BULLETIN NO. 68


relative to MSL. The database also records
uncertainty in the unit elevations. Synthetic
wells were added to the database to serve as
contour control along the margins of the study
area in order to facilitate a more accurate
interpolated surface. Quality assurance of all
well-head elevations was accomplished by
comparing the reported elevation with a 15 m
(49 ft) DEM. The data on which the maps and
cross sections are based is available from
http://ww.uflib.ufl.edu/ufdc/?b UF00087428&
v=00001

Map Interpolation and
Spatial Accuracy

 As the GIS database was developed,
preliminary contour maps were generated to
allow identification of anomalous elevations and
data gaps. ArcView with the Spatial Analyst
extension was used to generate the initial
contour maps. For each map, a grid surface
model was calculated using a surface
interpolation method. Contours were then
generated from the surface model. The inverse
distance weighted (IDW) interpolator was found
to provide a very useful surface model for
identifying problem areas and outliers that
required additional research. As these issues
were identified, borehole cuttings were retrieved
from the FDEP-FGS core repository and re-
evaluated in context of: 1) proximal stratigraphy,
2) sample quality, and 3) stratigraphic boundary
uncertainty. This process of generating and
reviewing maps and re-examining samples was
repeated numerous times through the course of
this study. Arthur and Pollock (1998) suggested
that the IDW interpolation method is not suitable
for geologic mapping of stratigraphic data. The
IDW method was used only as an iterative
review tool. Interim maps for the initial phases
of this project were generated using the spline
interpolator. Arthur and Pollock (1998) found
the spline-tension method preferable to IDW
because spline yielded surface models that were
more accurate and geologically characteristic.

 Upon review of subsequent interim maps
generated for this project, several shortcomings
in application of the spline interpolator were
recognized. These include false highs and lows


in the model surface that do not reflect any well
control and loss of contour accuracy along the
margins of the maps. Moreover, as the
interpolator was adjusted to yield smoother
contours, the interpolated surface became less
accurate. With the exception of accuracy at a
given well location, it was also difficult to assess
the error associated with the maps in data-poor
areas. Due to the "overshooting contour" effect
(e.g., false topographic highs), shallow map
units may exceed land surface elevations in
certain areas, especially along areas of high
topographic relief.

 Prior to generation of the final maps, a re-
assessment of interpolation methods was
completed. Through an iterative process,
kriging was identified as the most robust and
accurate surface interpolator. The GIS project
was then migrated to ESRI ArcMap 8.3. The
map-unit elevation database was imported into
ArcMap and projected in the FDEP standard
projection (custom Albers equal-area conic
projection). Individual surfaces were then
interpolated using the ordinary kriging function
within the ESRI Geostatistical Analyst.
Iterations of krige data models were produced
for each map unit to minimize the map error.
Prediction errors and other descriptive statistics
for each map were recorded and used to identify
appropriate contour intervals (see Contour
interval selection, p. 30). When evaluating krige
statistics, the variability of the prediction (as
standard deviation) is overestimated when the
root mean squared prediction error (RMS) is less
than the average standard error (ASE) of the
prediction error (Table 3; Johnston et al., 2001).

 The krige error automatically reported by the
software is based on the error within a
rectangular area defined by the distribution of
the data (i.e., wells). For nearly all of the maps
in this report, these results are misleading
because the datasets have an irregular spatial
distribution. As a result, for each kriged surface,
a prediction standard error map was also created
and masked to the spatial extent of the unit being
mapped. The average standard error of this
prediction standard error map was then
calculated and recorded (Table 3).