Statistical Analysis of Uncertainty Propagation and Model Accuracy

Abstract

Until recently the subjects of model uncertainty and prediction accuracy were largely ignored by water-quality modelers. There were many reasons for this, including a widespread conviction that model predictions could be made as accurate as desired simply by increasing the detail and complexity of the governing equations. Enthusiasm for complex model structures led to a proliferation of sophisticated ecosystem models, which grew larger and larger and included more and more biological compartments, chemical interactions, etc. Unfortunately, increases in model size and complexity did not necessarily provide the expected improvements in prediction accuracy. If anything, they made the models more difficult to use and the results harder to interpret. It became apparent that the primary factor limiting model performance in many applications was not lack of detail but rather insufficiently accurate model inputs.