A reverse engineering approach to optimal design of site investigation schemes and monitoring networks
Zusammenfassung der Projektergebnisse
The goal of experimental design is to investigate a system such that maximum information is achieved from a given, limited budget for investigation. Looking at groundwater as a drinking water resource, its quality and safety is hard to assess and predict mainly because subsurface materials are heterogeneous, and there are too few data to resolve this heterogeneity. Experimental design can help to optimize subsurface exploration, maximizing the desired prediction confidence of groundwater quality in spite of limited exploration budgets. In specific, optimal design can provide optimal data collection strategies in order to test, verify, discriminate and calibrate increasingly complex and data-hungry models. However, existing computational techniques for experimental design are heavily limited by their computer demands. Alternatively, existing methods sacrifice accuracy in trade for computational efficiency, e.g., by working with linearization techniques. This project developed highly accurate and highly flexible methods for optimal design of experiments. The major focus was not only on computational efficiency and accuracy, but also on versatility (the methods can be applied to many different systems in many different scientific disciplines) and robustness (the method works well even under adverse conditions). To achieve this goal, the new methods include novel concepts and computational tools to predict and evaluate the amount of information that can be gained from collecting additional data. The methods were at first developed for applications in the area of contaminant hydrogeology. In order to prove their versatility, they were successfully applied to other scientific fields, including thermodynamics, surface water – groundwater interaction and soil-plant modelling. The main conclusions from the project include: 1. The additional accuracy of the developed methods leads to significantly more effective experimental designs, because conventional methods are blind to certain types of information that is offered by data. 2. The core idea of the project was to reverse the manner of thinking about information in data: instead of assessing how possible data values (yet to be collected) could contribute to knowledge on a model prediction, we assessed how the possibly observable data values would change if we already knew the actual system behavior that the model tries to predict. While this is a non-intuitive new way of thinking, it offers many computational advantages on a methodological level. 3. Optimal experimental design is a form of decision making that involves an initially insufficient state of knowledge – if one would already know enough about a system, one would not plan to collect additional data. We found that a complex and allegedly superior type of optimization (called global optimization) uses its superiority when applied to the situation with insufficient knowledge about the system. Instead, it is more advantageous to use a piece-wise and simpler type of optimization that is allegedly inferior to global optimization, and then to use the piece-wise parcels of data collected in between to improve the knowledge state. The results have successfully been transferred to other projects within the same discipline, and also to other disciplines. Many questions for future research have been identified, and have delivered contributions to several current proposal initiatives that range from individual projects to large-scale efforts.
Projektbezogene Publikationen (Auswahl)
- A flexible Bayesian assessment for the expected impact of data on prediction confidence for optimal sampling designs, European Geosciences Union (EGU) - General Assembly, 2010
P. Leube, A. Geiges and W. Nowak
- A reverse engineering approach to optimal design of site investigation schemes and monitoring networks, presented at International Conference on Simulation Technology, Stuttgart, 14-17. Jun, 2011
A. Geiges and W. Nowak
- Entropy-based Utility Functions and efficient implementations. Abstract H21H-04, presented at 2011 Fall Meeting, AGU, San Francisco, Calif., 5-9 Dec., 2011
A. Geiges and W. Nowak
- A reverse analysis framework for the assessment of data worth in optimal design. Abstract H11K-05, presented at 2012 Fall Meeting, AGU, San Francisco, Calif., 3-7 Dec., 2012
A. Geiges and W. Nowak
- Bayesian assessment of the expected data impact on prediction confidence in optimal sampling design. Water Resources Research 48(2), 2012, W02501
P. Leube, A. Geiges, W. Nowak
(Siehe online unter https://doi.org/10.1029/2010WR010137) - Challenges for the sequential interaction between optimal design of field campaign and model calibration for non-linear systems, Abstract H24F-03, presented at 2013 Fall Meeting, AGU, San Francisco, Calif., 9-13 Dec., 2013
A. Geiges, W. Nowak and Y .Rubin
- Optimizing monitoring design to increase predictive reliability of groundwater flow models at different scales. European Geosciences Union (EGU) - General Assembly, 2013
T. Wöhling, M. J. Gosses, M. L. Perez, A. Geiges, C. R. Moore, K. Osenbrück, D. M. Schott
- Towards optimizing experiments for maximum-confidence model selection between different soilplant models. Procedia Environmental Sciences 19(0), 514–523, 2013
Th. Wöhling, A. Geiges, W. Nowak, S. Gayler, P. Högy, H. D. Wizemann
(Siehe online unter https://doi.org/10.1016/j.proenv.2013.06.058) - A reverse engineering approach to optimize the design of experiments and field campaigns, presented at XX. International Conference on Computational Methods in Water Resources (CMWR), Stuttgart, 10-13 Jun., 2014
A. Geiges and W. Nowak
- Comparing linear and nonlinear methods for more reliable predictive uncertainty quantification and optimal design of experiments. 2014 Fall Meeting, AGU, 2014
T. Wöhling, A. Geiges, M. Gosses, W. Nowak
- Efficient concepts for optimal experimental design in nonlinear environmental systems = Effiziente Konzepte für die optimale Versuchsplanung in nichtlinearen Umweltsystemen. Ph.D. dissertation School of Civil and Environmental Engineering, University of Stuttgart, 2014. - 978-3-942036-42-9 (Mitteilungen / Institut für Wasser- und Umweltsystemmodellierung, Universität Stuttgart; 238)
A. Geiges
(Siehe online unter https://dx.doi.org/10.18419/opus-612) - Eine Gruppenbeitragsmethode zur Berechnung der Viskosität mittels PCP-SAFT und Entropie-Skalierung, Thermodynamik-Kolloquium, VDI-Gesellschaft, Stuttgart, 2014
O. Lötgering-Lin, A. Geiges, M. Hopp, W. Nowak, J. Groß
