A complementary modelling approach to manage uncertainty of computationally expensive models

Area of Science:

• Environmental modeling
• Water resource management
• Computational hydrology

Background:

• Models in the water domain often deviate from reality, leading to uncertainty in results.
• Understanding model error is crucial for assessing confidence and optimizing parameter estimation.
• Traditional model calibration methods can be insufficient, especially for complex, computationally intensive models.

Purpose of the Study:

• To systematically analyze uncertainties in water domain models.
• To evaluate common approaches for model parameter estimation.
• To propose an alternative, complementary modeling strategy to enhance accuracy and confidence.

Main Methods:

• Analysis of model error to quantify discrepancies between models and physical processes.
• Discussion of four conventional methods for estimating model parameters.
• Development of a hybrid approach combining a physically-based model with a Support Vector Machine (SVM) model.

Main Results:

• The study highlights the limitations of traditional calibration methods when faced with significant model uncertainty.
• The proposed complementary approach integrates diverse modeling techniques to improve overall solution reliability.
• Combining physically-based and SVM models offers a more robust estimation of optimal model parameters.

Conclusions:

• Uncertainty analysis is vital for interpreting model outputs in water resource applications.
• A complementary modeling strategy, integrating physical and machine learning approaches, can overcome limitations of traditional methods.
• The proposed hybrid model enhances confidence in parameter estimation for complex environmental systems.