Comparative Error Analysis of Six Predator‐Prey Models

Six models were devised for a hypothetical predator—prey system. The models differed in functional forms, but were calibrated to display similar output for a standard annual cycle. Parameters were assumed to be measured with error, and this uncertainty was propagated through the models by Monte Carlo simulation. Total uncertainty on model output was partitioned, for each model, into six component sums of squares, and components were examined as they changed over the simulated annual cycle. The results indicate that the models differ in their error propagation properties with more complex functions often generating greater uncertainty on output. The contribution of a specific model term to overall error is dependent on its mathematical form (especially, the partial derivative of the term with respect to the parameter) and also its position within the total model. No single model showed the smallest output uncertainty for all possible modeling objectives. The results indicate that the most useful data for testing a model should be gathered during periods of the year when the state variables are changing most rapidly.