kriging estimates for all points in a 26-by-26 grid

with a grid size of about 61-by-61 m. A gray-

values in a more quantitative manner, the investi-

scale map of the kriging water levels is shown in

gator needs to establish some assurance that the

Figure 5-3a and basic univariate kriging estimate

measured data and the reduced kriging errors are

statistics are listed in Table 5-2a (water level A).

approximately normally distributed and also that

The kriging results a are a good representation of

the assumption of stationary residuals after drift

the results from other more elaborate studies.

removal is correct. If the investigator is confident

about these assumptions, then the basic statistical

principles involving confidence intervals can be

estimates are shown in Figure 5-3b. The magni-

applied. In this example, the standard deviation of

tude of kriging standard deviations can provide

about 0.35 throughout most of the map indicates

investigators with a direct indication of where the

that there is a 95-percent chance that the true value

uncertainty associated with kriging estimates is

at a location where there is a kriging estimate will

relatively high or low. The areas of greatest uncer-

be within about 0.70 (twice the kriging standard

tainty for the kriged water levels are in the upper

deviation) of the kriging estimate.

right and lower left corners of the map, where

standard deviations are as high as about 1.4 and

0.8. Not surprisingly, these areas are where the

density and the accuracy of kriging estimates, two

density of the measured data is relatively low.

new maps were developed. To make the first map,

Throughout much of the remainder (about 70 per-

a decrease in network density was effected by

cent) of the map, the kriging standard deviation is

removing nine measured locations from the north-

almost constant at about 0.35.

west part of the area (Figure 5-1b) where sampling

5-5

Integrated Publishing, Inc. |