Dealing with uncertainty in ecosystem models: lessons from a complex salmon model.

Area of Science:

• Ecology
• Environmental Science
• Ecological Modeling

Background:

• Ecosystem models are crucial for environmental assessment and management.
• Increasing model complexity complicates the assessment of parameter uncertainty and its impact on predictions.
• Sensitivity analysis is vital for evaluating model reliability, especially for management decisions.

Purpose of the Study:

• To conduct a sensitivity analysis on the Ecosystem Diagnosis and Treatment (EDT) model.
• To evaluate how uncertainty in parameters and data inputs affects EDT model predictions of salmon productivity and capacity.
• To assess the robustness of EDT's prioritization of areas for conservation under uncertainty.

Main Methods:

• A novel "structured sensitivity analysis" approach was employed for the complex EDT model.
• Plausible ranges (small, medium, large) were defined for input data and model parameters.
• A Monte Carlo approach was used to explore output variations, prediction intervals, and sensitivity indices.

Main Results:

• EDT model predictions for salmon productivity and capacity were found to lack precision due to internal parameter uncertainty.
• The prioritization of specific areas for preservation or restoration by the EDT model proved more robust to input uncertainties.
• The EDT model may be more effective as a relative indicator of fish performance rather than an absolute measure.

Conclusions:

• Explicitly incorporating uncertainty and sensitivity analyses is crucial when using complex model outputs for secondary analyses or management tools.
• Sensitivity analyses should be a standard procedure for evaluating ecosystem models.
• The EDT model's strength lies in relative prioritization rather than precise absolute predictions.