Model parameter estimation with imprecise information

Area of Science:

• Environmental modeling
• Statistical inference
• Hydrology

Background:

• Model parameter estimation is a challenging inverse problem, especially with imprecise measurements.
• Classical statistical methods struggle with non-standard data like censored or binary observations.
• Accurate parameter estimation is crucial for reliable system performance assessment.

Purpose of the Study:

• To introduce Approximate Bayesian Computation (ABC) as a viable method for parameter estimation with imprecise data.
• To demonstrate ABC's application using a rainfall-runoff model.
• To evaluate the contribution of different observation types to parameter estimation using Shapley values.

Main Methods:

• Utilized Approximate Bayesian Computation (ABC) for model parameter estimation.
• Applied the method to a standard rainfall-runoff model.
• Employed Shapley values to analyze the importance of various data types in parameter estimation.

Main Results:

• ABC successfully performed parameter estimation with imprecise data (censored and binary).
• The study illustrated the practical advantages and limitations of using ABC in this context.
• Shapley values effectively identified key observational data driving parameter estimation.

Conclusions:

• Approximate Bayesian Computation provides a flexible and effective approach for model parameter estimation when dealing with imprecise observational data.
• The rainfall-runoff model case study highlights ABC's utility and potential challenges.
• Shapley value analysis offers valuable insights into data utility for robust parameter estimation.