What do we need to trust in models?
Abstract. Scientists working with numerical models may notice that their presentations of numerical results to non-specialists sometimes unfold substantial persuasive power. It seems obvious that someone has worked intensively on a topic, bundled information and solved complicated equations on a high-performance computer. The final result is a number, a curve or a three-dimensional representation. The computer has made no mistake, so the result can certainly be trusted. But can it? Those who do the modelling often know the weak points of their models and invest time in increasing the reliability of the model calculation. Trust in model calculations is usually based on rigorous quality assurance of data, programs, simulation calculations and result analyses. It requires appropriate handling of uncertainties. In view of the simplifications and idealizations of models it is also necessary to assess which model results are actually meaningful. Additionally, in most cases simplified or idealised models have been used and it is necessary to assess which model results are actually meaningful. We want to discuss what it takes to generate simulation results that can be considered reliable and how scientists can appropriately convey their confidence in their own models in discussions with the public. The framework of the discussion is provided by an introduction from Martin Navarro und Ingo Kock (BASE) and we are happy to have brief input from Thomas Nagel (TUBAF), Klaus-Jürgen Röhlig (TUC) and Wolfram Rühaak (BGE).