This research presents a methodology and an application for predicting the risk from decisions made with probabilistic Agricultural Nonpoint Source Pollution Model (AGNPS) outputs. Assuming a worst-case scenario for data availability, uniformdistributions for all appropriate AGNPS inputs can be assumed. Output distributions can be produced using a Monte Carlo simulation method at increments of input variation, ranging from 0 to 50 variation at 5 increments. Decision risk calculations arebased on the non-parametric Bootstrap resampling technique. This methodology was applied to the Hazelton Drain subwatershed (400 ha) of the Sycamore Creek watershed (27410 ha) in southcentral Michigan. Eight AGNPS input file scenarios were created withthe Hydrologic Unit/Water Quality Tool. For each AGNPS scenario, the Monte Carlo method produced 1000 AGNPS simulations for each input variation level. This created 1000 AGNPS output files, from which were extracted the values for sediment yield, sediment nitrogen, sediment phosphorus, soluble nitrogen, soluble phosphorus, soluble COD, and runoff volume at the subwatershed outlet. Output distributions were created for each AGNPS scenario, at each input variation and for each output variable. Analysis ofthe output distributions and decision risk values revealed several findings. Decisions made using water-based outputs evidence much less risk than those decisions based on sediment-based outputs, because water-based outputs are calculated using fewerequations than sediment-based outputs. Decision risk will usually increase when the input variation increases and when the two output distributions being compared are closer or have more overlap. A precise value for input variation is not needed for mostdecisions based on these AGNPS outputs.
展开▼
