COMET – Computational Methods for Climate Impact Research


COMET is a cross-cutting research activity with a distinct focus on modelling, simulation and visual data analysis. It supports domain experts with innovative methods and techniques from both theoretical and applied computing science.

Speaker: Michael Flechsig

Team members: Nicola BottaThomas Nocke

Modelling chain

Motivation & Goals

PIK's mission substantially relies on systems and scenarios analysis, modelling, computer simulation and data integration. COMET supports the integration of these methodologies in PIK's modelling strategy. It provides methods for model specification, quality assurance, visualization of climate data and visual analytics. We

  • address crucial steps of the modelling chain in collaboration with international partners,
  • develop novel concepts and implement specific tools, and
  • apply them in cooperation with all PIK Research Domains.

Formal Specification

Climate research relies on increasingly complex models. While more complexity may make models more realistic, it also makes them more difficult to understand, implement and communicate. With increasing complexity, it becomes harder to assess whether a model implementation is correct, make sure that the input data is consistent with the model assumptions and independently reproduce and confirm results obtained by numerical simulation.

Managing complexity is one of the central themes of computing science. Until recent years, however, the type systems of mainstream programming languages were not expressive enough to support precise specifications.  Expressing the assumptions a model relies upon, specifying what a model implementation is required to do, proving that it actually does what it is required to do, was well beyond the reach of modelers and, in fact, even of computer scientists.

As functional languages and, more recently, dependently typed languages become mainstream programming tools, the situation changes dramatically. We have now methods and techniques which allow us to precisely express what our models are supposed to do, make model assumptions visible and prove (or, at least, systematically test) the correctness of our model implementations.

In collaboration with partners at Chalmers (Göteborg, Sweden) and St. Andrews (St Andrews, Scotland) we develop and apply specification methods for climate impact research. These include functional specifications, c++ generic algorithms and domain specific contracts (e.g., for multi-agent models of exchange [1]), dependently typed frameworks (e.g., for computing machine checkable optimal policies for monadic sequential decision problems [3]) and, in the near future, domain specific languages for modelling international environmental agreements.

Sensitivity & Uncertainty

The complexity of the Earth system, the intrinsic variability of processes and our limited knowledge about the considered sub-systems demand a careful quality assurance of the findings from modeling studies. In particular, quantification of uncertainty and identification of sensitive processes, parameters or initial values are of continuous importance when studying climate (impact) phenomena by simulation models.

To support modelers and analysts we have developed the multi-run simulation environment SimEnv [4] mainly for sensitivity and uncertainty analyses of model output. Generic experiment types are based on probabilistic, deterministic and Bayesian sampling schemes in factor spaces and allow for flexible experimentation and later analyses. SimEnv has been applied at PIK to 30+ models (e.g., [5]) with 13.000 experiments and a total of 3.5M single runs.

Selected Features of the Simulation Environment SimEnv

SimEnv workflow

  • Large-scale experiment settings:
    Support of high-dimensional (100+) model factor spaces and large-volume (GB range) of multi-dimensional (<=9) and -variate (100+) model output
  • Model interface for 10 programming languages
    Easy model coupling to the environment by implementing for each factor and each output field a function call from the SimEnv library
  • Flexible simulation load distribution strategies:
    Parallelization of the experiment on a compute cluster or on a multi-core machine
  • Interactive experiment post-processor
    Navigate the coupled factor – state space and derive interactively uncertainty / sensitivity measures from post-processed model output.
  • Coupled visualization wizard SimEnvVis

nullVisualization & Visual Analytics

Visualization is an established flexible tool to analyze climate related data and to communicate climate impact research findings to decision makers and the public. However, it faces multiple challenges:

  • the very large, heterogeneous data sets,
  • the multiple tasks to be performed [6], and
  • the complexity of the scientific knowledge.


Spherical node link representation of a climate network with 3D arcs


We tackle these problems by making state-of-the-art visualization techniques and tools – including own developments – easy to use for scientists:

  • visual analysis of ensemble climate simulations data/ uncertainty visualization (further information you can find here),
  • climate network visualization (NEONET), and
  • climate and impact knowledge dissemination/ web platforms (

Key Publications

  1. N. Botta, A. Mandel, C. Ionescu, M. Hofmann, D. Lincke, S. Schupp, C. Jaeger (2011) A functional framework for agent-based models of exchange. Appl. Math. and Comp., Vol. 218, 8.
  2. C. Ionescu, P. Jansson (2013) Dependently-typed programming in scientific computing: Examples from economic modelling. In R. Hinze (Ed.). 24th Symposium on Implementation and Application of Functional Languages (IFL 2012), LNCS. Springer.
  3. N. Botta, C. Ionescu, E. Brady (2013) Sequential decision problems, dependently typed solutions. Proceedings of the Conferences on Intelligent Computer Mathematics (CICM 2013), "Programming Languages for Mechanized Mathematics Systems Workshop (PLMMS)".
  4. M. Flechsig, U. Böhm, T. Nocke, C. Rachimow (2013) The multi-run simulation environment SimEnv. User’s Guide V3.1. PIK Potsdam,
  5. M. van Oijen, … M. Flechsig, … (2013)  Bayesian calibration, comparison and averaging of six forest models, using data from Scots pine stands across Europe. Forest Ecol. and Managem., 289: 255-268.
  6. H.-J. Schulz, T. Nocke, M. Heitzler, H. Schumann (2013) A design space of visualization tasks. IEEE Transactions on Visualization and Computer Graphics 19(12), 2366-2375, DOI: 10.1109/TVCG.2013.120, at IEEE InfoVis 2013.
  7. T. Nocke, M. Flechsig, U. Böhm (2007) Visual exploration and evaluation of climate-related simulation data. Proc. Winter Simulation Conference (WSC'07), Washington D.C.
  8. C. Tominski, J. Donges, T. Nocke (2011) Information visualization in climate research. In International Conference on Information Visualisation (IV’11), London.


Climate Risk for Asset Managers (CRAMs)
March 2017 – December 2018, Funded by: Climate-KIC
Contact: Thomas Nocke
Mixed Methods
Mixed Methods in den Geisteswissenschaften?
April 2017 – March 2020, Funded by: VW-Stiftung
Contact: Thomas Nocke

Document Actions