Download as pdf or txt
Download as pdf or txt
You are on page 1of 24

The Role of GIS and Expert Knowledge in 3-D Modelling,

Oak Ridges Moraine, Southern Ontario


C. Logan, H.A.J. Russell and D.R. Sharpe
Geological Survey of Canada, 601 Booth Street, Ottawa, ON, Canada, K1E 0E8
emails: clogan@nrcan.gc.ca; hrussell@nrcan.gc.ca; dsharpe@nrcan.gc.ca

F.M. Kenny
Ontario Ministry of Natural Resources, 300 Water Street, Peterborough, ON, Canada, K9J 8M5; email: frank.kenny@lrc.gov.on.ca

GSC Contribution #2003096

Abstract

A basin analysis approach is used to help understand a complex aquifer system in the Oak Ridges Moraine and Greater
Toronto areas, southern Ontario, Canada. The aquifer complex consists of a sequence of discontinuous strata that have
a prominent regional unconformity. To help visualize this architecture, a stratigraphic database has been developed and
used to construct a 3-D stratigraphic model, through selective integration of disparate data. To accurately interpret
borehole logs, geological context was supplied by using expert knowledge constrained with a conceptual stratigraphic
framework. Utilizing a digital stratigraphic training framework derived from manually coded, high-quality data, an
expert system automatically interpreted and coded a large number of low-quality water well records. The expert system
was designed to emulate the manual borehole interpretation process by applying knowledge-based geological rules,
within the constraints of the digital training framework. Issues of poorly constrained interpolation due to sparse data
are addressed by the integration of additional spatial rules defined by thematic map coverages within the expert system.
As quantitative hydrogeological modelling moves to more regional scales, geological knowledge input becomes increas-
ingly more valuable. The availability of seamless geological mapping improves 3-D modelling and helps to limit the
effect of deficiencies in data coverage and data quality, often encountered in regional hydrogeological studies.

Résumé

Une approche d'analyse par bassin est utilisée pour aider à comprendre le système aquifère complexe dans la moraine
de Oak Ridges et les régions du Grand Toronto, dans le Sud de l'Ontario au Canada. Le complexe aquifère est consti-
tué d'une séquence de strates discontinues comportant une discordance régionale saillante. Pour faciliter la visualisa-
tion de cette architecture, une base de données stratigraphiques a été constituée et utilisée pour la mise au point d'un
modèle stratigraphique 3D, par une intégration sélective de données disparates. Afin d'interpréter correctement les don-
nées des rapports de forage, le contexte géologique a été défini à partir de connaissances expertes adaptées à un cadre
stratigraphique conceptuel. Utilisant un cadre stratigraphique numérique d'apprentissage défini à partir de données de
haute qualité codifiées manuellement, un système expert a codifié et interprété automatiquement une grande quantité de
rapports de forage de mauvaise qualité. Le système expert a été conçu pour émuler le processus d'interprétation

519
GIS FOR THE EARTH SCIENCES

manuelle par l'application de règles de base de connaissances géologiques, dans le cadre du système numérique d'ap-
prentissage. Le problème des interpolations mal assurées dues à un manque de données est traité en intégrant d'autres
règles spatiales définies par des couvertures cartographiques thématiques au sein du système expert. Lorsque la mod-
élisation hydrogéologique quantitative passe à des échelles plus régionales, l'apport des connaissances géologiques
devient de plus en plus précieux. La disponibilité de cartes géologiques sans couture améliore la modélisation 3D et aide
à limiter l'effet d'hiatus dans la couverture et la piètre qualité des données qui afflige souvent les études hydro-
géologiques régionales.

INTRODUCTION els. Widely applied in the petroleum industry, a basin analysis


approach was used to develop a better understanding of the geolog-
Background ical system (e.g., Miall, 2000). This approach progressed from data-
base development to geological and hydrogeological model devel-
In the Oak Ridges Moraine (ORM) and Greater Toronto areas opment to support quantitative assessment of a regional or basin-
(GTA) of southern Ontario, approximately 30 years of municipally scale flow system (Sharpe et al., 2002a).
funded, site-scale hydrogeological studies were completed in the
absence of a regional geological perspective. Consequently, these Approaches to Stratigraphic Modelling
local studies contributed little to defining the extent of aquifers,
controls on groundwater flow systems, or groundwater resource The approach to conceptual stratigraphic correlation in sedimentary
potential at the regional scale. The lack of regional, conceptual geo- basins has evolved in recent years (e.g., Miall, 2000). In glacial set-
logical knowledge in many hydrogeological studies has been high- tings, a traditional approach entails lithostratigraphic mapping
lighted as a serious shortcoming (e.g., LeGrand and Rosen, 1998). based on till correlation (e.g., Karrow, 1967). This approach
Integration of existing geological knowledge and increased use of assumes that glacial advance-retreat events produce laterally con-
conceptual approaches will help these studies cope with inherent tinuous and correlatable units, in 'layer-cake' fashion. Recently,
basin heterogeneities and vagaries within earth systems as it has in bounding surfaces and disconformities have been recognized as
the petroleum industry (e.g., Cherry, 1996). In addition, the increas- important features of the glacial stratigraphic framework (e.g.,
ing need for regional, watershed (e.g., Gerber and Howard, 2002) to Martini and Brookfield, 1995; Sharpe et al., 2004).
basin-scale (e.g., MacFarlane et al., 1994) hydrogeological studies
and the proportionate decrease in data support, requires increased Rendering conceptual geological knowledge into a digital
collaboration between geologists and hydrogeologists toward form suitable for quantitative groundwater flow modelling has com-
developing improved stratigraphic and hydrostratigraphic models. monly been achieved by either correlating cross-sections or interpo-
lating point data from a stratigraphic database. Where a stratigraph-
Geologists and hydrogeologists often have significantly different ic succession could be identified based on adequate data, regional
approaches to the creation of stratigraphic/hydrostratigraphic models correlations were commonly completed by matching physical prop-
due to fundamental differences in their objectives (e.g., Anderson, erties and lithologic sections. This approach provides a direct
1989). Geologists attempt to understand the geological system by method of simplifying the level of geological heterogeneity by
mapping and developing process-based conceptual models, such as allowing the geologist to interpret the data as stratigraphic correla-
sedimentary facies or event-stratigraphic models of a basin (e.g., tions are made (e.g., Alms et al., 1996; Thorleifson et al., 2002).
Walker, 1992). Geologists use expertise and intuition to incorporate Once the cross-sections are made, however, interpreted boreholes
qualitative information and sparse quantitative data to develop strati- may not be digitally captured and therefore unavailable for future
graphic context. Conversely, hydrogeologists strive to mathematically model iterations. Interpretations are deterministic in nature and
model flow systems without necessarily developing a broader geolog- because the model's integrity depends largely on the geologist's
ical knowledge base. For example, hydrogeological models are com- ability to apply a set of correlation rules throughout the model area,
monly simplified to continuous, layer-cake models to accommodate there is potential for inconsistent stratigraphic correlations from
data handling issues specific to numerical modelling software (e.g., section to section and across model iterations.
MODFLOW). Numerical flow modelling may be improved by incor-
porating a more realistic stratigraphic model. A regional, 3-D strati- An alternative approach to regional stratigraphic correlation is
graphic model that is a distillation of geological observations and to rely on the spatial interpolation of stratigraphic assignments
knowledge provides a stratigraphic framework and thus a sound basis stored in a database (Mossop and Shetson, 1994). This method
for subsequent hydrogeological model development. relies on a well-structured database that contains contact informa-
tion on all potential stratigraphic units, regardless of whether or not
A study by the Geological Survey of Canada (GSC) and the they have a zero thickness in portions of the study area (Hughes,
Ontario Geological Survey of the 11 000 km2 ORM area (Figure 1) 1993). The stratigraphic database is then queried to produce point
was initiated to provide a regional geological framework to better datasets that are interpolated as individual layers. Although the
plan and guide hydrogeological investigations. In the ORM area, stratigraphic assignments in this type of database are also determin-
extensive geological study has resulted in a broad geological istic in nature, once made and vetted, they can be combined with
knowledge base (e.g., Eyles, 1997). In recognizing this, a primary new data and re-interpolated without having to repeat the borehole
task of this study was to assess the validity of existing conceptual interpretation and stratigraphic coding process. This format also has
geological models. This analysis was carried out in the context of the advantage of being suitable for a variety of interpolation tech-
evolving glacial processes and improved landform-sediment mod- niques (e.g., Yarus and Chambers, 1994).

520
SPECIAL PUBLICATION 44

Figure 1. The location of the study area in southern Ontario (inset) and simplified geology of the ORM-GTA region (modified from Sharpe
et al., 1997).

Objectives of Quaternary sediment that is up to 200 m thick (Russell et al.,


1998a). Studies of Lake Ontario bluffs defined a formal stratigraph-
This paper documents the development of a digital stratigraphic ic framework for the intercalated glacial and non-glacial sediment
model of a glacial sedimentary basin. We show how expert knowl- (Figure 2a; Karrow, 1967). This succession is up to 100 m thick,
edge was used to aid the interpretation and integration of diverse consists mainly of interstratified glacial lake sand, silt, clay and
data sources using a stratigraphic database approach to GIS model- interbedded diamicton (Eyles et al., 1985) and is here referred to as
ling. This integration involved the use of a training framework of Lower sediment (Figure 2). Overlying this is the regionally exten-
higher quality data to help interpret and assign stratigraphic coding sive drumlinized Newmarket Till that was deposited beneath the lat-
to lesser quality data using an expert system. Issues pertaining to the est (late Wisconsinan) ice sheet (Karrow, 1967). The Newmarket
model integrity, testing and refinement are discussed by reference to Till is readily identifiable as a thick-bedded, dense, sandy diamic-
a single geological unit and hydrogeological applications of the ton with <15% gravel and high seismic velocities. These properties
model in the region are reviewed. can be correlated regionally and thus the till serves as a regional
marker (Figure 2a; Sharpe et al., 1997). Sandy interbeds <5 m thick
may also exist in this unit as discontinuous lenses (Boyce and Eyles,
GEOLOGICAL SETTING 2000). Newmarket Till commonly forms drumlinized uplands dis-
sected by tunnel channels (Brennand and Shaw, 1994). Tunnel
The ORM-GTA in southern Ontario is underlain by Paleozoic rock channels are up to 7 km wide, 40 km long and 200 m deep (Russell
that is exposed along the Niagara Escarpment and nearby river val- et al., 2003b). ORM sediment, up to 150 m thick, infills the chan-
leys (Figure 1). The bedrock surface gently dips to the southwest nels and overlies the Newmarket Till (Barnett et al., 1998). Halton
and is cut by valleys, such as the buried Laurentian Valley, that Till and glacilacustrine silt and sand incompletely overlie these
extends from Georgian Bay to Lake Ontario (Brennand et al., older ORM, Newmarket Till and Lower sediment units. Thin allu-
1998). Across most of the area, bedrock is covered by a succession vium also occurs along modern incised river valleys. This strati-

521
GIS FOR THE EARTH SCIENCES

graphic setting is the basis for the evolving conceptual geological groundwater flow studies and models in the ORM area (Howard et
model of the study area (Figure 2). al., 1997; Meriano and Eyles, 2003).

Stratigraphic Framework Development of the Regional Stratigraphic Model


Previous Conceptual Models Stratigraphic correlation within sedimentary basins now incorpo-
rates event stratigraphic analysis (e.g., Miall, 2000). Furthermore,
The earliest geological models for the area were based on strati- recent advances in glacial process understanding (e.g., Shaw, 1996)
graphic data from the thick sedimentary successions exposed along may more accurately guide the construction of stratigraphic models
Lake Ontario bluffs (Coleman, 1932; Karrow, 1967), physiograph- thus increasing the geological content in subsequent hydrogeologi-
ic mapping (Chapman and Putnam, 1984), and from geological cal analysis. Consequently, a reassessment of the stratigraphic
mapping south of the ORM (Watt, 1957; Karrow, 1967). Tills and framework of the ORM area was appropriate.
inter-till sediment that can be dated were used as the basis for defin-
ing a formal stratigraphy (Figure 2; Karrow, 1967, 1974) that Regional analysis by Shaw and Gilbert (1990) found evidence
emphasized lithostratigraphy and climatic cycles as the driving for a number of meltwater floods that sculpted the glacial landscape
mechanisms. The extensive tabular nature of well-exposed lake and of south-central Ontario. Subsequent detailed field mapping
river bluff strata (Karrow, 1967) encouraged the use of an early con- (Sharpe et al., 1997) and seismic reflection surveys (Pugin et al.,
tinuous 'layer-cake' model that was subsequently applied to region- 1999) identified significant new elements in the stratigraphic archi-
al correlations of geophysical log signatures for distances >40 km tecture of the ORM region (Sharpe et al., 1996). A regional, late
(Fligg and Rodriques, 1983; Eyles et al., 1985). Early landform sed- Wisconsinan unconformity was recognized that consists of drumlin-
iment models used drumlin field extents to correlate near-surface ized Newmarket Till, tunnel channels, and sculpted bedrock
units, inland from the lake bluff sections and to infer a glacial (Barnett et al., 1998). The most important stratigraphic feature of
advance/re-advance pattern. These early geological concepts are, in the unconformity was the identification of the tunnel channels and
part, the basis for the use of 'layer-cake' geological models in their continuation beneath the ORM (Pugin et al., 1999; Barnett et

Figure 2. The regional stratigraphic framework of the study area. a) Lithostratigraphy and chronostratigraphy (modified from Karrow,
1974, ages from Barnett, 1992. b) Conceptual stratigraphic architecture (modified from Sharpe et al., 1997).

522
SPECIAL PUBLICATION 44

al., 1998; Russell et al., 2003a). The tunnel channels commonly ic modelling consistent with a basin analysis approach. One of the
erode through Newmarket Till, truncate the underlying succession first objectives of the study was the development of such a database
at multiple levels, and locally are floored by bedrock. This revised accessible to all project members (Russell et al., 1996). A simple
architecture and resulting 3-D geometry may be defined as discon- data model was developed using Microsoft® Access© to allow new,
tinuous layered stratigraphy having incised channels that form a cored boreholes and field mapping sites to be integrated with
'jigsaw-puzzle' (e.g., Weber and van Geuns, 1990). The spatial vari- archival borehole records (Figure 4).
ability resulting from the tunnel channel system is hydrogeological-
ly significant because it provides hydraulic connectivity between Thematic GIS Training Data
near surface aquifers and deeper confined aquifers (Desbarats et al.,
2001). Being the most readily observable portion of the sedimentary basin,
the mapped surface extent of stratigraphic units provides the most
The stratigraphic model also incorporates sedimentological consistent and reliable component of the stratigraphic model. To
variability. Thus, coarse tunnel channel sediment overlying the constrain the stratigraphic model at ground surface, two thematic
unconformity is transitional upward into sand and silt of the ORM GIS map coverages were developed: i) surficial geology map poly-
(Russell et al., 2003a; Russell and Arnott, 2003; Sharpe et al., gons, and ii) glacilacustrine and alluvial polygons.
2003). Because similar sediment facies occur in lower sequences,
knowledge of stratigraphic architecture and lithofacies is required i) Surficial geology map polygons were developed from geo-
to avoid incorrect stratigraphic interpretations in subsurface data. In logical mapping completed for eight 1:50 000 scale National Topo-
addition, specific stratigraphic units cannot be accurately resolved graphic System (NTS) map sheets. This mapping combined with
due to the lack of deep boreholes and a dependence on uncertain reinterpreted geology from two existing geological maps provided
water well material descriptions. Hence, all sediment stratigraphi- seamless, digital map coverage, thus establishing consistency
cally older than Newmarket Till was grouped as one unit - Lower throughout the ORM study area (Sharpe et al., 1997). To support
sediment (Figure 2).

ORM GIS MODELLING

The development of the ORM stratigraphic


model has involved a detailed process of
data acquisition, validation and integration.
Data sources were systematically integrated
using an expert system and GIS techniques
to produce a series of conformable strati-
graphic surfaces (Logan et al., 2001). To
make effective use of disparate data sources,
a hierarchical approach to data integration
was adopted. Data regarded as high-quality
based on the rigour with which it was col-
lected, interpreted and reported were used to
help interpret or "train" data of lesser-quali-
ty. As will be discussed in the following sec-
tions, Ontario Ministry of the Environment
(MOE) water well records, due to drill core
recovery and reporting problems, are regard-
ed as relatively low quality. Accordingly,
data were divided into 2 groups based on
overall quality: training data and water well
records. Training data are composed of two
components: subsurface data and thematic
GIS map coverages. In this section, we out-
line these data sources as well as the data-
base analysis techniques and GIS processes
that were used to produce the ORM 3-D
stratigraphic model (Figure 3).

Geospatial Database

Data Model

A comprehensive geospatial database is the Figure 3. Flow chart representing the main data components and process used in the ORM
foundation of stratigraphic/hydrostratigraph- stratigraphic modelling method.

523
GIS FOR THE EARTH SCIENCES

Figure 4. The ORM stratigraphic data model for a) high-quality training data, and b) low-quality water well data.

this mapping effort, approximately 9000 shallow (<5 m) field The volume of glaciolacustrine and alluvial units can be easi-
observations were collected. Thicker sequences (up to 40 m) were ly determined where hydrologically significant. Upper surfaces of
recorded at river sections, lake bluffs and quarries. Vertices extract- these units are defined by their mapped extent and the DEM. Their
ed from DEM-controlled polygon perimeters were used to ensure volume is then the residual from subtracting the youngest modelled
that the 3-D model conforms to surface mapping (Logan et al., surface (i.e., Halton Till) from the topographic DEM, where these
2001). sediments are mapped. There is, therefore, no need to interpolate
the glaciolacustrine and alluvial surface directly.
ii) Glaciolacustrine and alluvial units consist of relatively thin
sediment (<5 m) and are stratigraphically younger than the units Subsurface Training Data
modelled in this study. These units were not explicitly modelled
because their thin, highly discontinuous nature makes them hydro- Located predominantly in an urban setting, this study was able to
geologically insignificant. For modelled units, stratigraphic infer- utilize over 4800 archival boreholes from OGS regional mapping,
ences in the subsurface are made based on stratigraphy mapped at geotechnical engineering reports from urban infrastructure develop-
the surface. Consequently, areas obscured by glaciolacustrine and ment (e.g., foundation studies) and hydrogeological site investiga-
alluvial units represent gaps in the surface mapping control. These tions (Russell et al., 1996). These data plus 6 new and 18 existing
thin units can cover large areas of older strata and, depending on continuously cored boreholes, 50 line-km of seismic reflection data
their location, often one or more modelled unit is absent beneath and 857 field mapping sites formed the subsurface component of
them. The potential disconformity may not be detected in water the training data.
well logs unless these situations are recognized from surface map-
ping. To ensure an appropriate stratigraphic assignment of underly- Water Well Records
ing borehole units, a shallow glaciolacustrine and alluvial unit
thickness maximum is imposed and an older "surface" stratigraphic Given the size and complexity of the area, substantially more sub-
code is assigned to individual glaciolacustrine polygons. The new surface data was needed than could be extracted from the training
code identifies the next youngest possible strata directly below the data to support adequate resolution of a digital model. Over 58,000
disconformity. provincial water well records were obtained and incorporated into

524
SPECIAL PUBLICATION 44

the modelling process. In lesser-developed areas, water well records faces. By integrating map geology polygons into the 3-D model
often provided the only source of subsurface information for sever- building process, more data control and geological intuition is
al kilometres. Also an invaluable source of hydrogeological data, applied to surface interpolations than would otherwise be supplied
the water well database was, however, used with caution due to by subsurface point data alone.
issues of data quality, reliability, and resolution (Russell et al.,
1998b). Wash boring, the most common method of water well Spatial Data Preparation
drilling, employs a fluid to transport material to the surface.
Although efficient, this method does not provide a continuous core Most of the archival borehole records were obtained either in hard-
for accurate logging. Hence, for reliability concerns related to copy or in a variety of poorly documented, flat digital formats. A
drilling method and location accuracy, it was decided that water substantial amount of data standardization and quality assurance
well records would be used in a secondary role in developing the was undertaken before this data was integrated and assembled into
ORM 3-D model. database tables.

Data Coverage Missing Stratigraphy


The data available to define the ORM area stratigraphy decreases
Due to stratigraphic unit discontinuities resulting from erosional or
greatly with depth. Many water wells intercept shallow aquifers
non-depositional episodes (e.g., tunnel channels), often one or more
and, as a result, they do not penetrate the complete stratigraphic
stratigraphic units are missing in borehole logs. To maintain strati-
thickness of the ORM and older units (Figure 5). For example, over
graphic consistency throughout all subsurface data and to fully cap-
40% of the water wells with a depth of >10 m penetrate less than
ture the geological interpretation, zero thickness units were inserted
1/3 of the stratigraphy. Because few archival geotechnical/hydro-
in the stratigraphic database to represent missing stratigraphy, there-
logical studies catalogued the entire stratigraphy in their study
by ensuring the complete stratigraphy was recorded at each bore-
areas, high-quality borehole data is also scarce at depth. For exam-
hole. The addition of zero thickness records at stratigraphically
ple, the overall data point density ranges from ~2.6 / km2 for the top
appropriate depths in the data tables ensures that the interpolated 3-
of the ORM unit, to ~1.6 /km2 for the Lower sediment/Newmarket
D model will accurately show zero thickness at proper locations and
Till contact (Figure 6).
elevations while honouring stratigraphic relationships.
The reduction of data support with depth is also evident with
geology map polygon coverage. Geology map polygons represent Georeferencing
the exposed portion of stratigraphic units at ground surface. A larg-
er percentage of younger units are exposed than older units in the Although the North American Datum (NAD) 83 projection datum
ORM study area. For example, both the Newmarket Till and the was introduced in the mid-1980s, as late as the mid-1990s some
older Lower sediment units are approximately equally widespread topographic data was only available in NAD 27. Since archival data
throughout the study area, however only ~1% of the Lower sedi- used in the ORM study spans the transitional datum period, much
ment unit is mapped at surface compared to 23% of the Newmarket of the data had to be transformed from NAD 27 to NAD 83 to match
Till. Because the geometry of unit surface exposures can be direct- the modelling projection. Georeferencing of much of the archival
ly determined by the topographic DEM, the surfaces of younger data was complicated by unreported datums, non-standard and
units have substantially more data support than those of older sur- unreported projection systems (e.g., Modified Transverse Merca-

Figure 5. Example N–S cross-section across the crest of the ORM showing poor data depth coverage and version 1 stratigraphic surfaces.

525
GIS FOR THE EARTH SCIENCES

Figure 6. Example 20 km2 area showing data coverage for a) ORM surface, and b) Lower sediment surface.

tor), or a lack of a georeferenced location map. Due to the age of MOE UTM coordinates with 1:10 000 scale Ontario Basic Mapping
most reports (i.e., pre 1980s), borehole coordinate datums were (OBM) cadastral polygons coupled with a comparison of MOE ele-
assumed to be NAD 27 unless otherwise indicated. Archival reports vations and the ORM DEM were preformed using spatial queries in
that relied completely on georeferenced location maps of the study a GIS (Kenny et al., 1997). This process indicated that approximate-
site to show borehole locations were scanned and geocoded in a ly 30% of the water well dataset were potentially mislocated.
GIS. In other cases, coordinates were established by visual estima- However, due to the tendency of water wells to be clustered, a rea-
tion or converted from a local coordinate system to UTM projec- sonably high level of redundancy exists in the dataset. Because of
tion. Reported coordinates were visually verified by plotting on a this, a decision to omit well records with these potential location
digital topographic base. errors was made.

Despite the fact that water wells were captured in a digital for- For consistency in the stratigraphic model building process, a
mat with UTM coordinates, there often existed problems with loca- hydrologically conditioned DEM (Kenny et al., 1999) was used as
tion accuracy (Kenny et al., 1997). Ministry of the Environment an elevation datum (i.e., elevations were extracted from the topo-
(MOE) water well reports require only a hand-drawn diagram and graphic DEM and applied to all data components). In addition to
lot and concession numbers to indicate well location. Using this eliminating any possible elevation errors present in new or archival
information, coordinates and elevations are later determined and data, this level of standardization allowed for the seamless integra-
appended to a database by the MOE. Unfortunately, lot and conces- tion of geological map polygons whose elevations were also
sion identifiers are not unique within the Ontario municipal cadas- extracted from the DEM.
tral fabric and water well location diagrams often lack the informa-
tion necessary to accurately locate the well. In addition, data entry De-clustering
mistakes have caused obvious location errors ranging from 10s to
100s of kilometres. To identify location errors, a systematic location Training data is generally too sparse to warrant systematic de-clus-
checking protocol was implemented. A comparison of recorded tering. However, some clustered borehole logs from intensive site

526
SPECIAL PUBLICATION 44

studies were considered to be redundant at a regional scale and development efforts, Viewlog© borehole visualization and data
hence not digitally captured. Water well data, on the other hand, are management software was also used.
plentiful but not evenly distributed throughout the study area. They
tend to be clustered around rural communities, along roads and In the water well dataset, statistics based on database and GIS
around water bodies. Water well data redundancy and clustering queries were used to detect unreasonable unit configurations and
were reduced with the aid of an arbitrary 0.5 km2 grid. Within each sediment descriptions. For example, a comparison of sedimentolog-
grid cell, well records were ranked according to favourable attrib- ical logging of borehole core and archival data indicated that the
utes (i.e., depth, number of intervals, bedrock recorded). By elimi- thickness of fine-grained units, particularly clay was grossly over-
nating shallow and redundant wells, the total number of location- estimated (Figure 7). The reported unit thickness more commonly
verified water well records was reduced to nearly half while main- represents a stratigraphic unit thickness rather than that of a sedi-
taining a similar level of data coverage (Logan et al., 2002). mentological unit. This is particularly true for water well records in
which clay intervals can measure up to 30 m thick compared to sev-
Re-drilled Water Wells eral centimetres in sedimentologically logged core (Russell et al.,
1998b).
An additional step was taken to recover water well information that
would otherwise be lost as a result of de-clustering. In an effort to Standard Material Coding
improve water quality or capacity, some homeowners will have an
existing well extended to greater depth. In these cases, the driller Material descriptions were standardized to allow the disparate litho-
begins recording material descriptions at the bottom of the existing logic coding systems and free-form descriptions used in the archival
well and a new report is submitted to the MOE. The pre-existing sources to be integrated with new data. The water well records illus-
well depth is described as "previously dug/drilled" in the new trate the scope of this problem with descriptions assigned to three
report. However, in the MOE water well archival process, reports fields from a choice of 81 possible descriptors. This permitted a
for new well extensions were not combined or associated with the possible number of descriptive permutations exceeding 500,000. In
original report. Because the de-clustering process, as described the ORM dataset there were 1900 unique descriptions. This prob-
above, preferentially selected deeper wells in clustered groups, the lem was solved based on knowledge from geological mapping and
deeper well report would tend to be chosen over the shallower orig- a literature review of glacial environments. A standard coding
inal report. The "previously dug/drilled" interval of the selected scheme of eight principal sediment descriptors was used to recode
well record provides no information, how-
ever, so a process to identify and combine
these associated well records was devel-
oped. A spatial query was used to identify
and group wells that were within 500 m of
extended wells. If the total depth and "previ-
ously dug/drilled" interval of two wells in
local groups were within ± 2 m, the two
records were merged to provide a complete
stratigraphic description. This process im-
proved over 800 water well records.

Data Vetting

Visual inspection of computer-generated


borehole cross sections were the primary
means of inspecting training data, however
due to their large numbers, water well data
was assessed via comprehensive database
and GIS queries.

As a way of checking coded units and


stratigraphic assignments in the training data,
several iterations of borehole cross-section
inspections were completed during the mod-
elling process. In addition to detecting cod-
ing errors, these data visualizations served as
the primary means of testing and refining the
conceptual model. Initially these cross-sec-
tions were produced with unreleased utilities
written in AutoLISP® and MapBASIC®,
and later with BoreHole Mapper© MapInfo© Figure 7. A comparison of archival data and sedimentological data. Note the contrast in
add-on. To help streamline continuing model reported unit thickness, sediment classification and sedimentary structures.

527
GIS FOR THE EARTH SCIENCES

sediment descriptions from all data sources (Russell et al., 1998b). et al., 2001). This comparison indicates that archival data over
In some cases, multi-word descriptors were recoded to more mean- reports fine-grained sediment, often by an order of magnitude or
ingful geological terms (e.g., clay-gravel-silt recoded to diamicton). greater (Figure 9). For example, by comparing water well records
For the water well records, this material description reclassification with borehole core data, clay is overestimated by 1500% in the
was completed through an automated database operation, whereas western ORM likely at the expense of fine sand.
for training data it was applied interactively by geologists.

The integrity of the automated water well reclassification was


checked against other elements of the water well records (i.e., inter-
nal checks) and against manually coded training datasets (i.e., exter-
nal checks; Russell et al., 1998b). An example of an internal check
was a comparison of the principal material codes against screened
interval. Assuming well screens are placed at water bearing inter-
vals, permeable materials (e.g., sand or gravel) are expected within
screened intervals. This check found that 84% of the screens corre-
sponded to sand and gravel thus confirming the validity of the cod-
ing protocol. Two external tests completed on the recoded water
well records illustrate some of the problems with reliability, accura-
cy and material unit resolution. A comparison of water well record
units at a depth of one metre and the geology map indicate large
variations in the material distribution by geological map unit
(Figure 8). This is most likely due to poor descriptive accuracy in
well reports. However, part of this variation may reflect internal
map unit heterogeneity. A further test completed in the western
ORM compared material descriptions of all archival data with
recoded sedimentological descriptions of continuous core (Russell

Figure 9. a) A comparison of Oak Ridges Moraine composition


based on archival datasets and GSC data. Note radically different
reporting of fine-grained sediment, clay and silt between datasets.
b) Moraine composition using sediment facies coding of continuous
core and outcrop sections. Most clay in archival data (a) is most
probably sediment facies Sg and Sr. Agency abbreviations: MTO -
Ministry Transport Ontario (geotechnical); RMP - Regional
Municipality of Peel; RMY - Regional Municipality York; GSC-Bh -
Figure 8. External comparison of recoded water well material des- Geological Survey of Canada (sedimentology); IWA - Interim Waste
criptions at 1 m depth with geological map polygons (modified from Authority (hydrogeology); MOE - Ministry of Environment water
Russell et al., 1998a). Note a much better correlation for ORM wells. Material abbreviations: Cl - clay; Di - diamicton; Gr - grav-
aquifer material than for Newmarket Till aquitard. The grey band el; Sa - dune scale cross-stratified sand; Si - silt; Sg - graded fine
indicates material that best matches the surficial geology map unit. sand - silt; Sr - ripple scale cross-laminated sand (from Russell et
Note the high percentage of reported clay. al., 2001).

528
SPECIAL PUBLICATION 44

The water well material code reclassification system has been ically and coded with the same standard coding system used for
adopted by MOE, and subsequently, feedback from other groups training data. Due to the number of records, however, an automated
working with water well records has been received and assessed. approach was taken. In an attempt to mirror the process involved in
For example, a possible scenario for misidentifying unusually large manually interpreting training borehole stratigraphy, a systematic
boulders led to a revision of the reclassification rules (T. Warman, approach that consistently applied a set of geological rules within a
personal communication, 2002). To avoid abnormally thick grav- stratigraphic training framework was employed for interpreting
el/boulder units and excessively deep bedrock, thick intervals (>3 water well data (Figure 10). Using Microsoft® Access© and Visual
m) of MOE-coded bedrock followed by a thin (<1 m) non-bedrock BASIC for Applications©, an expert system was developed to
interval are now interpreted as actual bedrock and not as boulder or assign stratigraphic codes to water well records (Logan et al., 2002).
gravel. Although not intended nor designed for use outside of the
ORM study area, to date the knowledge-based rules seem to be The stratigraphic coding expert system relied on multiple
robust in Quaternary sediment elsewhere in Ontario. inputs: a training framework interpolated from high-quality, subsur-
face training data (Figure 10); thematic GIS training data; and stan-
Stratigraphic Interpretation and Coding dardized material coding applied to water well records. Expert geo-
logical knowledge controlled the model through these inputs and
In the context of a sound conceptual stratigraphic framework, mate- also by rules programmed into the expert system. The principal set
rial descriptions in high-quality training data were interpreted strati- of expert rules established the sediment descriptions allowable for
graphically by geologists. Thickness and elevation ranges defined by each stratigraphic unit (Table 1). The remainder handled specific
geological context as well as physical properties help geologists men- situations and refined coding parameters or program logic accord-
tally build a structural model to guide stratigraphic interpretations. ingly. Coding parameters include allowable sediment descriptions,
allowable stratigraphic assignment, maximum elevation, maximum
Prior to their inclusion in the stratigraphic model-building unit and interbed thickness. Table 2 lists some examples of specific
process, the water well dataset needed to be interpreted stratigraph- expert rules.

Figure 10. A schematic representation of the development of data-driven stratigraphic model with geological control. The stratigraphic
training framework interpolated from subsurface training data coupled with DEM-controlled geology polygons provide stratigraphic con-
text to help interpret water well material intervals.

529
GIS FOR THE EARTH SCIENCES

Training Framework Creation Triangular Irregular Network (TIN) interpolation (Figure 10). The
preliminary training surfaces were then used to identify stratigraph-
The perimeter of each stratigraphic unit at ground surface (Figure 1) ic intervals in boreholes that partially penetrate younger units and
was defined in a GIS using a polyline-to-point function to convert that are also deep enough to warrant deflecting the older surface
geology map polygon contacts into a set of points at 50 m spacing. downward. Once these 'push-down' points are identified, they are
The DEM elevation was then extracted and appended to each of added to the appropriate stratigraphic datasets and surfaces are re-
these points. The map polygon points were combined with strati- interpolated using the Natural Neighbour method. Following this,
graphic contacts from high-quality boreholes to assemble strati- surfaces were forced to conform to each other, the surface geology
graphic training datasets for each of the 5 stratigraphic units. A pre- and the topographic DEM via a series of GIS overlay operations.
liminary training surface was then produced for each unit using For exposed stratigraphic units, geology map polygons were used to
'cookie-cut' regions of the topographic DEM to overlay on appropri-
Table 1. Allowable sediment descriptions for expert system strati- ate surfaces. Having included the map polygon perimeter points in
graphic coding the surface interpolations, a seamless transition from subsurface
data to surface geology is attained. Overlaying and propagating
Allowable minimum grid cell elevations from the uppermost surface down-
Stratigraphic Code / Order Sediment Description ward then resolved any residual surface overlaps.

7 - Surficial miscellaneous (any non-bedrock material) A distance buffer grid, representing the minimum distance to a
training surface data node, was generated around the data subsets
6 - Glacilacustrine (GL) clay used to interpolate each surface. These raster surfaces were queried
silt at each water well location and extracted values were appended to
sand a database table. Using geospatial queries, additional information
was appended to this table such as: geology map polygon strati-
5 - Halton Till (HT) clay graphic code, maximum surface elevation and unit thickness rules.
silt This table formed the primary training data input for the expert sys-
clay-silt diamicton tem. It was used in conjunction with water well material intervals
sand-silt diamicton and programmatic expert knowledge-based rules (Tables 1 and 2) to
silt-sand diamicton interpret the stratigraphy of each well record.

4 - Oak Ridges Moraine (ORM) clay Stratigraphic Interpretation of Water Well


silt Material Code
sand
gravel
The expert system first examined water well material descriptions
and assigned a preliminary stratigraphic code based on allowable
3 - Newmarket Till (NT) sand-silt diamicton
sediment types as defined by expert rules. For each water well, the
silt-sand diamicton
stratigraphic unit mapped at ground surface was used as a bench-
mark to define the youngest possible stratigraphic assignment.
2 - Lower sediment (LS) (any non-bedrock material)
Accordingly, the first material interval in the well record is assigned
the stratigraphic code of the map polygon in which it is located. In
1 - Bedrock (B/R) (any bedrock)
unlithified/undeformed Quaternary sediment, as is the case in this

Table 2. Examples of expert rules used for water well stratigraphic coding. See Table 1 for stratigraphic unit abbreviation

Rule Condition Effect Purpose

Previous strat. code = GL AND Combine current depth interval with To disregard thin coarse-grained intervals
current material = sand/gravel AND next interval and use next material below thin GL units that may otherwise be
interval thickness < 3 m code for new combined interval. coded thin ORM.

Interval thickness <1.5 m AND Flag the next interval as a "map To ensure that no overly thin, near-surface
interval is a "map unit" (i.e., it exists unit" also. intervals are treated as "map unit" and receive
at a depth of 1 m) the strat. code from the geological map.

ORM strat. code is applied to any Increase allowable NT interbed To lessen the possibility of interpreting the
interval of the water well thickness from 5 to 8 m. presence of LS from NT at a high elevation
beneath the ORM.

Maximum LS elevation < well LS code is not permitted regardless of To eliminate unrealistically high LS elevation.
bottom elevation the effect of any other rule/constraint.

530
SPECIAL PUBLICATION 44

study, it can be assumed that older sediment will never overlie table and using rules based on geological knowledge, the expert
younger sediment. The expert system was, therefore, designed to system was able to interpret the well record's material intervals and
interpret either the same code as the previous interval or an older assign a plausible stratigraphy within a well-constrained strati-
stratigraphic code for deeper intervals in the well record based on graphic training framework.
material descriptions (Table 1). For example, all material intervals
for the entire depth of a well located within a polygon mapped as Model Generation
Lower sediment can only be interpreted by the system as either
Lower sediment or the next oldest, and last modelled unit – The process of building a 3-D basin model is greatly simplified by
bedrock. Initially, the transition from the current stratigraphy to the building a stratigraphically rigorous database. New data is easily
next older one is primarily determined by the allowable sediment integrated for subsequent modelling iterations because stratigraph-
descriptors (Table 1). ic interpretations or "picks" are stored and readily available. Basic
queries were used to extract a stratigraphic dataset from the data-
Training Surface Correction base for each of the 5 main units that were modelled. Using a sim-
ilar series of steps as described for the production of the training
Due to stratigraphic unit heterogeneity and potential reporting surfaces, a final set of surfaces were generated from X, Y, Z point
inconsistencies, the presence or absence of allowable sediment sets assembled from high-quality boreholes, low-quality water
types alone is not a definitive basis for stratigraphic assignment. A wells and thematic map polygon points (Logan et al., 2002). In a 3-
spatial stratigraphic context is essential for making reliable water D GIS or quantitative modelling application, these surfaces can be
well interpretations. In this system, spatial context is verified by rendered and used to build 3-D volume models (e.g., voxets or
comparing the preliminary stratigraphic assignment to training sur- SGrids in gOcad©; Figure 11) or individual stratigraphic unit
face elevations (Figure 10). A linear decay function based on dis- isopach grids in support of hydrogeological modelling and flow
tance to nearest training point determined the extent to which train- analysis.
ing surfaces governed the stratigraphic code assignment. Because
training surfaces only affected wells within a distance of 1-2 km Iterative Model Refinement
from a training surface data node, global expert knowledge rules
were relied on to constrain stratigraphic coding beyond the influ- A model is a representation of reality whose accuracy is dependent
ence of training data. The range of influence of the Halton Till and on the quality of observed data and the validity of assumptions that
ORM training surface nodes were set at 1 km, while a 2 km range are used to interpret it. For a conceptual model to be considered
was used for the others. By extracting information from the input geologically defendable, re-evaluation based on new data or insight

Figure 11. An example of stratigraphic surfaces converted to an array of regular, 3-D cells (i.e., voxet model) in gOcad©. Only Newmarket
Till and ORM voxet regions are shown.

531
GIS FOR THE EARTH SCIENCES

must be undertaken until the model can account for all observed directly measurable (e.g., Kassenar et al., 2003). Furthermore,
phenomena. For this reason, it was recognized that the ORM 3-D screen depths are presumably reported accurately due to the cost
model would evolve as versions were reviewed and new data accountability involved in casing and screening a well. Since 84%
became available. The stratigraphic database approach was chosen of reported screen intervals correspond to sand or gravel descrip-
to streamline the production of multiple model iterations because of tions (Russell et al., 1998a), we can assume that screens are set only
its ability to integrate new data without the need to revisit existing in aquifer material. Under this assumption, screens can be used as a
interpreted data. Each version of the model also provides an oppor- secondary check to refine the upper and lower contacts of the
tunity for checking borehole interpretation and code assignment. By Newmarket Till aquitard. This additional protocol was implement-
reviewing model cross-sections, rendered 3-D visualizations and ed in version 2. It affected the position of contacts in ~7% of water
sediment thickness maps, anomalies and unexpected trends served wells with units that overlap the version 1 Newmarket Till unit
as targets for closer scrutiny. In some cases, this prompted data re- (Logan et al., 2005).
interpretation or acquisition of new data. An iterative system of data
interpretation, modelling and data improvement was devised and a Tunnel Channels
version number for released models was established (Logan et al.,
2002). Tunnel channels are hydrogeologically significant features that have
complex geometries, undulating longitudinal profiles and variable
Version 1 of the model (Logan et al., 2002) is the culmination sediment fills (Sharpe et al., 2003). Channels generally have a
of numerous preliminary iterations. Subsequent improvements have north–south trend and are best defined north of the ORM where they
been incorporated in a version 2 based on the acquisition of new are not totally obscured by sediment in-filling (Figure 1). Because
data, and feedback from project geologists and collaborators in the they can act as areas of aquifer leakance, the potential impact of tun-
area (Logan et al., 2005). nel channels on regional hydrogeological flow systems can be sub-
stantial (Desbarats et al., 2001). To make the digital model conform
New Data more closely to the conceptual geological understanding, mapped
areas of tunnel channel features were used to further constrain
A deficiency of the version 1 stratigraphic model is the lack of sub- Newmarket Till distribution and thickness (Russell et al., 2003b).
surface data coverage in areas of thick sediment east of Rice Lake Tunnel channel features were assigned to one of four classes accord-
and at depth beneath the ORM unit. This reflects a scarcity of ing to similarities in geometry, orientation, and the interpreted extent
archival borehole records due to a lack of development or the exis- of underlying Newmarket Till (Russell et al., 2003b). For the version
tence of serviceable shallow aquifers. Shallow aquifer wells locat- 2 model, tunnel channel polygons were added as a new component
ed on the moraine provide very little information regarding the of thematic GIS training data and this restricted the thickness of
geometry of deeper units (Figure 5). Efforts were made throughout Newmarket Till allowed in tunnel channel areas.
the model development, however, to locate and integrate any new
or previously unavailable borehole records. Subsequent to version 1 MODELLING RESULTS
of the model, close to 5000 new water well records were added and
the high-quality training dataset was augmented with 178 new geot- Stratigraphic modelling in the ORM area has resulted in a digital 3-
echnical borehole logs. Unfortunately, very little new archival data D model of the approximately 11,000 km2 study area (Logan et al.,
were found for the data poor areas outlined above, however approx- 2002). In addition to the release of the digital surfaces, each of the
imately 180 new water wells drilled to bedrock were obtained in four major stratigraphic units above bedrock (Figure 2) have been
areas of thick ORM sediment. documented with a structural surface and isopach map: Halton Till
(Russell et al., 2005a), ORM (Russell et al., 2005b), Newmarket
Expert System Improvements Till (Sharpe et al., 2005a) and Lower Sediment (Sharpe et al.,
2005b). An assessment of the version 1 Newmarket Till isopach is
Over time, the expert system has been revised with new or modified highlighted below.
knowledge-based rules. Based on model inspections and testing,
various parameters were adjusted depending on local stratigraphic Newmarket Till Isopach Map
configurations such as maximum elevations, maximum thickness
and allowable interbed thickness. Many rules were refined based on The interpolation of Newmarket Till indicates that it covers ~80%
the iterative generation of model results and testing with field of the area and ranges in thickness up to 100 m (Figure 12; Sharpe
observations or areas of abundant training data. The repeatable et al., 2005b). Many patterns visible on this map may be explained
nature of the methodol used in the ORM study made it possible to geologically (Sharpe et al., 2002b); however, others appear to be
establish an efficient, iterative cycle of rule assessment and revi- related to variable data quality and coverage or to interpolation
sion. This expert feedback and program improvement is an impor- problems. The following is a selection of observations and their
tant component of an expert system allowing it to gain 'knowledge'. possible implications:

Subsurface Hydrological Conditioning i) Of the modelled Newmarket Till thickness, ~3 % is >50 m


(Figure 12). Such thicknesses are not present in training data
MOE water well records contain construction information that can or water well records and have not been observed in the field.
be used to further verify material descriptions and consequently Analysis of Newmarket Till and Lower sediment structural
stratigraphic assignments. For example, in contrast to estimates of surfaces indicates that the anomalous thickness is due to poor
the depth of material unit contacts, the depth of the well screen is data support (e.g., east of Rice Lake).

532
SPECIAL PUBLICATION 44

Figure 12. An isopach map and thickness histogram of Newmarket Till from version 1 stratigraphic model (Logan et al., 2002). The colour
ramp represents a range of thickness from high (red) to low (blue). White areas indicate no Newmarket Till. The histogram shows frequen-
cy of interpolated thickness. Section lines at AA', BB' and CC' are shown in Figure 13.

ii) Little or no Newmarket Till occurs in the southwestern part of v) Large valleys north of the moraine (i.e., Holland Marsh) com-
the region (Figures 12 and 13). This area slopes up to the base monly contain thin interpolative till wedges (Figures 12 and
of the Niagara Escarpment and has a thin sediment cover atop 14a). Here Newmarket Till was interpolated across parts of the
an elevated bedrock platform. Training data indicates that valley as a result of honouring sparse borehole data (Figure
Newmarket Till is sparse and truncated in the area. This is con- 14a). Alternatively, clay reported in some water well records
sistent with the meltwater erosion model for formation of the might have been erroneously coded as Newmarket Till. It is
regional late Wisconsinan unconformity (Russell et al., 2005b). not possible to assign the term 'clay' to a unique stratigraphic
unit since it can potentially occur in all modelled units.
iii) Newmarket Till does not occur in areas along the south side of Moreover, this overused description could actually indicate a
the ORM between Rice Lake and Lake Scugog (Figure 12). range of materials from till to fine-grained sediment (Russell et
Comparison with the surficial geology map indicates that this al., 2001). Expert knowledge and/or additional high-quality
is consistent with the mapped distribution of Lower sediment data are needed to resolve this ambiguity or data 'noise' (Figure
in this area. Hence Newmarket Till was either not deposited or 7).
it had been eroded.
vi) In a 20 km wide belt along the Lake Ontario shoreline, narrow
iv) Mottled patterns consisting of till thickness > 10 m surround- linear patterns trend toward the lake (Figure 12). These com-
ed by thinner areas of till (Figure 12) are probably related to monly correlate with modern river valleys formed by
both geological factors and interpolative artifacts. In areas of Holocene alluvial incision that dissected the Newmarket Till.
mapped Newmarket Till these mottled patterns likely represent Field evidence indicates that these valleys are commonly
geological features (e.g., drumlins) as opposed to data artifacts floored by Lower sediment that also occurs locally in valley
(Sharpe et al., 2002b). Elsewhere, in areas buried by ORM sides.
sediment, the pattern appears to be related to erosional win-
dows and the interpolative transition from areas of zero vii) Landscape elements <1 km are not clearly resolved where the
Newmarket Till thickness. interpolation is reliant on water well data (Figure 14). High-

533
GIS FOR THE EARTH SCIENCES

Figure 13. Cross-sections (AA', BB' and CC') drawn across the ORM version 1 structural model. East of Holland Marsh tunnel channel, the
AA' section illustrates that the DEM-controlled elevation of tunnel channel bases is low enough to have breached the Newmarket Till sheet.
In the southwest, section BB' shows thin sediment on sloping bedrock. Section CC' transects a Newmarket Till upland. The location of sec-
tion lines is shown in Figure 12.

quality, high-resolution data such as seismic profiles are to have formed drumlins (Shaw and Sharpe, 1987). Drumlins pro-
required to confirm the presence of landscape architecture duced during this phase of flow represent streamlined, erosional
such as drumlins and small tunnel channels (Figures 15c and bedforms. Differential erosion and flow evolution lead to shallow
d; Pugin et al., 1999). tunnel channels that are intertwined amongst the drumlins. This
landscape is identified on the isopach map as a mottled pattern rep-
Geological Interpretation of Newmarket Till resenting drumlinized uplands (Figures 12 and 14). Continued wan-
ing of the flow concentrated meltwater into deeper tunnel channels
Patterns in the Newmarket Till isopach and structural surface have that commonly eroded through Newmarket Till (Pugin et al., 1999;
a range of shapes, thickness, dimensions, and orientations that can Russell et al., 2003a). These large tunnel channels are up to several
be interpreted geologically. The Newmarket Till structural surface km wide and 10s of km long and define the extent of inter-channel
defines the late Wisconsinan unconformity (Figure 2). This uplands up to 5-10 km across and 10-20 km long. Field evidence
unconformity consists of a number of architectural elements that indicates that deep tunnel channels contain no Newmarket Till,
can be recognized in the isopach map (Figures 12 and 14). however as mentioned, sparse and/or 'noisy' data may cause inter-
Ordered by decreasing scale, they are: i) inter-channel uplands, ii) polative artifacts that result in outliers of Newmarket Till within the
tunnel channels, iii) portions of Holocene river valleys and iv) tunnel channel margins. On the isopach map, the tunnel channels
drumlins. are recognized by linear trends of thin (i.e., <10 m) and absent
Newmarket Till. Most inter-channel uplands are characterized by
The regional event-stratigraphic model interprets the uncon- thick till (i.e., 10-50 m) that is higher in elevation than the interven-
formity as the product of a regional subglacial meltwater discharge ing channels. South of the moraine, tunnel channel forms are wider
(Shaw and Gilbert, 1990). This event produced different types of and shallower or have thicker till substrate (>10 m). Modern
erosional landforms as the flow evolved and waned. Initial region- streams draining to Lake Ontario have deeply incised narrow val-
al flow occurred as a broad, shallow, high-energy sheet interpreted leys through the Newmarket Till (Sharpe et al., 1997).

534
SPECIAL PUBLICATION 44

Figure 14. A comparison of identical example areas showing a) Newmarket Till thickness and b) the DEM for the Holland Marsh area. Note
the correlation of thin Newmarket Till with identified tunnel channels. Locally within the uplands, spindle-shaped features on the isopach
map (a) correspond with drumlin forms in (b).

535
536
GIS FOR THE EARTH SCIENCES

Figure 15. The stratigraphic architecture of Newmarket Till has been assessed from a) outcrop, b) borehole sediment and geophysical logging, c) reflection seismic profiling
and d) stratigraphic interpretation (from Sharpe et al., 2005b).
SPECIAL PUBLICATION 44

DISCUSSION tant control on modelling an aquifer complex (Fogg, 1986).


The model provides a geological framework for understanding
To date, the stratigraphic model has been applied to problems at the probable distribution and scale of such windows. This may
both the regional and site scale. Additionally, the modelling exer- lead to more realistic flow modelling than attempting to satis-
cise in this area shares many similarities with other glaciated areas fy numerical modelling requirements by arbitrarily inserting
and the prevailing theme of using water well records in hydrogeo- hydraulic connection (e.g., Bester et al., 2002). Cross-sections
logical studies (Berg and Thorleifson, 2001; Thorleifson and Berg, along the deep tunnel channel fills and along inter-channel
2002). Consequently, work in this study potentially has applications uplands illustrate the radically different flow system geometry
to other areas. First, four example applications of the stratigraphic (Figure 13).
model are shown, and second, the wider application of GIS strati-
graphic methods developed in this area are discussed. Geological Model Assessment

Application of the ORM Stratigraphic Model The validity of a geological model depends on the quality, quantity
and distribution of data and the expertise of the geologist that inter-
The ORM stratigraphic model resulted in a series of five structural prets it. The development of a conceptual model early in a study is
stratigraphic surfaces and four derived isopach maps for the essentially a hypothesis (e.g., LeGrand and Rosen, 2000; Flint et al.,
Quaternary succession (Figure 2). These surfaces can provide a 2001). As such, it should be tested through data collection, experi-
stratigraphic or hydrostratigraphic framework for future hydrogeo- mentation (e.g., numeric modelling) and peer review. In the Oak
logical analyses, such as: Ridges area, the use of recent glacial process models and landform
interpretations led to a new event-stratigraphy and the interpretation
i) Geostatistical water table mapping – The ORM stratigraphic of a late Wisconsinan regional unconformity (Figure 2; Barnett et
model permits the generation of piezometric surfaces for indi- al., 1998; Sharpe et al., 2004). Supporting this conceptual model,
vidual hydrostratigraphic units and provides for the first time elements of the unconformity (e.g., tunnel channels) were con-
the opportunity for a regional assessment of water levels. For firmed by field mapping (Sharpe et al., 1997), seismic profiles
example, over 1500 water wells screened in the ORM unit (Pugin et al., 1999) and continuous core (Russell et al., 2003b).
were modelled for the Uxbridge area. This data was integrated Furthermore, because the conceptual model forms the framework
with stream discharge training data (Hinton, 1995) and topo- for interpreting archival subsurface data, the integrity of the ORM
graphic data (Kenny et al., 1999). All data were interpolated stratigraphic model is inherently linked to the integrity of the ORM
using kriging with external drift where the topographic data conceptual model. For structural stratigraphic models, as was done
was used as the external drift component (Desbarats et al., in this study, visual inspection can be used to determine if they con-
2002). form to observable features. Alternatively, they can be tested by
comparison with independent data sources (e.g., Kassenar et al.,
ii) Modelling regional aquitard thickness and leakance – 2003). Quantitative modelling results can be compared with field
Aquitard leakance is a measure of aquitard integrity as a flow observations to determine if conceptual model-based assumptions
barrier (Desbarats et al., 2001). This is significant for wellhead are valid (e.g., Herzog et al., 2002). Collection and use of high-qual-
protection in areas where a regional aquitard has been locally ity data allows testing and revision of the conceptual model that will
eroded by tunnel channels forming hydraulic windows (Figure lead to improvements in geological understanding (e.g., Sharpe et
2). A regional aquitard was modelled as a hydrostratigraphic al., 2002a).
unit that consisted of Newmarket Till and/or adjacent underly-
ing clay units of Lower sediment. A Sequential Simulator Regional Map Data Coverage
Simulation used ~12,000 sites to generate 50 realizations and
a probability map of aquitard thickness (Desbarats et al., At the regional scale, the scarcity of existing high-quality data and the
2001). Leakance was then mapped by upscaling the thickness high cost of acquiring new subsurface data makes it necessary to fully
and conductivity data. exploit thematic geology maps. This map data is commonly under
utilized in modelling exercises. In the hands of experienced users,
iii) Aquifer assessment – An isopach map of ORM sediment however, these datasets may yield a host of embedded information
derived from the version 1, 3-D model (Logan et al., 2002) pertaining to stratigraphic relationships and formative processes. Full
presents the first data-derived visualization of this high-use use of map data provides access to regional coverages that are com-
aquifer complex (Turner, 1977). Aquifer thickness variation is monly accompanied by extensive supporting point data.
controlled by the irregular topography of the underlying
regional unconformity and moraine elevation. The thickness In the ORM study, surficial geology maps (Sharpe et al., 1997)
distribution is highly directional and relates to the filling of provided a key control for the stratigraphic architectural model
deeply scoured ~N–S channels that were overlain with E–W (Figure 2). Additionally, following terrain analysis, a derivative tun-
oriented depositional wedges visible on the topographic DEM. nel channel map (Russell et al., 2003a) was subsequently used as a
thematic dataset for version 2 of the stratigraphic model. These the-
iv) Flow system analysis – The ORM stratigraphic model, with matic map data were input to the model as elevation-controlled,
implicitly defined hydraulic connectivity 'windows', provides a stratigraphically coded point data. As an alternate approach, length
3-D grid and conceptual framework for subsequent numeric and width scales can be extracted from map data and used as a con-
flow modelling in the area (e.g., Kassenaar et al., 2003). The ditional dataset in geostatistical operations (e.g., Weissmann et al.,
ability to recognize vertical hydraulic connectivity is an impor- 1999).

537
GIS FOR THE EARTH SCIENCES

Integration of map data greatly increased the spatial control for and a GIS. The construction of a geospatial database to support
the training surfaces beyond the initial borehole data density of <1 accurate stratigraphic modelling depends, however, on the avail-
site per km2. The geological architecture of glacial deposits is rela- ability of a complete and reliable data coverage. Due to the scale of
tively flat lying compared to many folded and faulted bedrock regional studies, adequate data coverage is often difficult to obtain.
regions, however, rapid facies changes, abrupt transitions between Extensive use of surface geological mapping and conceptual model
adjacent depositional settings, and incised erosional features neces- development offsets this problem by constraining the stratigraphic
sitate relatively abundant data. For example, to resolve tunnel chan- framework for GIS modelling.
nels of <1 km width, a data coverage of ~2 to 3 deep, boreholes per
km2 is required (e.g., Weber and van Geuns, 1990). Such a data den- A well-structured stratigraphic database was an effective tool
sity can only be approached in regional studies with the integration to support stratigraphic modelling. A series of data vetting and for-
of archival data with map data. To produce the most credible geo- matting steps, database queries and GIS operations were developed
logical model possible, regional scale models need to exploit and to integrate variable quality data and use it with thematic GIS lay-
integrate a range of disparate data that includes primary geological ers to produce the ORM 3-D structural model. Although the use of
and hydrogeological data, thematic map data and archival data abundant, low-quality water well records was essential to build the
within a hierarchal data structure. ORM stratigraphic model, great care was exercised to identify and
filter out data errors. To integrate water well data, an expert system
Potential for Future Advancements was used to emulate the process used by geologists for interpreting
subsurface data. In this system, high-quality data and expert knowl-
Each surface of the ORM model was rendered individually and con- edge-based rules were used to train low-quality water well records.
strained by using a variety of GIS map-layer queries to insure con- In this way, more data support was added to the interpolated sur-
formity. For interpolation, both TIN and Natural Neighbour meth- faces while limiting the potential introduction of gross accuracy
ods were used; however, any exact interpolator could be used to problems. The expert system approach has the additional advan-
achieve similar results (e.g., kriging, conditional simulation). tages of being fully consistent in all well interpretations and it also
Kriging is a powerful geostatistical estimation technique, however allows controlled experiments to be run and re-run to evaluate the
it is computationally intensive and it assumes data stationarity. effects of new or updated data, rules and geological conceptualiza-
Kriging may be more appropriately applied to smaller, more geo- tion.
logically homogeneous zones (e.g., tunnel channel and inter-chan-
nel areas) in the ORM study area. Various types of conditional sim- Reliable data interpretation cannot be ensured in the absence of
ulation modelling could be employed to produce multiple realiza- an accurate conceptual geological model. However, any model
tions of each unit as was completed for the Newmarket Till aquitard should be recognized as being only one possible interpretation of
(e.g., Desbarats et al., 2001). In addition, such an approach would many. As such, a model needs to be treated with skepticism and
allow a quantitative assessment of uncertainty for the respective reviewed by objective testing against high-quality field data. A num-
surface geometries (e.g., Desbarats et al., 2001). ber of rigorous interpolative techniques may be used to produce a
stratigraphic model (e.g., TIN, kriging, conditional simulation).
A digital stratigraphic model should ultimately capture addi- However in any modelling exercise, a clear description of the con-
tional resolution from sedimentary deposits, in particular, the inter- ceptual geological model, related GIS modelling assumptions and
nal characterization of mapped units or architectural elements. The subsequent model limitations should be considered and documented.
heterogeneity of these stratigraphic elements has been recorded as
sediment facies in the ORM (e.g., Russell et al., 2005b) but they ACKNOWLEDGMENTS
have not yet been used explicitly in the modelling. Sediment facies
and their respective depositional environments are recognized to The authors would like to thank Andy Bajc, Donald Keefer, and
play an important role in characterizing the distribution of hydraulic Ross Knight for highly constructive comments and insight that
conductivity (Fogg, 1986). For most of the aquifers, the most abun- greatly improved this paper. Additionally, an informal review from
dant data available are the sediment facies derived directly from Steve Holysh was very much appreciated. This work was funded
detailed core logging or interpreted from geophysical data. The dis- under the Groundwater program of the Geological Survey of
tribution of facies within units can be modelled using a variety of Canada, Contribution #2003096.
techniques, including: transitional probability geostatistics (e.g.,
Weissmann and Fogg, 1999), indicator kriging or simulation meth- REFERENCES
ods (e.g., Ritzi et al., 1994). Applied to sedimentary aquifers these
techniques provide a powerful means of estimating aquifer hetero- Alms, R., Klesper, C., and Siehl, A., 1996, Three-dimensional mod-
geneity even if generic 'book' values are used for hydraulic conduc- elling of geological features with examples from the Cenozoic
tivity of sediment facies. lower Rhine Basin, in Forster, A., and Merriam, D.F., eds.,
Geological Modelling and Mapping: Plenum Press, p. 113-
CONCLUSIONS 133.
Anderson, M.P., 1989, Hydrogeologic facies models to delineate
A basin analysis approach and a modelling method that organizes large-scale spatial trends in glacial and glaciofluvial sedi-
and integrates variable data at regional scales has proven to be a ments: Geological Society of America Bulletin, v. 101, p. 501-
practical means of understanding and mapping glaciated terrain in a 511.
regional hydrogeology study. This paper has documented the devel- Barnett, P.J., 1992, Quaternary geology of Ontario, in Thurston,
opment of a digital geological model using a stratigraphic database P.C., Williams, H.R., Sutcliffe, R.H., and Stott, G.M., eds.,

538
SPECIAL PUBLICATION 44

Geology of Ontario: Ontario Geological Survey, Map 2560, Fogg, G.E., 1986, Groundwater flow and sand body interconnected-
scale 1:50,000. ness in a thick, multi-aquifer system: Water Resources
Barnett, P.J., Sharpe, D.R., Russell, H.A.J., Brennand, T.A., Gorrell, Research, v. 22, no. 5, p. 679-694.
G., Kenny, F., and Pugin, A., 1998, On the origins of the Oak Gerber, R.E., and Howard, K.W.F., 2002, Hydrogeology of the Oak
Ridges Moraine: Canadian Journal of Earth Sciences, v. 35, p. Ridges Moraine aquifer system: implications for protection
1152-1167. and management from the Duffins Creek watershed: Canadian
Berg, R.C., and Thorleifson, L.H., (convenors) 2001, Geological Journal of Earth Sciences, v. 39, p. 1333-1348.
Models for Groundwater Flow Modelling: Illinois State Herzog, B.L., Larson, D.R., Abert, C.C., Wilson, D., and Roadcap,
Geological Survey, Open File Series 2000-1. G.S., 2002, Hydrostratigraphic modelling of a complex glacial-
Bester, M.L., Molson, J.W., and Frind, E.O., 2002, Numerical sim- drift aquifer system for importation into MODFLOW: Ground
ulation of road salt impact at the Greenbrook Well Field, Water, v. 41, p. 57-65.
Kitchener, Ontario, in Stolle, D., Piggott, A.R., and Crowder, Hinton, M.J., 1995, Measuring stream discharge to infer the spatial
J.J., eds., Ground and Water: Theory to Practice: Proceedings distribution of groundwater discharge, in Proceedings of the
of the 55th Canadian Geotechnical and 3rd Joint IAH-CNC watershed management symposium: Canada Centre for Inland
and CGS Groundwater Specialty Conferences, Niagara Falls, Waters, Burlington, Ontario, Canada, Dec. 6-8, 1995, p. 27-32.
Ontario, October 20-23, 2002, p. 449-456. Howard, K.W.F., Eyles, N., Smart, P.J., Boyce, J.I., Gerber, R.E.,
Boyce, J.I., and Eyles, N., 2000, Architectural element analysis Salvatori, S.L., and Doughty, M., 1997, The Oak Ridges
applied to glacial deposits: internal geometry of a late Moraine of southern Ontario: a ground-water resource at risk,
Pleistocene till sheet, Ontario, Canada: Geological Society of in Eyles, N., ed., Environmental Geology of Urban Areas,
America Bulletin, v. 112, p. 98-118. GEOText 3: Geological Association of Canada, p. 153-172.
Brennand, T.A., Moore, A., Logan, C., Kenny, F.M., Russell, Hughes, J.D., 1993, GEOMODEL; an expert system for modelling
H.A.J., Sharpe, D.R., and Barnett, P.J., 1998, Bedrock topog- layered geological sequences applied to the assessment of
raphy of the Greater Toronto and Oak Ridges Moraine areas, coalfields, in Kirkham, R.V., Sinclair, W.D., Thorpe, R.I., and
southern Ontario: Geological Survey of Canada, Open File Duke, J.M., eds., Mineral Deposit Modelling: Geological
3419, scale 1: 200 000. Association of Canada, Special Paper 40, p. 707-734.
Brennand, T.A., and Shaw, J., 1994, Tunnel channels and associat- Karrow, P.F., 1967, Pleistocene Geology of the Scarborough Area:
ed landforms, south-central Ontario: their implications for ice- Ontario Ministry of Natural Resources, Report 46.
sheet hydrology: Canadian Journal of Earth Sciences, v. 31, p. Karrow, P.F., 1974, Till stratigraphy in parts of southwestern Ontario:
505-522. Geological Society of America Bulletin, v. 85, p. 761-768.
Chapman, L.J., and Putnam, D.F., 1984, The Physiography of Kassenaar, D., Holysh, S., Gerber, R.E., and Sharpe, D.R., 2003,
Southern Ontario, Special Volume 2: Ontario Geological Hydrostratigraphic interpretation in a complex aquifer system,
Survey, 270 p. Oak Ridges Moraine, Ontario, in 56th Canadian Geotechnical
Cherry, J.A., 1996, Politics and economics; geological research Conference and 4th joint IAH-CNC and CGS Groundwater
bridging the gulf, the near-term future, Part 2; Perspective 5, Speciality Conferences: Winnipeg, Manitoba, Sept. 28-Oct. 1.
From site investigations to site remediation; implications for Kenny, F.M., Chan, P., and Hunter, G., 1997, Quality control of the
hydrogeology: GSA Today, v. 6, p. 13-15. positional accuracy of records in Ontario's water well database
Coleman, A.P., 1932, The Pleistocene of the Toronto region: using automated GIS techniques, in Conference Proceedings
Ontario Department of Mines, Map no. 42g, scale 1:63,360. Geomatics in the Era of RADARSAT GER 97: Ottawa, May
Desbarats, A.J., Hinton, M., Logan, C., and Sharpe, D., 2001, 24-30, 1997.
Geostatistical mapping of leakance in a regional aquitard, Oak Kenny, F.M., Paquette, J., Russell, H.A.J., Moore, A.M., and
Ridges Moraine, Ontario, Canada: Hydrogeology Journal, v. 9, Hinton, M.J., 1999, A digital elevation model of the Greater
p. 79-96. Toronto area, southern Ontario and Lake Ontario bathymetry:
Desbarats, A.J., Logan, C.E., Hinton, M.J., and Sharpe, D.R., 2002, Geological Survey of Canada, Ontario Ministry of Natural
On the kriging of water table elevations using collateral infor- Resources, and Canadian Hydrographic Service; Geological
mation from a digital elevation model: Journal of Hydrology, Survey of Canada, Open File D3678.
v. 255, p. 25-38. LeGrand, H.E., and Rosen, L., 1998, Putting hydrogeological site
Eyles, N., 1997, Environmental Geology of Urban Areas, GEOText studies on track: Ground Water, v. 36, p. 193-194.
3: Geological Association of Canada, 590 p. LeGrand, H.E., and Rosen, L., 2000, Systematic makings of early
Eyles, N., Clark, B.M., Kaye, B.G., Howard, K.W.F., and Eyles, stage hydrogeological conceptual models: Ground Water, v.
C.H., 1985, The application of basin analysis techniques to 38, p. 887-893.
glaciated terrains; an example from the Lake Ontario Basin, Logan, C., Russell, H.A.J., and Sharpe, D.R., 2001, Regional three-
Canada: Geoscience Canada, v. 12, p. 22-32. dimensional stratigraphic modelling of the Oak Ridges
Fligg, K., and Rodrigues, B., 1983, Geophysical well log correla- Moraine area, southern Ontario, in Current Research 2001-D1:
tions between Barrie and the Oak Ridges Moraine: Water Geological Survey of Canada, p. 19.
Resources Branch, Ontario Ministry of the Environment, Map Logan, C., Russel, H.A.J., and Sharpe, D.R., 2005, Regional 3-D
2273. structural model of the Oak Ridges Moraine and Greater
Flint, A.L., Flint, L.E., Bodvarsson, G.S., Kwicklis, E.M., and Toronto area, southern Ontario: Version 2.1: Geological
Fabryka-Martin, J., 2001, Evolution of the conceptual model Survey of Canada, Open File 5062, 1 CD-ROM.
of unsaturated zone hydrology at Yucca Mountain, Nevada: Logan, C., Sharpe, D.R., and Russell, H.A.J., 2002, Regional 3D
Journal of Hydrology, v. 247, p. 1-30. structural model of the Oak Ridges Moraine and Greater

539
GIS FOR THE EARTH SCIENCES

Toronto area, southern Ontario: version 1.0: Geological Russell, H.A.J., Sharpe, D.R., and Logan, C., 2005b, Structural
Survey of Canada, Open File 4329. model of Oak Ridges Moraine and Greater Toronto areas,
MacFarlane, P.A., Doveton, J.H., Feldman, H.R., Butler jr., J.J., southern Ontario: Oak Ridges Moraine: Geological Survey of
Combes, J.M., and Collins, D.R., 1994, Aquifer/Aquitard units Canada, Open File 5065.
of the Dakota aquifer System in Kansas: Methods of delin- Russell, H.A.J., Sharpe, D.R., Logan, C., and Brennand, T.A., 2001,
eation and sedimentary architecture effects on ground-water Not without sedimentology: guiding groundwater studies in
flow and flow properties: Journal of Sedimentary Petrology, v. the Oak Ridges Moraine, southern Ontario: Illinois State
64, p. 464-480. Geological Survey, Open File Series 2000-1.
Martini, I.P., and Brookfield, M.E., 1995, Sequence analysis of Sharpe, D.R., Barnett, P.J., Brennand, T. A., Finley, D., Gorrell, G.,
Upper Pleistocene (Wisconsinan) glaciolacustrine deposits of Russell, H.A.J., and Stacey, P., 1997, Surficial geology of the
the north-shore bluffs of Lake Ontario, Canada: Journal of Greater Toronto and Oak Ridges Moraine area, southern
Sedimentary Research, v. 65B, p. 388-400. Ontario: Geological Survey of Canada, Open File 3062, scale
Meriano, M., and Eyles, N., 2003, Groundwater flow through 1:200 000.
Pleistocene glacial deposits in the rapidly urbanizing Rouge Sharpe, D.R., Dyke, L.D., Hinton, M.J., Pullan, S.E., Russell,
River-Highland Creek watershed, City of Scarborough, south- H.A.J., Brennand, T.A., Barnett, P.J., and Pugin, A., 1996,
ern Ontario, Canada: Hydrogeology Journal, v. 11, p. 288-303. Groundwater prospects in the Oak Ridges Moraine area, south-
Miall, A.D., 2000, Principles of Sedimentary Basin Analysis: New ern Ontario: application of regional geological models, in
York, Springer-Verlag, 616 p. Current Research 1996-E: Geological Survey of Canada, p.
Mossop, G.D., and Shetsen, I., 1994, Geological Atlas of the 181-190.
Western Canada Sedimentary Basin. On-line version: Sharpe, D.R., Hinton, M.J., Russell, H.A.J., and Desbarats, A.J.,
Edmonton, Alberta Geological Survey, http://www.ags.gov.ab. 2002a, The need for basin analysis in regional hydrogeological
ca/AGS_PUB/ATLAS_WWW/(1998,12,09). studies: Oak Ridges Moraine, Southern Ontario: Geoscience
Pugin, A., Pullan, S.E., and Sharpe, D.R., 1999, Seismic facies and Canada, v. 29, p. 3-20.
regional architecture of the Oak Ridges Moraine area, southern Sharpe, D.R., Pugin, A., Pullan, S.E., and Gorrell, G., 2003,
Ontario: Canadian Journal of Earth Sciences, v. 36, p. 409-432. Application of seismic stratigraphy and sedimentology to
Ritzi, R.W., Jayne, D.F., Zahradnik, A.J., Field, A.A., and Fogg, regional investigations: an example from Oak Ridges Moraine,
G.E., 1994, Geostatistical modelling of heterogeneity in southern Ontario, Canada: Canadian Geotechnical Journal, v.
glaciofluvial, buried-valley aquifers: Ground Water, v. 32, p. 40, p. 711-730.
666-674. Sharpe, D.R., Pugin, A., Pullan, S.E., and Shaw, J., 2004, Regional
Russell, H.A.J., and Arnott, R.W.C., 2003, Hydraulic jump and unconformities and the sedimentary architecture of the Oak
hyperconcentrated flow deposits of a glacigenic subaqueous Ridges Moraine area, southern Ontario: Canadian Journal of
fan: Oak Ridges Moraine, southern Ontario: Journal of Earth Sciences, v. 41, p. 183-198.
Sedimentary Research, v. 73, p. 887-905. Sharpe, D.R., Russell, H.A.J., and Logan, C., 2002b, Geological
Russell, H.A.J., Arnott, R.W.C., and Sharpe, D.R., 2003a, Evidence characterization of a regional aquitard: Newmarket Till, Oak
for rapid sedimentation in a tunnel channel, Oak Ridges Ridges Moraine area, southern Ontario, in Stolle, D., Piggott,
Moraine, southern Ontario, Canada: Sedimentary Geology, v. A.R., and Crowder, J.J., eds., Ground and Water: Theory to
160, p. 33-55. Practice: Proceedings of the 55th Canadian Geotechnical and
Russell, H.A.J., Brennand, T.A., Logan, C., and Sharpe, D.R., 3rd joint IAH-CNC and CGS Groundwater Specialty
1998a, Standardization and assessment of geological descrip- Conferences, Niagara Falls, Ontario, October 20-23, p. 219-226.
tions from water well records: Greater Toronto and Oak Ridges Sharpe, D.R., Russell, H.A.J., and Logan, C., 2005a, Structural
Moraine areas, southern Ontario, in Current Research 1998-E: model of Oak Ridges Moraine and Greater Toronto areas,
Geological Survey of Canada, p. 89-102. southern Ontario: Lower sediment: Geological Survey of
Russell, H.A.J., Logan, C., Brennand, T.A., Hinton, M., and Sharpe, Canada, Open File 5067.
D.R., 1996, A regional geoscience database: An example from Sharpe, D.R., Russell, H.A.J., and Logan, C., 2005b, Structural
the Oak Ridges Moraine NATMAP / Hydrogeology Project, in model of Oak Ridges Moraine and Greater Toronto areas,
Current Research 1996-E: Geological Survey of Canada, p. southern Ontario: Newmarket Till: Geological Survey of
191-200. Canada, Open File 5066.
Russell, H.A.J., Logan, C., Moore, A., Kenny, F.M., Brennand, Shaw, J., 1996, A meltwater model for Laurentide subglacial land-
T.A., Sharpe, D.R., and Barnett, P.J., 1998b, Sediment thick- scapes, in McCann, S.B., and Ford, D.C., eds., Geomorph-
ness of the Greater Toronto and Oak Ridges Moraine areas, ology Sans Frontieres: John Wiley & Sons Ltd., p. 181-236.
southern Ontario: Geological Survey of Canada, Open File Shaw, J., and Gilbert, R., 1990, Evidence for large-scale subglacial
2892, scale 1:200,000. meltwater flood events in southern Ontario and northern New
Russell, H.A.J., Sharpe, D.R., Brennand, T.A., Barnett, P.J., and York State: Geology, v. 18, p. 1169-1172.
Logan, C., 2003b, Tunnel channels of the Greater Toronto and Shaw, J., and Sharpe, D.R., 1987, Drumlin formation by subglacial
Oak Ridges Moraine areas, southern Ontario: Geological meltwater erosion: Canadian Journal of Earth Sciences, v. 24,
Survey of Canada, Open File 4485. p. 2316-2322.
Russell, H.A.J., Sharpe, D.R., and Logan, C., 2005a, Structural Thorleifson, L.H., and Berg, R.C., (convenors) 2002, Three-dimen-
model of Oak Ridges Moraine and Greater Toronto areas, sional geological mapping for groundwater applications:
southern Ontario: Halton Till: Geological Survey of Canada, workshop extended abstracts: Geological Survey of Canada,
Open File 5064. Open File 1449.

540
SPECIAL PUBLICATION 44

Thorleifson, L.H., Matile, G.L.D., Keller, G.R., and Pyne, D.M., Weber, K.J., and van Geuns, L.C., 1990, Framework for construct-
2002, Construction of a geological model for southern ing clastic reservoir simulation models: Journal of Petroleum
Manitoba for groundwater modelling, in Thorleifson, L.H., Technology, p. 1248-1253 and p. 1296-1297.
and Berg, R.C., eds., Three-Dimensional Geological Mapping Weissmann, G.S., Carle, S.F., and Fogg, G.E., 1999, Three-dimen-
for Groundwater Applications: Workshop Extended Abstracts: sional hydrofacies modelling based on soil surveys and transi-
Geological Survey of Canada, Open File 1449, p. 75-78. tion probability geostatistics: Water Resources Research, v. 35,
Turner, R.E., 1977, Oak Ridges Moraine Aquifer Complex: Ontario p. 1761-1770.
Ministry of the Environment, Water Resources Branch, Weissmann, G.S., and Fogg, G.E., 1999, Multi-scale alluvial fan
Hydrogeological Map 78-2. heterogeneity modelled with transition probability geostatis-
Walker, R.G., 1992, Facies, facies models and modern stratigraphic tics in a sequence stratigraphic framework: Journal of
concepts, in Walker, R.G., and James, N.J., eds., Facies Hydrology, v. 226, p. 48-65.
Models: Response to Sea Level Changes: Geological Yarus, J.M., and Chambers, R.L., 1994, Stochastic modelling and
Association of Canada, St. John's, Newfoundland, p. 1-14. geostatistics: principles, methods, and case studies, AAPG
Watt, A.K., 1957, Pleistocene geology and groundwater resources computer applications in geology, no 3: The American
of the township of North York, York County, in Anonymous, Association of Petroleum Geologists, p. 379.
ed., Annual Report 1955: Ontario Department of Mines.

541
GIS FOR THE EARTH SCIENCES

542

You might also like