Quarterly water-level measurements were analyzed to assess the effectiveness of a monitoring network of 26 wells in Huron County, Michigan. Trends were identified as constant levels and autoregressive components were computed at all wells on the basis of data from 1993 to 1997 using structural time series analysis. Fixed seasonal components were identified at 22 wells and outliers were identified at 23 wells. The 95-percent confidence intervals were forecast for water-levels during the first and second quarters of 1998. Intervals in the first quarter were consistent with 92.3 percent of the measured values. In the second quarter, measured values were with the forecast intervals only 65.4 percent of the time. Unusually low precipitation during the second quarter is thought to have contributed to the reduced reliability of the second-quarter forecasts. Spatial interrelations among wells were investigated on the basis of the autoregressive components, which were filtered to create a set of innovation sequences that were temporally uncorrelated. The empirical covariance among the innovation sequences indicated both positive and negative spatial interrelations. The negative covariance components are considered to be physically implausible and to have resulted from random sampling error. Graphical modeling, a form of multivariate analysis, was used to model the covariance structure. Results indicate that only 29 of the 325 possible partial correlations among the water-level innovations were statistically significant. The model covariance matrix, corresponding to the model partial correlation structure, contained only positive elements. This model covariance was sequentially partitioned to compute a set of partial covariance matrices that were used to rank the effectiveness of the 26 monitoring wells from greatest to least. Results, for example, indicate that about 50 percent of the uncertainty of the water-level innovations currently monitored by the 26-well network could be described by the 6 most effective wells.
Additional publication details
USGS Numbered Series
A temporal and spatial analysis of ground-water levels for effective monitoring in Huron County, Michigan