Globalsensitivitätsanalysen

Globalsensitivitätsanalysen, often abbreviated as GSA, are a class of methods used to assess the influence of input parameters on the output of a model. Unlike local sensitivity analysis, which examines the effect of changing one input parameter at a time while holding others constant, GSA considers the effects of all input parameters and their interactions simultaneously. This is particularly important for complex models where input variables may be uncertain or correlated.

The primary goal of GSA is to identify which input factors are most influential in determining the model's output uncertainty. This helps in understanding the model's behavior, simplifying it by focusing on key parameters, and guiding data collection efforts to reduce uncertainty in critical areas. GSA can also reveal non-linear relationships and interactions between input variables that might be missed by simpler methods.

Various GSA methods exist, each with its strengths and weaknesses. Variance-based methods, such as the Sobol' indices, are widely used. These methods decompose the total variance of the model output into contributions from individual input factors and their interactions. Other approaches include regression-based methods and derivative-based global sensitivity measures. The choice of method often depends on the nature of the model, the type of input distributions, and the computational resources available. GSA is applied across many disciplines, including environmental science, engineering, economics, and climate modeling, to enhance model reliability and inform decision-making.