-Earth resolution was increased from ~125km to ~25km, although the main climatological biases remained largely unaffected across these resolutions.
2. Only one of the coupled historical simulations (that performed by the EMBRACE-updated EC-Earth3 model) did not exhibit a double ITCZ problem in the equatorial Pacific. Hence while significant improvements were seen in a number of aspects of coupled equatorial ocean processes this remains an outstanding bias.
3. The representation of sub-tropical stratocumulus clouds continues to be relatively poor in most ESMs.
4. As previously reported (Mueller and Seneviratne, 2014), most models continue to overestimate evapotranspiration in summer, especially over Europe, Africa, China, Australia, Western North America.
5. Most models continue to show a positive bias in precipitation over high northern continental land regions.
Project Results:
Please refer to the attached file: Main Results and Potential Impact.pdf
Potential Impact:
Please refer to the attached file: Main Results and Potential Impact.pdf
List of Websites:
http://www.embrace-project.eu/