Table 1.
List of symbols.
Fig 1.
Location of the study area within Flanders, Belgium (A) and the Nete basin (B), and location of CPTs and cored boreholes for the coarse (C) and fine sampling grid (D).
Fig 2.
Geological (A) and hydrogeological map (B) of the study area and its surroundings, respectively based on DOV [47] and Gulinck [48].
Fig 3.
Two example CPT logs displaying normalized cone tip resistance (Qt) and normalized friction ratio (Fr).
Their location is indicated in Fig 4. Stratigraphy is based on nearby (< 10 m) boreholes “Dessel-2” (A) and “Kasterlee-1” (B) (see Fig 1).
Fig 4.
A) Conceptual lithostratigraphic profiles through the study area. B) Top view of the location of the profiles in A and C with respect to the geometry of the top of the aquitard. C) Sideview of the CPT data (40x height exaggeration with panel dimension ~10 km x 40 m) projected orthogonally onto the NE-SW dipping plane (which corresponds to the NE-SW conceptual profile in A), with logarithmic normalized cone tip resistance (Qt) and friction ratio (Fr).
Fig 5.
SBT classification charts of Robertson et al. [4] (A) and Robertson [5] (B).
Data (~ 480,000 data points) from this study are shown as red dots.
Table 2.
SBT classification based on SBT index (Ic) ranges [47].
Fig 6.
SBTs density plot based on the SBT index Ic.
Data are from this case study with ~ 480,000 observations.
Table 3.
Number of SBT classes for the different classification approaches.
Fig 7.
Examples of an automatically selected horizon mapped by using model-based clustering and the kernel density estimates of the z coordinates of the two contrasting classes.
Fig 8.
Biplots for all 11 SBT classifications.
The first two principal components (Comp.1, Comp.2) of the nine sediment properties (“Geoz” = zstrat, “Por” = porosity, “Dens” = bulk density, “LogK” = logarithmic hydraulic conductivity, “Clay” = clay content, “Glau” = glauconite content, and “CEC” = cation exchange capacity). The SBT data are represented as individual data points and cluster centres (black numbers). The x- and y-coordinates of these points are multiplied by 3.5 to illustrate more clearly the relationship with the sediment properties. The size of the numbers is proportional to the amount of data points in the cluster.
Fig 9.
Marginal distributions for SBT classifications along the stratigraphic depth zstrat.
Stratigraphic boundaries are overlain based on an average stratigraphic column (top row, last column), and are only indicative.
Fig 10.
Side views of the CPT dataset (40x height exaggeration; ~10 km x 40 m) projected on a hypothetical plane approximately perpendicular to the layer dip, with colour-coded SBT classes.
Fig 11.
Four examples of typical SBT variograms.
The full list of variograms is provided as supplementary material (S1, S2 and S3 Figs).
Fig 12.
Example SBT logs for the CPT data displayed in Fig 3.
Fig 13.
Scatterplot of lithostratigraphic mapping of the top of the aquitard (in m below sea level) versus the manually interpreted top of the aquitard [17, 18], using A) x-means clustering and B) model-based clustering.
Fig 14.
Contour maps of the top of the aquitard, using A) a manual approach, B) the model-based and C) the x-means clustering.
Differences between the automatically and the manual derived reference values are presented in D) for the model-based and E) for the x-means clustering. Locations outside of the CPT characterization area are influenced by extrapolation.