Premium
Bayesian model averaging for groundwater head prediction and uncertainty analysis using multimodel and multimethod
Author(s) -
Li Xiaobao,
Tsai Frank T.C.
Publication year - 2009
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2008wr007488
Subject(s) - uncertainty analysis , hydraulic conductivity , variance (accounting) , bayesian probability , occam , statistics , bayesian inference , computer science , groundwater model , head (geology) , groundwater , econometrics , environmental science , mathematics , aquifer , soil science , groundwater flow , engineering , geology , geotechnical engineering , geomorphology , accounting , business , soil water , programming language
This study introduces a Bayesian model averaging (BMA) method that incorporates multiple groundwater models and multiple hydraulic conductivity estimation methods to predict groundwater heads and evaluate prediction uncertainty. BMA is able to distinguish prediction uncertainty arising from individual models, between models, and between methods. Moreover, BMA is able to identify unfavorable models even though they may present small prediction uncertainty. Uncertainty propagation, from model parameter uncertainty to model prediction uncertainty, can also be studied through BMA. This study adopts a variance window to obtain reasonable BMA weights for the best models, which are usually exaggerated by Occam's window. Results from a synthetic case study show that BMA with the variance window can provide better head prediction than individual models, or at least can obtain better predictions close to the best model. The BMA was applied to predicting groundwater heads in the “1500‐foot” sand of the Baton Rouge area in Louisiana. Head prediction uncertainty was assessed by the BMA prediction variance. BMA confirms that large head prediction uncertainty occurs at areas lacking head observations and hydraulic conductivity measurements. Further study in these areas is necessary to reduce head prediction uncertainty.