From the Average Nearest Neighbor test, there
is statistically significant clustering in the
drumlins (Graph 1).
There was a strong linear correlation found
between the area of a unit in the bedrock
geology and the number of drumlins it
contained. The Student’s T-Test resulted in a
14% likelihood that the density of drumlins was
due to random chance. Since it was larger than
5%, it is not statistically significant-i.e.,
drumlins are evenly distributed over each
type of bedrock.
There was a less strong linear correlation found between the area of a unit in the surficial
geology and the number of drumlins it contained. Furthermore, for the surficial geology, the
Students T test gave a 2.3% likelihood that the density of drumlins was the result of random
chance, and since it is less than 5%, it is statistically significant. i.e., drumlins are not
evenly distributed over each type of surficial geology. Indeed, a disproportionately large
number of drumlins are found above till. This is unsurprising since drumlins are made of till.
A disproportionately small number of drumlins are found over drift poor sediment.
The underlying process for the formation of drumlins is still unknown. Various
explanations have been suggested as to how drumlins begin to form, including catastrophic
flooding of highly pressurized glacial water [1], or small changes in the overlying glacier
creating distinct points of nucleation [2]. However, the available studies did not investigate
how the underlying bedrock affects drumlin formation. By using statistical methods, it may
be possible to determine whether there is structure within the spatial distribution of
drumlins or correlation with the underlying bedrock which would better allow the process
of nucleation to be investigated.
Drumlins in Newfoundland and Labrador were selected as the study dataset due to the
large number of drumlins in the area and quality of data. Drumlins were modeled as point
features, and statistical tests were performed to observe (1) whether the distribution of
drumlins was random, and (2) if the density of drumlins over any given rock type was
statistically different from the mean density.
The drumlin data was extracted from the Geological Survey of Canada glacial data by
performing a SQL query. After inspecting the data, the area of study (required for the density
tests) was chosen as the main landmass from the same dataset.
The first statistical test was Average Nearest Neighbor, which tests how the points are
distributed by taking the average distance of the nearest neighbor from every point.
Comparing this against knowns distribution can help determine if the points are randomly
distributed or have structure associated with them, such as clustering or dispersion. The
“Multi Distance Spatial Analysis (Ripley’s K Function)” was also used to test clustering or
dispersion at a variety of scales, but due to software issues the tool did not yield useable
results.
Next, it was investigated whether the drumlins had higher densities over any geologic units.
For both surficial geology and bedrock geology, the number of drumlins in any given rock
unit was found, and compared against the area of the rock unit in the area of study. A density
of drumlins over each rock unit was found. The effectiveness of the linear correlation was
found by using Students T- Test [3] to compare the average density of drumlins over each
rock type to the average density of drumlins over the full area of study.
The distribution of drumlins in Newfoundland and Labrador are clustered, with no relation
to the underlying bedrock. There is a relation to the underlying surficial geology, in that
drumlins are more likely to occur above till. The analysis of the distribution of drumlins
over the underlying rock could be improved if more distinctive criteria were selected, i.e. by
choosing more specific categories of underlying rock.
Additional data linking drumlins with the hardness of the rocks they are on top of could be
used to produce a more subtle analysis. Furthermore, a similar type of analysis could be
performed on a larger area with more accuracy, and could also be performed on any
presumed random point feature, for instance on the distribution of glacial erratics.
Introduction
Data Sources:
Klassen, R.A., Paradis, S., Bolduc, A.M., and Thomas, R.D.
1992: Glacial landforms and deposits, Labrador, Newfoundland and
eastern Québec; Geological Survey of Canada, Map 1814A,
scale 1:1 000 000.
Wheeler, J.O., Hoffman, P.F., Card, K.D., Davidson, A., Sanford,
B.V., Okulitch, A.V., and Roest, W.R. (comp.)
1997: Geological Map of Canada, Geological Survey of Canada,
Map D1860A.
Publications:
[1]Shaw, John, and Robert Gilbert. "Evidence for large-scale
subglacial meltwater flood events in southern Ontario and
northern New York State." Geology 18.12 (1990): 1169. Web.
[2] John Menzies, Dale P. Hess, Jessey M. Rice, Kaleb G. Wagner,
Edouard Ravier, A case study in the New York Drumlin
Field, an investigation using microsedimentology, resulting
in the refinement of a theory of drumlin formation,
Sedimentary Geology, Volume 338, 1 June 2016, Pages 84-
96, ISSN 0037-0738, https://doi.org/10.1016/
j.sedgeo.2016.01.017.
[3]Wong, David S., and Jay Lee. Statistical Analysis of Geographic
Information. Hoboken: John Wiley & Sons, 2005. Print.
Drum’lin up some Statistics: Geostatistics of Drumlins
in Newfoundland and Labrador.
By Zachary Kaplan
Sources
Methods
Results
Map 1: Map of the area of study and all of the drumlins within that area.
Graph 1: The result of the Average Nearest Neighbor test
Map 2: Map of the drumlins overlaid above the bedrock
units
Graph 2: plot of number of drumlins in a given bedrock
unit vs. its area
Graph 3: plot of the number of drumlins in a given
surficial geology unit vs. its area
Map 3: Map of the drumlins overlaid above the surficial
geology units
Conclusions