The Expected Value of Sample Information Analysis for Nonpoint Water Quality Management

There is considerable interest in watershed-based water quality protection. However, the approach can be highly information intensive, necessitating decisions about the types and amounts of data used to guide decisions. This study examines the Bayesian value of different types and amounts of sample information for reducing nutrient pollution in the Conestoga watershed of Pennsylvania, focusing on nitrogen from agricultural sources. Uncertainty is modeled from the perspective of a social planner seeking to maximize the economic efficiency of water quality control. A nested Monte Carlo procedure combined with an Evolutionary Optimization Strategy with Covariance Matrix Adaptation is used to compute resource allocation that optimizes the expected net benefit after updating for varying sample sizes and information types (broadly classified as pertaining to abatement costs, pollution fate and transport, and benefits of environmental protection). The results provide insights the returns from information investments to improve water quality management.