Bayesian Optimization Framework for Cost-Effective Control and Research of Non-Point-Source Sediment

Abstract

Rural nonpoint sources of water pollution are particularly difficult to control, with relatively little progress having been made compared to point sources. Management choices are difficult because of large uncertainties in both the monitoring of nonpoint pollution and the effectiveness of various actions to reduce that pollution. This study includes a proposed framework for selecting the optimal combination of research, monitoring, and management actions. The approach combines Bayesian inference and multiobjective linear programming to explicitly represent uncertainty in the effectiveness and cost of controls and to quantify the value of reducing uncertainty through research and monitoring. The authors illustrate the framework using the problem of reducing turbidity from rural sediment sources in the Minnesota River basin. The results show that a combination of research methods in different subbasins usually yields the most valuable information and is predicted to result in benefits via reduced cost and increased effectiveness of sediment reduction.