Groundwater monitoring and water level prediction have been challenging issues due to the complexity of groundwater movement. Simplified numerical simulation models have been used to represent the groundwater system; these models however only provide a conservative approximation of the system and may not always capture the local variations. Several other efforts such as coupling groundwater models with hydrological models and using geostatistical methods are being practiced to accurately predict the groundwater levels. In this study, the authors present a novel application of a geostatistical tool on residuals of the groundwater model. The kriging method was applied to the residuals of the numerical model (MODFLOW) generated by the TWDB (Texas Water Development Board) for the Edwards–Trinity (Plateau) aquifer. The study was done for the years 1995 through 2000 where 90% of the observation data was used for model simulation followed by cross-validation with the remaining 10% of the observations. The kriging method reduced the average absolute error of approximately 31 m (for MODFLOW simulation) to less than 5 m. Furthermore, the residuals’ average standard error was reduced from 9.7 to 4.7. The use of the kriging technique thus can provide improved monitoring of groundwater levels resulting in more accurate potentiometric surface maps.
