State-of-the-Art techniques for sensitivity analysis in geochemical modeling: exploring methods and tools: SANGUR : Systematic sensitivity analysis for mechanistic geochemical models using field data from crystalline rock

This report, a key deliverable of the SANGUR project, provides a comprehensive analysis of sensitivity analysis (SA) methodologies. Global Sensitivity Analysis (GSA) receives particular attention, with the report exploring various approaches, their pros and cons, and their suitability for geochemical modeling in this domain. The report delves into the fundamentals of GSA, highlighting its importance in evaluating the reliability of geochemical models by understanding the influence of input parameters on model outputs. A systematic investigation of popular GSA methods is presented, including correlation/regression analysis, variance-based methods (Sobol’s method, FAST/EFAST), highdimensional model representation (HDMR), and stepwise regression. Exploring these methods and their relative strengths and weaknesses helps researchers in selecting the most appropriate GSA approach for geochemical modeling tasks. The choice will depend on factors like the complexity of the model, the number of input parameters, and the desired level of detail in the sensitivity analysis. Additionally, the report explores crucial sampling methodologies for SA, including Monte Carlo (MC), Latin Hypercube Sampling (LHS), and Quasi-Monte Carlo (QMC) techniques. Each method’s principles and applicability in the context of geochemical modeling are explained. Recognizing the importance of robust uncertainty quantification practices, the report explores available GSA software tools. It examines a diverse range of options, encompassing Python libraries (SALib, Chaospy) and comprehensive toolkits (Dakota, UQTk, OpenTURNS). While offering researchers valuable resources, one observes the absence of a universal standard. The report encourages researchers to carefully consider their modeling needs and preferences when selecting a tool, as the optimal choice depends on the specific modeling task. Finally, the unique challenges associated with geochemical models are explored, particularly nonlinearities and parameter interdependencies, and their impact on SA methodologies. A dedicated section discusses these complexities and their implications for selecting appropriate SA techniques.