Parameter Estimation for Groundwater Models under Uncertain Irrigation Data.

Area of Science:

• Hydrogeology
• Environmental Modeling
• Geostatistics

Background:

• Groundwater model accuracy is crucial for subsurface characterization.
• Uncertainty in source/sink terms, like irrigation data, can bias parameter estimates and predictions using standard regression methods.

Purpose of the Study:

• To quantify bias in groundwater model parameters and predictions resulting from irrigation data errors.
• To present a novel inverse modeling technique, input uncertainty weighted least-squares (IUWLS), for unbiased parameter estimation with uncertain source/sink data.

Main Methods:

• Developed and applied the input uncertainty weighted least-squares (IUWLS) method, incorporating generalized least-squares with objective function weights adjusted for pumping uncertainty.
• Conducted analytical and numerical experiments using Republican River Basin irrigation data.
• Compared IUWLS with ordinary least-squares (OLS) under varying irrigation data uncertainty and calibration conditions.

Main Results:

• Ordinary least-squares (OLS) calibration resulted in statistically significant bias (p < 0.05) in estimated parameters and predictions, persisting across different calibration datasets and sizes.
• The proposed IUWLS method effectively minimized bias by directly accounting for irrigation pumping uncertainties during calibration.

Conclusions:

• Standard regression-based inverse modeling techniques are susceptible to bias when source/sink data are uncertain.
• The IUWLS method provides an effective and computationally efficient approach for unbiased groundwater model parameter estimation in the presence of input data uncertainty.