In decision-making situations characterized by uncertainty and risk, there often exists an opportunity to gather additional information, contributing to better informed decisions. Acquiring more information, while potentially reducing uncertainty and enhancing decision outcomes, often entails significant costs, emphasizing the importance of optimal information acquisition. Choosing a data acquisition strategy entails not only deciding whether to invest in information but also involves identifying the type of information to acquire and determining the optimal locations for collecting that information. 

The motivating example for the proposed research project centers on the influence of quick clay, a sensitive glaciomarine clay, on slope stability, involving the risk of landslides and prompting decision-makers to contemplate potential actions such as slope reinforcement or mitigation measures. To improve decision-making in this context, it is crucial to acquire essential information about the presence and distribution of quick clay. In this context, the primary aims of this project focus on devising effective methodologies to identify optimal data acquisition strategies. The work is segmented into two key parts: one dedicated to the Value of Information (VoI), and the other concentrating on the Expected Value of Partial Perfect Information (EVPPI). In the first subproject, the focus is on assessing the VoI through the application of an innovative Energy-Based Model (EBM) approach, introduced for the efficient estimation of rare event probabilities. The EBM approach stands out by transforming a potentially high-dimensional inverse problem into the optimization of a one dimensional function and can be applied in VoI analysis to effectively calculate failure probabilities. Efficiency in the VoI context is improved by utilizing estimates of this one-dimensional function across different data scenarios. In the second subproject, the value of perfect local information is quantified using the EVPPI.

While traditionally employed in engineering applications to quantify the value of information regarding system components, the EVPPI can also be applied in a spatial context to assess the value of information about a specific location. With a focus on sequential data acquisition, approaches for EVPPI evaluation that involve efficient sampling and Gaussian approximations are tested. The methods developed in this project are expected to significantly speed up VoI and EVPPI calculations, with the hope that their implementation can enable more frequent usage in practice. While the quick clay scenario is used as an illustrative example, the methodology presented is applicable across a wide range of decision contexts.