Available via license: CC BY-NC 4.0
Content may be subject to copyright.
ORIGINAL RESEARCH
Influence of artefacts in marine digital terrain models on
habitat maps and species distribution models: a multiscale
assessment
Vincent Lecours
1,2,3
, Rodolphe Devillers
2,3
, Evan N. Edinger
3
, Craig J. Brown
3,4
&
Vanessa L. Lucieer
5
1
Fisheries & Aquatic Sciences, School of Forest Resources & Conservation, University of Florida, 7922 NW 71st Street, Gainesville 32653, Florida
2
Department of Geography, Marine Geomatics Research Lab, Memorial University of Newfoundland, 232 Elizabeth Avenue, St. John’s,
Newfoundland and Labrador, Canada A1B 3X9
3
Department of Geography, Marine Habitat Mapping Research Group, Memorial University of Newfoundland, 232 Elizabeth Avenue, St. John’s,
Newfoundland and Labrador, Canada A1B 3X9
4
Applied Research, Nova Scotia Community College, 80 Mawiomi Place, Dartmouth, Nova Scotia, Canada B2Y 0A5
5
Institute for Marine and Antarctic Studies, University of Tasmania, 20 Castray Esplanade, Battery Point, Tasmania 7004, Australia
Keywords
Artefacts, error propagation, habitat
mapping, multibeam bathymetry, species
distribution model, terrain analysis
Correspondence
Vincent Lecours, Fisheries & Aquatic Sciences,
School of Forest Resources & Conservation,
University of Florida, Gainesville, FL 32653.
Tel: +1 352 273 3617; Fax: +1 352 392
3672; E-mail: vlecours@ufl.edu
Funding Information
This project was funded by the Natural Sciences
and Engineering Research Council of Canada
(NSERC) and the Canadian Foundation for
Innovation (CFI). Memorial University Libraries
provided funding through their Open Access
Author Fund.
Editor: Duccio Rocchini
Associate Editor: Mat Disney
Received: 13 March 2017; Revised: 15 May
2017; Accepted: 21 May 2107
doi: 10.1002/rse2.49
Abstract
Remote sensing techniques are currently the main methods providing elevation
data used to produce Digital Terrain Models (DTM). Terrain attributes (e.g. slope,
orientation, rugosity) derived from DTMs are commonly used as surrogates of spe-
cies or habitat distribution in ecological studies. While DTMs’ errors are known to
propagate to terrain attributes, their impact on ecological analyses is however
rarely documented. This study assessed the impact of data acquisition artefacts on
habitat maps and species distribution models. DTMs of German Bank (off Nova
Scotia, Canada) at five different spatial scales were altered to artificially introduce
different levels of common data acquisition artefacts. These data were used in 615
unsupervised classifications to map potential habitat types based on biophysical
characteristics of the area, and in 615 supervised classifications (MaxEnt) to predict
sea scallop distribution across the area. Differences between maps and models built
from altered data and reference maps and models were assessed. Roll artefacts
decreased map accuracy (up to 14% lower) and artificially increased models’ per-
formances. Impacts from other types of artefacts were not consistent, either
decreasing or increasing accuracy and performance measures. The spatial distribu-
tion of habitats and spatial predictions of sea scallop distributions were always
affected by data quality (i.e. artefacts), spatial scale of the data, and the selection of
variables used in the classifications. This research demonstrates the importance of
these three factors in building a study design, and highlights the need for error
quantification protocols that can assist when maps and models are used in deci-
sion-making, for instance in conservation and management.
Introduction
In the last decades, remote sensing has become the main
method used for collecting elevation data used in the pro-
duction of Digital Terrain Models (DTM). All DTMs
carry a certain level of error (Gessler et al. 2009) caused
by random noise, systematic errors and artefacts (Wise
2000). Artefacts were characterized by Reuter et al.
(2009)) as “distinct erratic features” that are made of
improbable and incorrect values. Artefacts can be found
in DTMs collected from any remote sensing systems
(Fisher and Tate 2006; Sofia et al. 2013) and at all scales.
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use,
distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes.
1
Artefacts can be introduced by the interpolation method
used to create the DTM (Sofia et al. 2013), the motion
and location of the acquisition platform (Harrison et al.
2009), timing or log frequency issue in the surveying sys-
tem (Lecours and Devillers 2015) or a lack of or an inap-
propriate correction of ionospheric and atmospheric
conditions (Li and Goldstein 1990). Artefacts can be
problematic as they influence data quality more than
other types of errors like noise and imprecise measure-
ments (Rousseaux 2003) and can be very subtle in the
DTM (Filin 2003), making them “the most significant
errors in a spatial or statistical analysis because they are
not easily detected yet introduce significant bias” (Brown
and Bara 1994).
DTMs are now commonly used in Geographic Informa-
tion Systems (GIS) to derive terrain attributes (e.g. slope,
orientation, rugosity) that can be used as surrogates for
other phenomena in fields like ecology (Bolstad et al. 1998)
and biogeography (Franklin 2013). Artefacts in DTMs were
shown to sometimes propagate to the derived terrain attri-
butes (Sofia et al. 2013; Lecours et al. 2017a) and are likely
to impact subsequent analyses (Arbia et al. 1998; Heuvelink
1998). Mapping and quantifying error propagation
throughout analysis have received some attention in the
geospatial literature (e.g. Fisher and Tate 2006; Wilson
2012) but are rarely performed by DTM users from other
disciplines (van Niel and Austin 2007). Quantifying error
propagation from DTM is especially relevant for the pro-
duction of species distribution models (SDM) and habitat
maps (van Niel et al. 2004; Peters et al. 2009) that often
combine terrain attributes with other environmental data
(Franklin 1995; Guisan and Zimmermann 2000; Williams
et al. 2012; Leempoel et al. 2015). These maps and models
are regularly used to support decision-making in conserva-
tion (Miller 2010; Guisan et al. 2013). However, a lack of
understanding of errors, their propagation and spatial dis-
tribution in maps may result in inaccurate maps and mod-
els that could lead to inappropriate decisions (Beale and
Lennon 2012), and negative impacts on biodiversity or
stakeholders (Beven 2000; Regan et al. 2005; Etnoyer and
Morgan 2007). However, issues related to spatial data error
are often overlooked (but see van Niel et al. 2004 and Livne
and Svoray 2011). To our knowledge, the influence of data
acquisition artefact errors in DTMs has never been assessed
on maps resulting from SDM or habitat mapping exercises.
The objective of this study was to describe the impact
of some common remotely sensed data acquisition arte-
facts on marine habitat maps and SDMs. Our specific
objectives were to (1) quantify the impact of artefacts on
habitat maps accuracy and SDMs performance, to (2)
assess if impacts are dependent on spatial scale, and to
(3) assess if impacts can be attenuated when combining
the affected data with other environmental data of better
quality. Our hypotheses were that artefacts do negatively
affect habitat maps and SDMs, that the impacts are
greater at finer scales, and that the addition of relatively
better quality data reduces the impacts of artefacts on
maps and models.
Material and Methods
Case study and data
This article explored the impact of DTM artefacts on
habitat maps using a case study from the marine environ-
ment. The marine realm provides an ideal case as it has
been suggested that underwater DTMs, or Digital Bathy-
metric Models (DBM), may be more prone to errors and
artefacts than terrestrial DTMs (Hughes-Clarke et al.
1996; Passalacqua et al. 2015; Lecours et al. 2016a).
DBMs are often the only available datasets used to char-
acterize deep-water environments due to difficulties to
observe and sample other environmental characteristics
(Solan et al. 2003; Robinson et al. 2011). If multibeam
echosounders (MBES) are currently the best technology
enabling the collection of large DBMs (Kenny et al.
2003), most bathymetric surfaces generated from these
systems still contain some artefacts (Hughes-Clarke
2003a; Roman and Singh 2006). Since these artefacts are
often within hydrographic error standards (Hughes-
Clarke 2003a) and appear even when appropriate calibra-
tion and corrections are made (Erikstad et al. 2013), they
are often considered inherent to the data and tend to be
overlooked by DBM end-users.
This article used bathymetric data for German Bank,
off Nova Scotia (Canada), in the eastern Gulf of Maine
(Fig. 1). The surveyed area covers 3650 km
2
of the Sco-
tian Shelf and has been extensively studied in previous
works (e.g. DFO, 2006; Brown et al. 2012; Todd et al.
2012; Smith et al. 2017). Bathymetric data were collected
by the Canadian Hydrographic Service (CHS) and were
corrected in post-processing for tide, motion, and sound
velocity. The corrected soundings were used to generate
reference DBMs at five different spatial resolutions: 10 m,
25 m, 50 m, 75 m and 100 m in the bathymetric process-
ing software CARIS HIPS and SIPS v.9.0. These five refer-
ence DBMs were assumed to be free of artefacts, and
following methods described in Lecours et al. (2017a), 10
different amplitudes of heave, pitch, roll and time arte-
facts were artificially introduced in them by altering the
calibration measures of the different surveys (Table 1).
The ten levels of amplitude for each type of artefacts were
derived from the standard deviation (r; Table 1) of the
ship’s recorded range of motion at the time of surveys.
As described in Lecours et al. (2017a), these common
artefacts were selected based on their different theoretical
2ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
impact on data; pitch impacts bathymetric data in both
horizontal and vertical planes, heave impacts them in a
vertical plane, roll affects soundings that are further away
from the nadir in a vertical plane –consequently affecting
areas that overlap between different survey lines –and
time causes a relative shift of adjacent lines in the hori-
zontal plane (see Hughes-Clarke 1997, 2002, 2003a,b; Lur-
ton 2010). Similar artefacts can also be found in other
types of remote sensing like LiDAR (Brown and Bara
1994; Filin 2003; Lichti and Skaloud 2010). The artefacts
were systematically introduced to provide controlled
conditions that enable comparisons of results, as often
performed in evaluations of the impact of error on
analyses (Reuter et al. 2009).
Six terrain attributes that together summarize topo-
graphic variability were derived from the reference and
altered DBMs using the TASSE toolbox for ArcGIS
(Lecours 2015): slope, easterness and northerness, topo-
graphic mean, rugosity and topographic position (see
Lecours et al. 2016b, 2017b). Backscatter data (i.e. acous-
tic reflectance) were simultaneously recorded with the
bathymetric data. The backscatter data were processed
and transformed by Brown et al. (2012) into three deriva-
tive layers that inform on seafloor properties (e.g. surficial
geology, porosity): Q1, Q2 and Q3. Finally, two sets of
ground-truth data from Brown et al. (2012) were used:
(1) 3190 geo-referenced photographs of the seafloor clas-
sified into five habitat types (reef, glacial till, silt and
mud, silt with sediment bed forms, sand with sediment
bed forms and highly abundant sand dollars (Echinarach-
nius parma)), and (2) 4816 geo-referenced sea scallop
observations (Placopecten magellanicus). Details on how
these data were collected and processed and examples of
photographs of the seafloor can be found in DFO (2006)
and Brown et al. (2012).
Habitat Maps and SDMs
Using the 205 sets of bathymetric and terrain attribute sur-
faces (i.e. one set for each of the 10 levels of artefacts, for
the four types of artefacts, at five different resolutions, in
addition to a reference set for each of the five resolutions),
habitat maps and SDMs were produced for three scenarios.
First, maps and models were generated using only the
bathymetry and the six terrain attribute surfaces, thus
Figure 1. Digital bathymetric model of the
German Bank study area.
Table 1. Levels of artefacts introduced in the five reference DBMs.
Standard deviations (r) were derived from the recorded motion at
time of survey.
Level of induced artefact
r2r3r4r5r
Heave (m) 0.33 0.66 0.99 1.32 1.65
Pitch (°)1.65 3.30 4.95 6.60 8.25
Roll (°)1.01 2.02 3.03 4.04 5.05
Time (sec) 0.25 0.50 0.75 1.00 1.25
A positive pitch indicates that the bow is up and a positive roll means
that the port side is up.
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 3
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
accounting only for terrain morphology (i.e. hereafter
referred to as “7 layers” scenario). Then, maps and models
were produced using all the available data (bathymetry,
slope, easterness, northerness, topographic mean, rugosity,
topographic position, Q1, Q2 and Q3) (i.e. “10 layers” sce-
nario). The addition of backscatter data in this scenario
was done to address our third hypothesis regarding the
addition of relatively better quality data to the mapping
process. Finally, maps and models were built using only
non-correlated variables (i.e. “8 layers” scenario). On Ger-
man Bank, the steepest areas are also the ones with the
highest rugosity, resulting in a high correlation between the
slope and rugosity data layers. Also, bathymetry is highly
correlated with topographic mean as they are closely
related (see Lecours et al. 2017b). Rugosity and topo-
graphic mean were therefore not used for the last sets of
maps and models. The data for the eight layers scenario
were thus bathymetry, slope, easterness, northerness, topo-
graphic position, Q1, Q2 and Q3. Overall, 615 habitat
maps and 615 SDMs were produced and analysed by using
the 205 sets of data for each of the three scenarios with
both approaches.
The method used to generate habitat maps is based on
the concept of benthoscape (Zajac 2008), a representation
of the biophysical characteristics of an area generated by
adopting a landscape style approach similar to when maps
of landscape features are generated from terrestrial data-
sets. Such approach was used by Brown et al. (2012) to
map features on the seafloor that could be resolved
within the acoustic remotely sensed data, thus not
attempting to delineate seafloor attributes beyond what
the remote sensing techniques were capable of resolving.
This approach segments the different data layers into a
statistically optimum number of classes that are then spa-
tially compared to the geo-referenced photographs and
recombined based on best match with the different habi-
tat types (see Brown et al. 2012). The Modified k-Means
unsupervised classification tool of Whitebox GAT v.3.2
was used to produce these maps, and confusion matrices
were built to calculate the kappa coefficient of agreement
of each map (Boyce et al. 2002).
SDMs were generated based on maximum entropy
(MaxEnt), a common and effective method (Phillips et al.
2006; Monk et al. 2010), that used the sea scallop obser-
vations to segment the environmental data and quantify
sea scallop habitat suitability across the area. SDMs were
computed using the MaxEnt software v.3.3.3k with the
same settings as in Brown et al. (2012). Area under the
curve (AUC) derived from threshold independent receiver
operating curves were also measured to quantify the per-
formance of the models and enable comparisons (Phillips
et al. 2006): AUC
Train
was measured to evaluate the good-
ness-of-fit of models to the training data, and AUC
Test
was used to evaluate the ability of models to perform well
on an independent dataset (i.e. validation samples) (Fitz-
patrick et al. 2013). These two measures were combined
to compare models’ performance, robustness and general-
izability (Vaughan and Ormerod 2005; Warren and Seifert
2011). AUC
Diff
–the difference between AUC
Train
and
AUC
Test
–was used to quantify generalizability (i.e.
transportability, transferability): a high value is an indica-
tion that a model over-fitted the training data and does
not replicate well to a different dataset. Details on these
measures can be found in Lecours et al. (2016b). Finally,
correlations between model outputs were calculated to
evaluate spatial similarity of predictions.
Results
Habitat maps
The average kappa coefficients of agreement of all maps
produced with altered data and their standard deviation
are presented in the supporting information (Table S1),
while the individual kappa of the 615 habitat maps are
presented in Figure 2. In general, habitat maps produced
using 10 layers provided the best classifications, followed
by those using eight layers and those built from only seven
layers. However, map accuracy varied less for the scenario
with eight layers (i.e. it was more consistent). Heave was
generally the artefact type that made map accuracy vary
the least, while roll artefacts usually made map accuracy
vary the most. In average (Table S1), only three sets of
maps showed a scale-dependent pattern for which maps
produced from finer-scale data were more impacted by
artefacts than maps made from broader-scale data. These
sets all belong to the scenario with 10 layers and were
maps impacted by pitch, roll and time. A visual assessment
of the results showed that the presence of artefacts in data
used to produce habitat maps has a noticeable influence
on the spatial distribution of the habitats (cf. Fig. 3).
When matching the total area misclassified because of
artefacts–that is, when comparing the classifications made
from altered data to one made from reference data–to the
difference in kappa coefficient between these maps, results
show that a little difference in kappa coefficient can trans-
late into large differences in spatial output (cf. Fig. 3C).
Except for maps affected by roll artefacts, the reference
maps did not always produce the best outcome in terms
of accuracy: while all habitat maps impacted by any level
of roll artefact performed worse than the reference maps,
an important number of maps made from altered data
had a higher kappa coefficient than their corresponding
reference maps. This was observed regardless of scale and
scenario. Overall, 47% of the habitat maps altered by
pitch had a higher kappa coefficient than their
4ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
corresponding reference map, and that percentage was
higher for time (50%) and heave (55%). No particular
scale-dependent patterns were observed, except for habitat
maps made from seven layers and impacted by pitch, for
which a greater amount of maps performed better than
the reference maps at broader scales.
55
60
65
70
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
Level of heave (m)
Level of time (s)
Level of pitch (°)
Level of roll (°)
55
60
65
70
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
55
60
65
70
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
55
60
65
70
–9–8–7–6–5–4–3–2–1 0 1 2 3 4 5 6 7 8 9
55
60
65
70
–9–8–7–6–5–4–3–2–1 0 1 2 3 4 5 6 7 8 9
40
45
50
55
60
65
70
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6
40
45
50
55
60
65
70
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6
55
60
65
70
–1.5 –0.5 0.5 1.5
55
60
65
70
–1.5 –0.5 0.5 1.5
55
60
65
70
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
55
60
65
70
–9–8–7–6–5–4–3–2–1 0 1 2 3 4 5 6 7 8 9
40
45
50
55
60
65
70
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6
55
60
65
70
–1.5 –0.5 0.5 1.5
7 layers 8 layers 10 layers
7 layers 8 layers 10 layers
7 layers 8 layers 10 layers
7 layers 8 layers 10 layers
Kappa coefficient (%) Kappa coefficient (%) Kappa coefficient (%) Kappa coefficient (%)
10 m 25 m 50 m 75 m 100 m
Figure 2. Kappa coefficients of agreement of the 615 habitat maps.
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 5
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
(A)
(B)
(C)
Figure 3. Examples of habitat maps produced with eight layers at 50 m resolution, overlaid by the ground-truth data. The colour of the ground-
truth data matches the classification’s colour when appropriately classified. (A) shows the reference map that was built with data that were
assumed free of artefacts. (B) shows maps built from data that were impacted by different types of artefacts. (C) shows the spatial distribution of
the change in habitat map classification between the maps from (B) and the map from (A). Red pixels indicate change while grey pixels are those
that were classified as the same habitat type in the two compared classifications.
6ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
Species distribution models
Figure 4 shows how the SDMs’ performance and robust-
ness change as artefacts are introduced in the input DBMs
for the eight layers scenario. For all types of artefacts, no
pattern could be observed regarding whether some scales
were more impacted than others, or whether a greater
level of artefact resulted in higher or lower performance
or robustness.
In general, results show that introducing heave artefacts
decreased models’ performance and tend to also decrease
models’ robustness. One major exception was observed to
these patterns: at 75 m resolution, 7 and 10 layers models
performed better than the reference models. The two ref-
erence models in these cases had a higher standard devia-
tion and a lower AUC
Test
. Overall, about 87% of models
impacted by heave had a lower performance index than
their comparable reference model. About 39% of models
with pitch artefacts had a higher AUC
Test
than their
respective reference model while 36% of them were more
robust, which resulted in 41% of these models with a
higher performance index than their reference model
(Fig. 4). Roll artefacts boosted model performance, as
87% of models built from altered data had higher
AUC
Test
measures than the reference models. However,
only 26% of the models impacted by roll artefacts were
more robust than the reference models. The combination
of these two metrics into the performance index indicated
that 56% of models impacted by artefacts had a higher
index than the reference models. In terms of time arte-
facts, 29% of the models that were built from altered data
performed better than the reference models and 11% were
more robust, resulting in 26% of them having a higher
performance index than reference models.
In terms of generalizability, it was difficult to find any
consistent general patterns except for those models
affected by roll: 95% of them showed a greater generaliz-
ability index than the reference models. On the other
end, the presence of heave artefacts decreased the general-
izability in 80% of the cases, compared to 53% of models
affected by pitch and 66% of those affected by time
(Fig. 5).
Regarding spatial outputs, the presence of artefacts
always introduced discrepancies in the distribution of rel-
ative habitat suitability (cf. Figs. 6 and S1). While the
average discrepancies could be globally small (e.g. 2.2%),
they could be locally important (e.g. 58.9%). It is also
interesting to note that a high measure of correlation
between a model impacted by artefacts and its reference
model did not necessarily involve high similarity between
those. For instance, a correlation coefficient of 0.963 still
resulted in 23% of the area for which differences in rela-
tive habitat suitability were >5% (Fig. 6).
In general, models accounting only for topography and
depth (seven layers) were consistently the most affected
by artefacts in comparison to the reference models
10 m 25 m 50 m 75 m 100 m
Performance index
Level of heave (m)
85
90
95
100
105
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
85
90
95
100
105
–9 –8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9
85
90
95
100
105
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6
85
90
95
100
105
–1.5 –1 –0.5 0 0.5 1 1.5
Level of roll (°)
Level of pitch (°)
Level of time (s)
Figure 4. Change in performance index (ratio of AUC
Test
on standard deviation) as the level of artefacts in the data changes, for the models
built from eight layers. Models that perform better have a high AUC
Test
and models that are more robust have a low standard deviation. High-
performance models are often less robust than less performing models: the performance index thus captures the trade-off between performance
and robustness, with higher values of performance index indicating a better trade-off.
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 7
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
(Fig. S1), while those built with uncorrelated variables
(eight layers) were often the least impacted. The ranges in
correlation coefficients (Fig. S1) were usually not very big
for heave artefacts and were in average lower for roll arte-
facts. Overall, roll seemed to have the most impact on the
spatial distribution of relative habitat suitability of sea
scallop. No clear pattern was observed in terms of scale,
although roll artefacts seemed to produce models that
were more similar to the reference ones at coarser scales,
and the extreme scales (10 and 100 m) seemed to be a bit
more impacted by heave, time and pitch than the inter-
mediate scales (25–75 m).
Discussion
Impacts of artefacts on habitat maps and
SDMs
Our first hypothesis was that artefacts in bathymetry
that propagate to terrain attributes would impact habi-
tat maps and SDMs in a negative way. Results show
that this is not always the case. While we were expect-
ing map accuracy to decrease as a function of level of
artefacts, only maps impacted by roll demonstrated such
relationship. Results show that the other types of arte-
facts sometimes artificially increased map accuracy,
although not in a predictable way. A higher level of
artefact did not necessarily result in a better or worse
map or model than a lower level of artefact. About half
of the habitat maps produced with data altered by
heave, pitch and time artefacts performed better than
the reference maps. While these results suggest a ran-
dom pattern, they may also have been influenced by
the approach used to quantify map accuracy. Since the
habitats are represented on maps as clusters of pixels
showing similar characteristics, they share some charac-
teristics with areal data. Artefacts might thus influence
the boundaries of these “zones” more than the area
inside them. Because the ground-truth data are points
that are more likely to fall within the middle of a zone
than at its boundary, the kappa coefficients of agree-
ment may not capture the change in boundary. A spa-
tial assessment of the differences between the different
habitat maps, as performed for instance in Figure 3C
and in Diesing et al. (2014), could help better capture
the influence of artefacts on the delineation of the dif-
ferent habitat zones. Considering the amount of maps
produced in this study, this would be computationally
intensive but such an approach should be considered in
future work. These results however yield an important
conclusion regarding the methods commonly used in
the literature to quantify classification and habitat map
accuracy: measures using point data to validate classifi-
cations of zones may be biased by not capturing the
variability of the classifications along zone boundaries.
The analysis of SDMs yielded similar conclusions to the
analysis of habitat maps but from different types of arte-
facts. Heave artefacts had generally a negative impact on
10 m 25 m 50 m 75 m 100 m
Generalizability index
Level of heave (m)
Level of roll (°)
Level of pitch (°)
Level of time (s)
5
10
15
20
25
–6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6
5
6
7
8
9
10
–9 –8 –7 –6 –5 –4 –3 –2 –1 0 1 2 3 4 5 6 7 8 9
5
6
7
8
9
10
–1.5 –1 –0.5 0 0.5 1 1.5
5
6
7
8
9
10
–2 –1.5 –1 –0.5 0 0.5 1 1.5 2
Figure 5. Change in generalizability index (ratio of AUC
Train
on AUC
Diff
) as the level of artefacts in the data changes, for the models built from
eight layers. Higher values indicate more generalizable or replicable models.
8ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
the performance of models, with some exceptions (e.g.
75 m resolution models–which may indicate that this par-
ticular scale does not capture the relevant drivers of species
distribution). Some models impacted by pitch and time
performed better than the reference models. Models
impacted by roll artefacts clearly contradicted our hypothe-
sis: the performance of most of these models was artifi-
cially increased by the presence of roll artefacts. This could
be explained by the fact that sea scallops distribution is
driven by rugosity (Brown et al. 2012), and artefacts like
roll and pitch artificially increase the rugosity of an area.
Models produced with data impacted by these artefacts
would thus artificially increase the relative habitat suitabil-
ity of sea scallops across the entire area, resulting in a
higher prediction success when validated against the test
data. The increase in high values of relative habitat suit-
ability was confirmed by visual comparison (cf. Fig. 6) but
also by the high differences in spatial correlation recorded
for pitch and particularly for roll (cf. Fig. S1).
Our second hypothesis stated that the impacts of artefacts
in bathymetry and terrain attributes should be greater at
finer scales. This hypothesis was based on the fact that the
propagation of artefacts from DTM to terrain attributes was
previously found to be scale-dependent (cf. Lecours et al.
2017a). Results from both unsupervised and supervised clas-
sifications did not confirm, neither did they refute, this
hypothesis as no particular scale-dependent patterns could
be identified. The difference between the scale-dependent
propagation of DTM artefacts in terrain attributes and the
scale-independent propagation of these artefacts in habitat
Habitat suitability
Value
High : 1
Low : 0
Differences
Less than 5%
5% to 10%
10% to 15%
15% to 20%
20% to 25%
More than 25%
Heave Pitch
Roll Time
Minimum: 0.0%
Maximum: 58.9%
Mean: 2.2%
Standard deviation: 3.2%
Minimum: 0.0%
Maximum: 63.0%
Mean: 3.4%
Standard deviation: 4.1%
Minimum: 0.0%
Maximum: 74.6%
Mean: 8.6%
Standard deviation: 8.0%
Minimum: 0.0%
Maximum: 58.1%
Mean: 2.2%
Standard deviation: 3.2%
No artefact
Proportion of change >5%: 58%
Correlation with reference map: 0.832
Proportion of change >5%: 11%
Correlation with reference map: 0.980
Proportion of change >5%: 11%
Correlation with reference map: 0.980
Proportion of change >5%: 23%
Correlation with reference map: 0.963
Figure 6. Differences in distribution of relative habitat suitability of sea scallops between models affected by artefacts and a reference model
(top). The scenario represented is the one with eight layers at 50 m resolution. The level of error represented is the highest one (5r, Table 1).
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 9
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
maps and SDMs may be explained by the integration of a
biological/ecological context. The presence of artefacts in
finer-scale data may not result in a poor habitat classification
if these data and the scales at which they were collected and
analysed do not have an ecological meaning or do not match
the ecological scale of the phenomenon being studied, thus
being unsuitable regardless of their quality.
Finally, our third hypothesis was that the addition of rela-
tively better quality data would reduce the impacts of arte-
facts on maps and models. Results suggest that this
hypothesis is true for the habitat maps, as maps built with
the relatively good quality backscatter data were generally
more accurate. It however remains unclear whether this
improvement was caused by the quality of the data or their
nature (i.e. backscatter), which in this particular case was
known to be ecologically relevant (Brown et al. 2012). The
latter option is the most likely, considering results from
Lecours et al. (2016b) that showed that maps and models of
the same area produced only with backscatter and depth
data performed very well. Further work is thus required to
validate or invalidate this hypothesis with more certainty. In
addition, results showed that the range in measures of accu-
racy was more stable when uncorrelated data were used,
which could be an indication that a better choice in input
variable has the potential to stabilize and attenuate the
impacts of artefacts.
Spatial errors in ecology: comparisons with
other studies
In terms of data quality, the ecological literature has been
oriented mostly towards measurement uncertainty, and
the work performed on errors has largely focused on the
positional accuracy of species observations (e.g. Moudry
and
S
ımov
a 2012). The impact of DTM artefacts has been
studied before in a geomorphology and geomorphometry
context (e.g. Bonin and Rousseaux 2005) but rarely in
ecology. Of note is the work by van Niel et al. (2004) and
van Niel and Austin (2007) that studied the effect of error
in DTM on terrain attributes and predictive vegetation
modelling. Despite different approaches and types of error
studied, these studies and the current one yielded similar
conclusions regarding the fact that errors do propagate
throughout analyses, and affect distribution models
although not in an easily predictable way. In another eco-
logical study that looked at uncertainty and error propa-
gation, Livne and Svoray (2011) identified the need to
focus on assessing the behaviour of ecological models to
spatial errors at different spatial resolutions. While this
was addressed in the current study, results did not indi-
cate any scale-dependent pattern.
In the marine environment, researchers are aware of
artefacts as they are often, although not always (e.g.
Lucieer et al. 2012), acknowledged (e.g. Blondel and
G
omez Sichi 2009). When acknowledged, their implica-
tions for the ecological analysis being performed are often
not discussed (e.g. Kostylev et al. 2001). The presence of
artefacts in MBES data sometimes prevents their use or
the use of their derived terrain attributes in ecological
applications (e.g. Clements et al. 2010). When such data
are still used, artefacts have been linked to habitat mis-
classifications (e.g. Micallef et al. 2012; Costa and Battista
2013), to noise in results from unsupervised classifications
(e.g. Galparsoro et al. 2015), and to difficulties associated
with identification of seabed features (e.g., Dolan and
Lucieer 2014), among other consequences. As part of
their seabed mapping guidelines, the Norwegian Hydro-
graphic Service indicated that “seabed features shall not
be camouflaged by artefacts and artefacts must not appear
as seabed features” (NHS, 2013), and that artefacts in the
processed bathymetry “shall be kept at an insignificant
level not disturbing the seabed image” (NHS, 2013).
However, no procedures are indicated to assist in making
decisions regarding how to deal with artefacts when they
cannot be removed. This overview of the literature reflects
the lack of understanding of how artefacts impact ecologi-
cal analyses and interpretations, and the lack of knowl-
edge on how to respond to the presence of artefacts. The
work by Zieger et al. (2009) is however noteworthy as
they used terrain attributes and seafloor classification to
identify artefacts in flat areas before correcting for the
misclassifications caused by artefacts. Such methods may
however be inefficient in more complex areas as the clas-
sifications may be unable to distinguish which bathymet-
ric patterns are artefacts and which are actual natural
features.
While this study has focused on artefacts in multibeam
bathymetric data, backscatter data are also often impacted
by artefacts (e.g. Collier and Brown 2005; Che Hasan
et al. 2012). Like for bathymetric data, some of these arte-
facts can be removed in post-processing (e.g. De Falco
et al. 2010; Lamarche et al. 2011) but a complete removal
is not always achieved. Backscatter data with artefacts
have been widely used (e.g. Rattray et al. 2009; Roberts
et al. 2009) as they may still yield useful observations.
Other times however, they are judged unusable for the
mapping or modelling exercise (e.g. Holmes et al. 2008).
It has been recognized that there is a broad misunder-
standing of backscatter within the end user community
(Lurton and Lamarche 2015). In this study, backscatter
data were used to evaluate the impact of adding relatively
better quality data to poor quality data within the same
analysis. As done with bathymetry in this study, future
work should evaluate the impacts of artefacts in backscat-
ter data on habitat maps and SDMs. It is to be expected
that like for bathymetry and terrain attributes, artefacts in
10 ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
backscatter data will have a greater impact if sediment
properties are ecologically relevant to the species, area or
problem studied. For instance, Copeland et al. (2013)
noted that artefacts in the backscatter data resulted in an
apparent striping pattern in their habitat classifications.
Implications for ecological applications
The results of this study have critical implications for eco-
logical studies that use DTMs and their derived terrain
attributes in their applications, which is a common prac-
tice (Bouchet et al. 2015; Lecours et al. 2016a). The use
of environmental variables such as terrain attributes has
been shown to improve predictions accuracy in SDMs
(Dobrowski et al. 2008). However, this study showed that
when artefact errors are present in DTMs, there is a
trade-off between the improved prediction that would be
gained from including the DTM and its derived terrain
attributes and the risk to produce inaccurate predictions.
Results showed that such predictions are not necessarily
revealed as lower or absence of predictions, but can be
important inflation in predictions. For instance, artefacts
may alter the quantification of species-environment rela-
tionships by artificially increasing the importance of
rugosity in habitat characterization. When rugosity is
known to be a surrogate of a particular species distribu-
tion, this leads to an overestimation of the suitable habi-
tat for that species.
Studies that include any assessment of data quality are
rare (van Niel and Austin 2007). The current study high-
lighted the sensitivity of maps and models to the observa-
tional scale and spatial errors like artefacts. Many calls have
been made in the literature for the quantification of uncer-
tainty and error propagation throughout ecological analy-
ses (e.g. Guisan et al. 2006; Lecours et al. 2015), and tools
have been proposed to deal with uncertainty (e.g. the Data
Uncertainty Engine by Brown and Heuvelink 2007) but not
with errors. The ecological community that makes use of
GIS tools and remote sensing techniques is usually aware of
this need but such protocol are not yet implemented in any
workflow. As stated by Li et al. (2012): “there are user com-
munities who may be aware of spatial data quality issues
but may not have at their disposal techniques and tools for
data quality assurance.” Such tools, associated with proper
standards, protocols and metadata, are becoming crucial to
enable a proper incorporation of error modelling in the dif-
ferent applications workflow. This will eventually lead to
results and interpretation that are grounded on solid foun-
dations, and more informed decisions. While it is impossi-
ble to avoid error and uncertainty in ecological analyses, it
is also important that practitioners stop avoiding it. An
acknowledgement of errors like artefacts and a discussion
on their potential impact on analyses will increase the
chances to make more informed decisions when these data
and analyses are used in contexts like conservation plan-
ning.
Conclusions
DTMs and terrain attributes are now commonly used in
ecological studies. Despite an awareness of the presence of
errors like artefacts in these data, their quality is rarely
assessed, acknowledged or discussed. The goal of this
study was to develop evidence linking the presence of
artefacts in DTMs with the accuracy of analyses per-
formed in ecological applications. Results demonstrated
that artefacts do impact habitat maps and SDMs,
although not in a predictable way. Roll artefacts showed
the most predictable influence, decreasing the accuracy of
habitat maps and artificially increasing the performance
and generalizability of SDMs. Other types of artefacts
sometimes increased map accuracy and model perfor-
mance and generalizability or decreased them. These con-
clusions may however change if different data were used;
perhaps that with higher-resolution data (e.g. 0.5–2 m),
the relative magnitude of the artefacts would be more
important and produce a much larger and consistent
effect. Results showed that the importance of the impacts
of artefacts on ecological applications strongly depend on
whether or not the methods are grounded in ecological
relevance, particularly in terms of the choice of variables
and the spatial scale of the data. While the influence of
errors on an analysis depends on the type and require-
ments of the analysis (Friedl et al., 2001), results gained
in this study are transposable to other applications that
use remotely sensed data like LiDAR-derived DTMs and
encounter similar artefacts. This study also highlighted
requirements for error quantification tools to become
widely available to scientists and practitioners with a wide
range of background and expertise. This will improve
standards and protocols and lead to more quality-aware
decisions in contexts like conservation.
Acknowledgements
We thank Emma LeClerc for her insightful comments on
early drafts of this manuscript, and two anonymous
reviewers for their comments on the final version. We
also thank Jessica A. Sameoto, Dr. Stephen J. Smith and
the Canadian Hydrographic Service (Department of Fish-
eries and Oceans Canada), for sharing their data of Ger-
man Bank. This project was funded by the Natural
Sciences and Engineering Research Council of Canada
(NSERC) and the Canadian Foundation for Innovation
(CFI). Memorial University Libraries provided funding
through their “Open Access Author Fund”
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 11
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
Data Accessibility
Data are available upon request from the Department of
Fisheries and Oceans, Canada, for researchers who meet
the criteria for access. Requests can be sent to
chsinfo@dfo-mpo.gc.ca.
References
Arbia, G., D. Griffith, and R. Haining. 1998. Error propagation
and modelling in raster GIS: overlay operations. Int. J.
Geogr. Inform. Sci. 12, 145–167.
Beale, C. M., and J. J. Lennon. 2012. Incorporating uncertainty
in predictive species distribution modelling. Philosophical
Transactions of the Royal Society B 367, 247–258.
Beven, K. 2000. On model uncertainty, risk and decision
making. Hydrol. Process. 14, 2605–2606.
Blondel, P., and O. G
omez Sichi. 2009. Textural analyses of
multibeam sonar imagery from Stanton Banks, Northern
Ireland continental shelf. Appl. Acoust. 70, 1288–1297.
Bolstad, P. V., W. Swank, and J. Vose. 1998. Predicting
Southern Appalachian overstory vegetation with digital
terrain data. Landscape Ecol. 13, 271–283.
Bonin, O., and F. Rousseaux. 2005. Digital terrain model
computation from contour lines: how to derive quality
information from artifact analysis. GeoInformatica 9,253–
268.
Bouchet, P. J., J. J. Meeuwig, C. P. Salgado Kent, T. B.
Letessier, and C. K. Jenner. 2015. Topographic determinants
of mobile vertebrate predator hotspots: current knowledge
and future directions. Biol. Rev. 90, 699–728.
Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A.
Schmiegelow. 2002. Evaluating resource selection functions.
Ecol. Model. 157, 281–300.
Brown, D. G., and T. J. Bara. 1994. Recognition and reduction
of systematic error in elevation and derivative surfaces from
71/2-minute DEMs. Photogramm. Eng. Remote Sensing 60,
189–194.
Brown, J. D., and G. B. M. Heuvelink. 2007. The Data
Uncertainty Engine (DUE): a software tool for assessing and
simulating uncertain environmental variables. Comput.
Geosci. 33, 172–190.
Brown, C. J., J. A. Sameoto, and S. J. Smith. 2012. Multiple
methods, maps, and management applications: purpose
made seafloor maps in support of ocean management. J. Sea
Res. 72,1–13.
Che Hasan, R., D. Ierodiaconou, and L. Laurenson. 2012.
Combining angular response classification and backscatter
imagery segmentation for benthic biological habitat
mapping. Estuar. Coast. Shelf Sci. 97,1–9.
Clements, A. J., , J. A. Strong, C. Flanagan, and M. Service
2010. Objective stratification and sampling-effort allocation
of ground-truthing in benthic-mapping surveys. ICES J.
Mar. Sci. 67:628–637.
Collier, J. S., and C. J. Brown. 2005. Correlation of sidescan
backscatter with grain size distribution of surficial seabed
sediments. Mar. Geol. 214, 431–449.
Copeland, A., E. Edinger, R. Devillers, T. Bell, P. LeBlanc, and J.
Wroblewski. 2013. Marine habitat mapping in support of
marine protected area management in a subarctic fjord: Gilbert
Bay, Labrador, Canada. J. Coast. Conserv. 17,225–237.
Costa, B. M., and T. A. Battista. 2013. The semi-automated
classification of acoustic imagery for characterizing coral
reef ecosystems. Int. J. Remote Sens. 34, 6389–6422.
De Falco, G., R. Tonielli, G. Di Martino, S. Innangi, S.
Simeone, and I. M. Parnum. 2010. Relationships between
multibeam backscatter, sediment grain size and Posidonia
oceanica seagrass distribution. Cont. Shelf Res. 30,
1941–1950.
DFO. 2006. Presentation and review of Southwest Nova Scotia
benthic mapping project. DFO Canadian Science Advisory
Secretariat, Proceedings Series 2006/047.
Diesing, M., S. L. Green, D. Stephens, R. M. Lark, H. A.
Stewart, and D. Dove. 2014. Mapping seabed sediments:
comparison of manual, geostatistical, object-based image
analysis and machine learning approaches. Cont. Shelf Res.
84, 107–119.
Dobrowski, S. Z., H. D. Safford, Y. B. Cheng, and S. L. Ustin.
2008. Mapping mountain vegetation using species
distribution modeling, image-based texture analysis, and
object-based classification. Appl. Veg. Sci. 11, 499–508.
Dolan, M. F. J., and V. L. Lucieer. 2014. Variation and
uncertainty in bathymetric slope calculations using
geographic information systems. Mar. Geodesy 37, 187–219.
Erikstad, L., V. Bakkestuen, T. Bekkby, and R. Halvorsen.
2013. Impact of scale and quality of digital terrain models
on predictability of seabed terrain types. Mar. Geodesy 36,
2–21.
Etnoyer, P., and L. E. Morgan. 2007. Predictive habitat model
for deep gorgonians needs better resolution: comment on
Bryan & Metaxas (2007). Mar. Ecol. Prog. Ser. 339, 311–312.
Filin, S. 2003. Recovery of systematic biases in laser altimetry
data using natural surfaces. Photogramm. Eng. Remote
Sensing 69, 1235–1242.
Fisher, P. F., and N. J. Tate. 2006. Causes and consequences of
error in digital elevation models. Prog. Phys. Geogr. 30,
467–489.
Fitzpatrick, M. C., N. J. Gotelli, and A. M. Ellison. 2013.
MaxEnt versus MaxLike: empirical comparisons with ant
species distributions. Ecosphere 4,1–15.
Franklin, J. 1995. Predictive vegetation mapping: geographic
modelling of biospatial patterns in relation to environmental
gradients. Prog. Phys. Geogr. 19, 474–499.
Franklin, J. 2013. Species distribution models in conservation
biogeography: developments and challenges. Divers. Distrib.
19, 1217–1223.
Friedl, M. A., K. C. McGwire, and D. K. McIver. 2001. An
overview of uncertainty in optical remotely sensed data. Pp.
12 ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
258–283 in C. T. Hunsaker, M. F. Goodchild, M. A. Friedl,
and T. J. Case, eds. Spatial uncertainty in ecology:
Implications for remote sensing and GIS applications.
Springer-Verlag, New York.
Galparsoro, I., X. Agrafojo, M. Roche, and K. Degrendele.
2015. Comparison of supervised and unsupervised
automatic classification methods for sediment types
mapping using multibeam echosounder and grab sampling.
Ital. J. Geosci. 134,41–49.
Gessler, P., R. Pike, R. A. MacMillan, T. Hengl, and H. I
Reuter 2009. The future of geomorphometry. Pp. 637–652.
in T. Hengl, H. I. Reuter, eds. Geomorphometry: Concepts,
Software, Applications. Developments in Soil Science, 33.
Amsterdam, Elsevier.
Guisan, A., and N. E. Zimmermann. 2000. Predictive habitat
distribution models in ecology. Ecol. Model. 135, 147–186.
Guisan, A., A. Lehmann, S. Ferrier, M. Austin, J. M. C.
Overton, R. Aspinall, et al. 2006. Making better
biogeographical predictions of species’ distributions. J. Appl.
Ecol. 43, 386–392.
Guisan, A., R. Tingley, J. B. Baumgartner, I. Naujokaitis-Lewis,
P. R. Sutcliffe, A. I. T. Tulloch, et al. 2013. Predicting
species distributions for conservation decisions. Ecol. Lett.
16, 1424–1435.
Harrison, J. W., F. P. Ferrie, S. W. Hefford, C. Samson, K.
Kusevic, P. Mrstik, et al. 2009. Finding anomalies in high-
density LiDAR point clouds. Geomatica 63, 397–405.
Heuvelink, G. B. M. 1998. P. 127 Error propagation in
environmental modelling with GIS. Taylor and Francis,
London.
Holmes, K. W., K. P. van Niel, B. Radford, G. A. Kendrick,
and S. L. Grove. 2008. Modelling distribution of marine
benthos from hydroacoustics and underwater video. Cont.
Shelf Res. 28, 1800–1810.
Hughes-Clarke, J. E. 1997. Are you really getting “full bottom
coverage”? Available at: http://www.omg.unb.ca/~jhc/covera
ge_paper.html (accessed 23 August 2016).
Hughes-Clarke, J. E. 2002.The challenge of technology:
improving sea-floor mapping methodologies. Available at:
http://www.omg.unb.ca/AAAS/UNB_Seafloor_Mapping.html
Hughes-Clarke, J. E 2003a. Pp. 1–12. A reassessment of vessel
coordinate systems: what is it that we are really aligning? US
Hydrographic Conference 2003. The Hydrographic Society of
America, Rockville, Maryland.
Hughes-Clarke, J. E 2003b. Dynamic motion residuals in
swath sonar data: ironing out the creases. Int. Hydrogr. Rev.
4,6–23.
Hughes-Clarke, J. E, L. A. Mayer, and D. E. Wells. 1996.
Shallow-water imaging multibeam sonars: a new tool for
investigating seafloor processes in the coastal zone and
on the continental shelf. Mar. Geophys. Res. 18, 607–
629.
Kenny, A. J., I. Cato, M. Desprez, G. Fader, R. T. E.
Sch€
uttenhelm, and J. Side. 2003. An overview of seabed-
mapping technologies in the context of marine habitat
classification. ICES J. Mar. Sci. 60, 411–418.
Kostylev, V. E., B. J. Todd, G. B. J. Fader, R. C. Courtney, G.
D. M. Cameron, and R. A. Pickrill. 2001. Benthic habitat
mapping on the Scotian Shelf based on multibeam
bathymetry, surficial geology and sea floor photographs.
Mar. Ecol. Prog. Ser. 219, 121–137.
Lamarche, G., X. Lurton, A.-L. Verider, and J.-M. Augustin.
2011. Quantitative characterisation of seafloor substrate and
bedforms using advanced processing of multibeam
backscatter –application to Cook Strait, New Zealand. Cont.
Shelf Res. 31, S93–S109.
Lecours, V. 2015. Terrain Attribute Selection for Spatial
Ecology (TASSE), v.1.0, https://doi.org/10.13140/rg.2.2.
35147.18728.
Lecours, V., and Devillers R. 2015. Assessing the spatial data
quality paradox in the deep-sea. Pp. 1–8. In R. E. Sieber, ed.
Proceedings of spatial knowledge and information –Canada
2015. SKI-Canada, Banff, Alberta.
Lecours, V., R. Devillers, D. C. Schneider, V. L. Lucieer, C. J.
Brown, and E. N. Edinger. 2015. Spatial scale and
geographic context in benthic habitat mapping: review and
future directions. Mar. Ecol. Prog. Ser. 535, 259–284.
Lecours, V., M. F. J. Dolan, A. Micallef, and V. L. Lucieer.
2016a. A review of marine geomorphometry, the
quantitative study of the seafloor. Hydrol. Earth System Sci.
20, 3207–3244.
Lecours, V., C. J. Brown, R. Devillers, V. L. Lucieer, and E. N.
Edinger. 2016b. Comparing selection of environmental
variables for ecological studies: a focus on terrain attributes.
PLoS ONE 11, e0167128.
Lecours, V., R. Devillers, V. L. Lucieer, and Brown, C. J..
(2017a). Artefacts in marine digital terrain models: a
multiscale analysis of their impact on the derivation of
terrain attributes. DOI: 10.1109/TGRS.2017.2707303
Lecours, V., R. Devillers, A. E. Simms, V. L. Lucieer, and C. J.
Brown. 2017b. Towards a framework for terrain attribute
selection in environmental studies. Environ. Model. Softw.
89,19–30.
Leempoel, K., C. Parisod, C. Geiser, L. Dapra, P. Vittoz, and
S. Joost. 2015. Very high-resolution digital elevation models:
are multi-scale derived variables ecologically relevant?
Methods Ecol. Evol. 6, 1373–1383.
Li, F. K., and R. M. Goldstein. 1990. Studies of multibaseline
spaceborne interferometric synthetic aperture radars. IEEE
Trans. Geosci. Remote Sens. 28,88–97.
Li, D., J. Zhang, and H. Wu. 2012. Spatial data quality and
beyond. Int. J. Geogr. Inf. Sci. 26, 2277–2290.
Lichti, D. D., and J. Skaloud. 2010. Registration and
calibration. Pp. 83–133. In G. Vosselman and H.-G. Maas,
eds, Airborne and terrestrial laser scanning. CRC Press Inc,
Boca Raton, Florida.
Livne, E., and T. Svoray. 2011. Components of uncertainty
in primary production model: the study of DEM,
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 13
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
classification and location error. Int. J. Geogr. Inform. Sci.
25, 473–488.
Lucieer, V., N. Barrett, N. Hill, and S. L. Nichol. 2012.
Characterization of shallow inshore coastal reefs on the
Tasman Peninsula, Southeastern Tasmania, Australia.
Pp. 481–492. in P. T. Harris and E. K. Baker, eds. Seafloor
geomorphology as benthic habitat. Elsevier, Amsterdam.
Lurton, X. 2010. P 704. An introduction to underwater
acoustics: principles and applications, 2nd ed Springer/
Praxis Publishing, Berlin, Germany
Lurton, X., and Lamarche G.. 2015. Backscatter measurements
by seafloor-mapping sonars, guidelines and
recommendations. GeoHAB, 200 p.
Micallef, A., T. P. Le Bas, V. A. I. Huvenne, P. Blondel,
V. H€
uhnerback, and A. Deidun. 2012. A multi-method
approach for benthic habitat mapping of shallow coastal
areas with high-resolution multibeam data. Cont. Shelf Res.
39–40,14–26.
Miller, J. 2010. Species distribution modeling. Geogr. Compass
4, 490–509.
Monk, J., D. Ierodiaconou, V. L. Versace, A. Bellgrove,
E. Harvey, A. Rattray, et al. 2010. Habitat suitability for
marine fishes using presence-only modelling and multibeam
sonar. Mar. Ecol. Prog. Ser. 420, 157–174.
Moudry, V., and P.
S
ımov
a. 2012. Influence of positional
accuracy, sample size and scale on modelling species
distributions: a review. Int. J. Geogr. Inform. Sci. 26,
2083–2095.
NHS. 2013. (Norwegian Mapping Authority, Hydrographic
Service) Specifications for seabed mapping within the
MAREANO program. 17 p.
van Niel, K. P., and M. P. Austin. 2007. Predictive
vegetation modeling for conservation: impact of error
propagation from digital elevation data. Ecol. Appl. 17,
266–280.
van Niel, K. P., S. W. Laffan, and B. G. Lee. 2004. Effect of
error in the DEM on environmental variables for predictive
vegetation modelling. J. Veg. Sci. 15, 747–756.
Passalacqua, P., P. Belmont, D. M. Staley, J. D. Simley, J. R.
Arrowsmith, C. A. Bode, et al. 2015. Analyzing high
resolution topography for advancing the understanding of
mass and energy transfer through landscape: a review. Earth
Sci. Rev. 148, 174–193.
Peters, J., N. E. C. Verhoest, R. Samson, M. van Meirvenne,
L. Cockx, and B. De Baets. 2009. Uncertainty propagation
in vegetation distribution models based on ensemble
classifiers. Ecol. Model. 220, 791–804.
Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006.
Maximum entropy modeling of species distributions. Ecol.
Model. 190, 231–259.
Rattray, A., D. Ierodiaconou, L. Laurenson, S. Burq, and M.
Reston. 2009. Hydro-acoustic remote sensing of benthic
biological communities on the shallow South East
Australian continental shelf. Estuar. Coast. Shelf Sci. 84,
237–245.
Regan, H. M., Y. Ben-Haim, B. Langford, W. G. Wilson,
P. Lundberg, S. J. Andelman, et al. 2005. Robust decision-
making under severe uncertainty for conservation
management. Ecol. Appl. 15, 1471–1477.
Reuter, H. I., Hengl T., Gessler P., and P. Soille. 2009.
Preparation of DEMs for geomorphometric analysis.
Pp. 87–120. In T. Hengl, H. I. Reuter, eds.
Geomorphometry: Concepts, Software, Applications.
Developments in Soil Science, vol. 33. Elsevier, Amsterdam.
Roberts, J. M., A. J. Davies, L. A. Henry, L. A. Dodds, G. C. A.
Duineveld, M. S. S. Lavaleye, et al. 2009. Mingulay reef
complex: an interdisciplinary study of cold-water coral
habitat, hydrography and biodiversity. Mar. Ecol. Prog. Ser.
397, 139–151.
Robinson, L. M., J. Elith, A. J. Hobday, R. G. Pearson, B. E.
Kendall, J. P. Possingham, et al. 2011. Pushing the limits in
marine species distribution modelling: lessons from the land
present challenges and opportunities. Glob. Ecol. Biogeogr.
20, 789–802.
Roman, C., and H. Singh. 2006. Consistency based error
evaluation for deep sea bathymetric mapping with robotic
vehicles. Proceedings of the 2006 IEEE International
Conference on Robotics and Automation, pp. 3568-3574.
Rousseaux, F. 2003.
Etude de l’impact de la repr
esentation du
relief sur les applications. Revue Internationale de
G
eomatique 13, 493–509.
Smith, S. J., J. A. Sameoto, and C. J. Brown. 2017. Setting
biological reference points for sea scallops (Placopecten
magellinacus) allowing for the spatial distribution of
productivity and fishing effort. Can. J. Fish Aquat. Sci..
74:650–667. https://doi.org/10.1139/cjfas-2015-0595
Sofia, G., F. Pirotti, and P. Tarolli. 2013. Variations in
multiscale curvature distribution and signatures of
LiDAR DTM errors. Earth Surf. Proc. Land. 38, 1116–
1134.
Solan, M., J. D. Germano, D. C. Rhoads, C. Smith,
E. Michaud, D. Parry, et al. 2003. Towards a greater
understanding of pattern, scale and process in marine
benthic systems: a picture is worth a thousand worms.
J. Exp. Mar. Biol. Ecol. 285–286, 313–338.
Todd, B. J., V. E. Kostylev, and S. J. Smith. 2012. Seabed
habitat of a glaciated shelf, German Bank, Atlantic Canada.
Pp. 555–568. in P. T. Harris and E. K. Baker, eds. Seafloor
geomorphology as benthic habitat. Elsevier, Amsterdam.
Vaughan, I. P., and S. J. Ormerod. 2005. The continuing
challenges of testing species distribution models. J. Appl.
Ecol. 42, 720–730.
Warren, D. L., and S. N. Seifert. 2011. Ecological niche
modeling in Maxent: the importance of model complexity
and the performance of model selection criteria. Ecol. Appl.
21, 335–342.
14 ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London
Impacts of Artefacts in Habitat Mapping V. Lecours et al.
Williams, K. J., L. Belbin, M. P. Austin, J. L. Stein, and S. Ferrier.
2012. Which environmental variables should I use in my
biodiversity model? Int. J. Geogr. Inform. Sci. 26, 2009–2047.
Wilson, J. P. 2012. Digital terrain modeling. Geomorphology
137, 107–121.
Wise, S. 2000. Assessing the quality for hydrological
applications of digital elevation models derived from
contours. Hydrol. Process. 14, 1909–1929.
Zajac, R. N. 2008. Challenges in marine, soft-sediment
benthoscape ecology. Landscape Ecol. 23,7–18.
Zieger, S., T. Stieglitz, and S. Kininmonth. 2009. Mapping reef
features from multibeam sonar data using multiscale
morphometric analysis. Mar. Geol. 264, 209–217.
Supporting Information
Additional supporting information may be found online
in the supporting information tab for this article.
Table S1. Mean and standard deviation of kappa coeffi-
cients of agreement of the 10 maps made from altered data
for each type of artefact, each scenario and each scale.
Figure S1. Spatial variation in predictions of sea scallops
distribution as quantified by the range in correlation coeffi-
cients between models built from altered data and the refer-
ence models.
ª2017 The Authors. Remote Sensing in Ecology and Conservation published by John Wiley & Sons Ltd on behalf of Zoological Society of London 15
V. Lecours et al. Impacts of Artefacts in Habitat Mapping
