Comparison of Geostatistical Algorithms for Completing Groundwater Monitoring Well Timeseries Using Data of a Nearby River

Abstract

The benefit of river stage as secondary information in the kriging of groundwater level time-series is investigated for an unconfined, alluvial sand and gravel aquifer on Csepel Island in the Danube River. Factorial kriging analysis allows the filtering of the secondary information, which is then used in 3 forms: raw river data, the trend of the river data by itself or shifted by a well-specific lag time derived from river-well cross-correlograms. Cross-validation indicates that incorporation of river data using either kriging with an external drift or simple kriging with varying local means reduces the mean absolute error of prediction for 92% of the wells by an average of 18% relative to ordinary kriging. The Danube’s influence diminishes rapidly within the island, and two groups of wells are distinguished: one under the influence of the river and another, interior group. The kriging of the latter derives spurious benefit from the secondary information, possibly due to other seasonally varying influences.