Application of the Discrimination Inference to Reduce Expected Cost Technique (DIRECT) to a Contaminant Transport Problem.

Abstract:

There is growing recognition in the hydrologic community that deterministic hydrologic models are imperfect tools for decision support. Despite this insight, the state of practice for a hydrologic investigation follows this sequence: data collection, conceptual model development, numerical model development, and finally decision making based on model projections. This approach, based on relatively unconsidered design of data collection, may result in uninformative data. As a result, it is commonly repeated several times to resolve critical uncertainties. We present a novel two step multi-model approach to optimizing data collection to aid decision making, risk analysis. Here, we describe the application this approach (Discrimination Inference to Reduce Expected Cost Technique - DIRECT) for a contaminant transport problem. DIRECT has 7 steps. First, outcomes of concern were defined explicitly. Next a probabilistic analysis of the outcomes was conducted that incorporated multiple conceptual and parametric realizations. The likelihood of each model was assessed based on goodness of fit to existing data. A cost function was developed and used to define the projected costs based on the model-predicted outcomes of concern. Data collection was then optimized to identify the data that could test the models of greatest concern (cost) against the other models in the ensemble. Finally a field program was conducted that included gathering lithologic, hydrologic, and chemical data from 22 new wells that were drilled in projected high value locations. The additional data reduced the expected cost of model projections to an acceptable level for defining new site compliance conditions.