DFG
  1. Home
  2. Research
  3. Area S
  4. S1

S1: Diagnosis and Metrics in Climate Models

Principal investigators: Prof. Hans Burchard (Leibniz Institute for Baltic Sea Research Warnemünde), Prof. Veronika Eiring (MARUM/University of Bremen), Prof. Thomas Jung (Alfred Wegener Institute for Polar and Marine Research/University of Bremen), Prof. Nedjeljka Žagar (Universität Hamburg)

In a continuation and extension of the S1 subproject from the first phase of the CRC, S1 will assess the impact of model developments – proposed by the other project areas – on the quality of two climate models developed and used by the German scientific community, namely ICONa/ICON-o and OpenIFS/FESOM2. Major parts of the work will be organized around performing numerical simulations with new parameterizations, and development and application of new metrics, diagnostics and tools for model evaluation in terms of model biases and response to forcing.

Numerical simulations will be performed for a hierarchy of model setups, ranging from simplified channel models to realistic global coupled high-resolution simulations following CMIP6 protocols. These simulations will provide the basis for further model evaluation. Scalable open source tools for processing and evaluation of this data will be further developed and shared with the community. We will continue using diagnostics developed in the first phase, and implementing additional diagnostics already published in the literature. Moreover, we will develop new diagnostics that will help us to evaluate models in the context of their energetic consistency.

The new diagnostics will extend the validation of the scale-dependent spatial variability performed by the MODES software to the temporal domain. The extension will include a novel method that quantifies the effects of model biases on the simulated spatio-temporal variability. Remote effects of model biases – here called bias teleconnections – will be initially investigated in a GCM of intermediate complexity, then applied to ICON-a/ICON-o and OpenIFS/FESOM2 including new parameterizations.

Numerical mixing and dissipation are potential sources of large errors in ocean components of climate models. New numerical schemes developed in the CRC to tackle these shortcomings need to be properly evaluated. Therefore, methods to quantify numerical errors in water mass transformations will be further developed in M5 during the first two years and implemented and evaluated in FESOM2 during the last two years of the second phase of S1.

Results of ICON-a/ICON-o and OpenIFS/FESOM2 model simulations will be put into a broader climate modelling context (with emphasis on CMIP6 simulations). The Earth System Model Evaluation Tool (ESMValTool), a community diagnostic and performance metrics tool for routine evaluation of climate models, will be enhanced with the offline diagnostics developed within this project. By making additional use of a wide range of diagnostics already included in the new ESMValTool version 2.0 (v2.0), the impact of model developments will be thoroughly evaluated, also in the context of CMIP6 models.

Fig.1: Mean systematic errors of 500 hPa geopotential height fields (shading in dam) for winters (December–February) of the period 1962–2005 and various versions of the ECMWF model: (a) 29R2, (b) 30R1, (c) 31R1, (d) 32R1, (e) 32R3 and (f) 33R1. Also shown are mean fields (contours) obtained from a combination of ERA-40 (1962–2001) and operational ECMWF analysis data (2002–2005). Mean systematic errors significant at the 95% confidence level are hatched. Notice that negative values are contoured (as well as shaded). (From Jung et al. 2010; Q. J. R. Meteorol. Soc. 136: 1145–1160).

The hypothesis underlying the proposed Collaborative Research Centre is that a more consistent representation of parameterised processes and numerical formulations will result in better climate models. To test this hypothesis, therefore, it is critical to thoroughly evaluate the influence of enhanced model formulations, developed in the other project areas, on the realism of climate models. Here, we will focus on two next-generation climate models developed and used by the German climate research community, namely ICON-a/ICON-o and ECHAM6/FESOM.

Efforts towards developing a new generation of energetically consistent models require the availability of tools that can help to identify short-comings and quantify progress (e.g. Figure 1). Therefore, we will develop and apply new diagnostics and metrics that help us to understand the pathways of energy transport and energy conversion, analyse the energetic consistency of existing models and formulate measures taken up by the other subprojects to advance the energetic consistency and quality of parameterisations of climate models. Furthermore, we will develop effective metrics and diagnostics that can be used to quantify the realism of models and unravel the origin of model error. Metrics describing systematic model error in atmosphere, ocean and sea ice using some of the most advanced observational data sets for reference, will be elaborated. Furthermore, dedicated metrics will be developed that capture important weather and climate phenomena resulting from instabilities such as extratropical storms, Euro-Atlantic blocking, or monsoon dynamics. Finally, we will construct metrics able to analyse consistently a large class of extremes occurring across a wide range of spatial and temporal scales.

An important aspect will be the practical implementation of new tools that can be used for model evaluation. We will build on the newly developed Earth System Model Validation Tool (ESMValTool)—a community diagnostic and performance metrics tool for routine evaluation of climate models. By doing so, we can also benefit from other community efforts in developing evaluation tools. Moreover, we can add our new metrics and diagnostics to ESMValTool thereby increasing the applicability and international visibility—and hence the impact—of the research coming out of this subproject.

Fig. 2: Relative space–time root-mean-square deviation (RMSD) calculated from the climatological seasonal cycle of the CMIP5 simulations. The years averaged depend on the years with observational data available. A relative performance is displayed, with blue shading indicating better and red shading indicating worse performance than the median of all model results. Note that the colours would change if models were added or removed. A diagonal split of a grid square shows the relative error with respect to the reference dataset (lower right triangle) and the alternative dataset (upper left triangle). White boxes are used when data are not available for a given model and variable. The performance metrics are shown separately for atmosphere, ocean and sea ice (a), and land (b). Extended from Fig. 9.7 of IPCC WG I AR5 chap. 9 (Flato et al., 2013) and produced with recipe_perfmetrics_CMIP5.yml.; see details in Sect. 3.1.1.

To aid effective evaluation of the model development activities, coming out of the other research areas, and to guide further efforts, we will design protocols for numerical experimentation, including common model setups, together with the PIs from subproject S2. These protocols will allow direct comparability between different modelling experiments, ensure availability of model output needed to compute all relevant metrics and diagnostics and provide a framework to evaluate model changes in the context of CMIP6-type experiments with ICON-a/ICON-o and ECHAM6/FESOM.

The results will be made available through a web-based tool—the Diagnostics and Metrics Explorer—for effective analysis. We will take advantage of direct collaboration with ongoing Horizon2020 projects (CRESCENDO and PRIMAVERA) dealing with evaluation of Earth System Models in the context of CMIP6.

FESOM model

One of the two ocean models used in the TRR181 project is FESOM (Finite-Element/volumE Sea ice-Ocean Model), which is part of the AWI-CM climate model of the Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research (AWI).

FESOM is the only ocean model participating in CMIP6, that is formulated on an unstructured mesh. This allows scientists to flexibly increase horizontal resolution of the model in more energetically active areas of the ocean like the Gulf stream or Agulhas current. The resulting model resolves important details of ocean circulation, but is still computationally efficient.

FESOM ocean currents in Indian and Pacific oceans (100m)

The Earth’s energy budget and other funny aspects of the thermodynamics of the climate system

State-of-the-art climate models still struggle to reproduce a reasonably energetically consistent system, even though outstanding improvements have been achieved in the recent past.

Valerio Lembo, Postdoc in S1

The idea of this subproject is assessing the impact of introducing new numerical schemes and physical parametrizations developed in the TRR181 for the energy closure of state-of-the-art climate models. We provide diagnostic tools that allow for evaluation and intercomparison of climate models, starting from their outputted datasets.

It might sound trivial, expecting that the climate system, if in steady state, is also in thermodynamic equilibrium. This is at least what our studies of classical thermodynamics suggest. The problem is that the system constantly exchanges energy with its exterior, i.e. the outer space, and within its interior. In steady state conditions, the net exchange of energy with the exterior has to be null. In other words, the climate system is in thermodynamic equilibrium, once we averaged out the modulation of the solar energy input to an appropriately long timescale and all the energy exchanges occurring in its interior, shaping the solar “reflection” and the thermal energy output. This is a clear example of what is called a “non-equilibrium dissipative steady state thermodynamical system”.

State-of-the-art climate models still struggle to reproduce a reasonably energetically consistent system, even though outstanding improvements have been achieved in the recent past. This points to the very basic reasons for climate modeling, on one hand reflecting the lack of understanding of some processes involving energy exchanges and the limits of the discretization/truncation of the real world in finite dimension models, on the other hand preventing us from correctly evaluating the impact of the various forcings for reconstructed and projected climate change.

As TRR181, we are participating to the community effort called “ESMValTool”, whose aim is providing a set of standardized diagnostics for the evaluation of state-of-the-art and forthcoming multi-model ensembles. In our diagnostics, we try to address specifically the Earth’s energy budget and its atmospheric and oceanic components, and the atmospheric energy exchanges, including the Lorenz Energy Cycle, which describes the energy exchanges in the extratropical synoptic eddies. We also provide an estimate of the atmospheric material entropy production, i.e. the entropy production through irreversible processes, and the water mass budget, which is known to be one of the main sources of uncertainty for the modeled energy budget.

The diagnostic tool is currently being ported from version 1 to version 2 of ESMValTool, and will be hopefully soon publicly released. A report for the ESMValTool version 2 is being written, with contributions by all groups in the community, and another paper, focused on potential applications of the tool in various fields of climate science, will be submitted.

Metrics and Diagnostics for model improvements

The proper protocol and experiment setup for numerical experiments is crucial.

Nikolay Koldunov, Postdoc in S1

I am Nikolay Koldunov, Post Doc at MARUM and Alfred Wegener Institute. Since October I begin to work at Research Project S1: Diagnosis and Metrics in Climate Models. The main aim of the project is to integrate and synthesize work done in other parts of the TRR181. In particular we will provide metrics and diagnostics to help access the impact of model improvements suggested by TRR181 on quality of the climate models. One of the main challenges is to create model diagnostics that would not only quantify improvements, but also allow to clearly identify the cause of changes in model behavior. In this respect the proper protocol and experiment setup for numerical experiments is crucial and its development will be important part of my work. The resulting diagnostics will become available for the wider research community through the ESMValTool, that is going to be one of the main instruments of model analysis for the CMIP6 project.