Statistische Modellierung von Klimasimulationsergebnissen zur Berechnung der zugehörigen Strahlungswirkung

Kurzfassung

For many studies in climate science a major part of the experimental concept is the comparison of two or more simulations to answer the question: Is there a difference? Depending on the simulation setup and results different procedures have been developed and put to use over the last decades. Now more than ever due to the extensive production of huge datasets, more sophisticated question can be asked and need to be answered. This work will give an overview of the difficulties involved in providing answers, shortly reviews some of the existing methodology and proposes some method for further study. The main part consists of the application of three selected evaluation methods, univariate t-test, principal component analysis and regression models to a specified dataset, which will illustrate the strengths and weaknesses of each. Concluding this work, several recommendations will be given for a statistically sound evaluation strategy depending on the questions asked and datasets produced.