Why preferring parametric forecasting to nonparametric methods?

Area of Science:

• Ecology
• Ecological Modeling
• Forecasting

Background:

• Nonparametric forecasting methods may outperform parametric ones in noisy nonlinear systems.
• Challenges with parametric methods include inference instability in chaotic systems and the "true model myth" discrepancy.

Purpose of the Study:

• To argue for the continued use of parametric models in ecological forecasting.
• To highlight comparative advantages of parametric over nonparametric approaches.

Main Methods:

• Bayesian model checking procedures for diagnosing parametric forecasting failure.
• Estimation of forecasting uncertainty using virtual data from fitted parametric models.
• Illustration using the theta-logistic model.

Main Results:

• Parametric models allow for diagnosis of forecasting reliability.
• Uncertainty estimation is possible with reliable parametric forecasts.
• Nonparametric techniques currently lack reliability assessment.

Conclusions:

• Parametric ecological forecasting provides diagnosable reliability and quantifiable uncertainty.
• Ecologists should continue using parametric approaches until nonparametric reliability can be assessed.