An Integrated, Fast and Easily Useable Software Toolbox Allowing Comparative and Complementary Application of Various Parameter Sensitivity Analysis Methods

The analysis of parameter sensitivity in environmental models is an excellent technique to assess a model’s behavior, to determine its potential utility, to support its calibration, and to identify areas of improvement. Recent work on comparing sensitivity analysis methods shows that the methods available today are complementary, i.e. multiple methods should be used to assess a model. We present a software toolbox for global sensitivity analysis which supports the investigation of parameter sensitivity using different methods. The toolbox includes Regional Sensitivity Analysis, Morris Method, and a Sobols method. The majority of these methods require input data from a Monte-Carlo- Sampling which has to be carried out in advance, others demand for special properties of the sampling. Therefore, in most cases, huge computational effort has to be spent to generate several sampling data. To overcome this deficit the data from a single Monte- Carlo-Sampling is used to train an Artificial Neural Network (ANN) which imitates the original model. By using this approach, arbitrary samplings can be easily drawn from the ANN-based emulator. This approach also gives an objective measure of the quality of the sampling itself and provides criteria on how many samples are required to get representative results. The sensitivity toolbox is part of the OPTAS module in the Jena Adaptable Modelling System. We will present the developed sensitivity analysis toolbox and examples of its application to the hydrological model J2000 in a catchment located in Germany. Special attention is paid to the emulation of the model with the newly developed ANN approach which produced very promising results.