FLORIDA GEOLOGICAL SURVEY


 As an extra measure to assess map accuracy,
a script was written to allow comparison of the
observed map-unit elevations (i.e., top of the
Suwannee Limestone in a given well) with the
value of the final interpolated grid cell in which
the well is located. On Table 3, this data is
summarized in the "Grid to Point" column,
where the "mean" represents the average of the
difference between grid cells and map unit
elevations for each well located within its
respective grid cell. In every case, the mean
"Grid to Point" value is less than +3 ft, (0.9 m)
for all maps, and the standard deviation (s) is
less than 20 ft (6.1 m) for 18 of the 22 maps and
less than or equal to 30 ft (9.1 m) for the
remaining 4 maps (Table 3). Qualitative
evaluation of these errors suggests they are
normally distributed.

 The standard deviations for the "Grid to
Point" calculations (Table 3) are well within
acceptable limits, especially when considering:
1) geologic processes that can create
perturbations in a mapped surface (e.g., faults,
paleo-karst, paleo-environmental features [e.g.,
wave-cut scarps, river valleys]), 2) well location
error or uncertainty, 3) sample quality (e.g.,
cuttings interval and borehole cavings) and 4)
formation pick uncertainty (i.e., gradational
contacts, differences in professional opinion,
etc.).

 Once the kriged surfaces were optimized for
each map unit, the shallow surfaces (i.e., top of
the Peace River Formation) were compared to
land surface elevations. To accomplish this, the
kriged surface was converted to a raster file
(grid) using a 400 m2 (4305.6 ft2) cell size,
which was then subtracted from the FDEP-FGS
15 m (49.2 ft.) resolution DEM to remove
interpolated elevations that exceeded land
surface. The grid was clipped (i.e., masked) to
the lateral extent of its respective map unit. For
most of the mapped units, contours generated
from the kriged and the DEM-trimmed surfaces
were generally irregular or jagged. This
characteristic is not only atypical for maps
depicting subsurface elevations and thicknesses,


but it also overemphasizes the level of resolution
represented by the maps. To remove these
localized and misrepresentative contour
anomalies, the grids were smoothed using the
neighborhood statistics function in Spatial
Analyst. Color shading, contour lines and labels
were then added.

 It is noteworthy that substantial effort was
devoted to surface interpolations that would
allow weighting of data from cores
preferentially over that of cuttings and
geophysical logs. Combinations of grid
averages, weights, and point buffering applied
within various interpolators were evaluated
during extensive sensitivity analyses and
validation. In the final assessment, too many
negative attributes were associated with what
was considered the optimal core-weighting
technique. As a result, the well data from cores,
cuttings and geophysical logs are all considered
equally in the maps.

 Unlike hand-drawn contours, the smoothed
contours generated from krige-interpolated
surfaces do not always reflect highly anomalous
elevation data points. While this may be
considered a disadvantage, it is substantially
outweighed by numerous advantages of the
digital products developed during this study: 1)
the interpolated surfaces are supported by
accuracy and precision statistics, 2) the
calculated grids can be used in a variety of
groundwater flow models and 3D applications,
3) GIS compatibility exists, including inherent
scalability and flexibility, and 4) the maps can
be readily updated with new information. As a
result, manual modification (editing) of the
contours to reflect anomalous values would
create discrepancies between the maps and the
grid coverages. In cases where contour
adjustments were deemed necessary to reflect
sharp-relief surface trends (as opposed to single
anomalous well values), synthetic control points
were added to improve map accuracy while
maintaining consistency with the calculated
grids.