ArticlePDF Available

Abstract and Figures

Selecting appropriate environmental variables is a key step in ecology. Terrain attributes (e.g. slope, rugosity) are routinely used as abiotic surrogates of species distribution and to produce habitat maps that can be used in decision-making for conservation or management. Selecting appropriate terrain attributes for ecological studies may be a challenging process that can lead users to select a subjective, potentially sub-optimal combination of attributes for their applications. The objective of this paper is to assess the impacts of subjectively selecting terrain attributes for ecological applications by comparing the performance of different combinations of terrain attributes in the production of habitat maps and species distribution models. Seven different selections of terrain attributes, alone or in combination with other environmental variables, were used to map benthic habitats of German Bank (off Nova Scotia, Canada). 29 maps of potential habitats based on unsupervised classifications of biophysical characteristics of German Bank were produced, and 29 species distribution models of sea scallops were generated using MaxEnt. The performances of the 58 maps were quantified and compared to evaluate the effectiveness of the various combinations of environmental variables. One of the combinations of terrain attributes–recommended in a related study and that includes a measure of relative position, slope, two measures of orientation , topographic mean and a measure of rugosity–yielded better results than the other selections for both methodologies, confirming that they together best describe terrain properties. Important differences in performance (up to 47% in accuracy measurement) and spatial outputs (up to 58% in spatial distribution of habitats) highlighted the importance of carefully selecting variables for ecological applications. This paper demonstrates that making a subjective choice of variables may reduce map accuracy and produce maps that do not adequately represent habitats and species distributions, thus having important implications when these maps are used for decision-making.
Content may be subject to copyright.
Comparing Selections of Environmental
Variables for Ecological Studies: A Focus on
Terrain Attributes
Vincent Lecours
*, Craig J. Brown
, Rodolphe Devillers
, Vanessa L. Lucieer
, Evan
N. Edinger
1Department of Geography, Memorial University of Newfoundland, 232 Elizabeth Avenue, St. John’s,
Newfoundland and Labrador, Canada, 2Applied Research, Nova Scotia Community College, 80 Mawiomi
Place, Dartmouth, Nova Scotia, Canada, 3Institute for Marine and Antarctic Studies, University of Tasmania,
20 Castray Esplanade, Battery Point, Tasmania, Australia, 4Department of Biology, Memorial University of
Newfoundland, 232 Elizabeth Avenue, St. John’s, Newfoundland and Labrador, Canada
Selecting appropriate environmental variables is a key step in ecology. Terrain attributes
(e.g. slope, rugosity) are routinely used as abiotic surrogates of species distribution and to
produce habitat maps that can be used in decision-making for conservation or management.
Selecting appropriate terrain attributes for ecological studies may be a challenging process
that can lead users to select a subjective, potentially sub-optimal combination of attributes
for their applications. The objective of this paper is to assess the impacts of subjectively
selecting terrain attributes for ecological applications by comparing the performance of dif-
ferent combinations of terrain attributes in the production of habitat maps and species distri-
bution models. Seven different selections of terrain attributes, alone or in combination with
other environmental variables, were used to map benthic habitats of German Bank (off
Nova Scotia, Canada). 29 maps of potential habitats based on unsupervised classifications
of biophysical characteristics of German Bank were produced, and 29 species distribution
models of sea scallops were generated using MaxEnt. The performances of the 58 maps
were quantified and compared to evaluate the effectiveness of the various combinations of
environmental variables. One of the combinations of terrain attributes–recommended in a
related study and that includes a measure of relative position, slope, two measures of orien-
tation, topographic mean and a measure of rugosity–yielded better results than the other
selections for both methodologies, confirming that they together best describe terrain prop-
erties. Important differences in performance (up to 47% in accuracy measurement) and spa-
tial outputs (up to 58% in spatial distribution of habitats) highlighted the importance of
carefully selecting variables for ecological applications. This paper demonstrates that mak-
ing a subjective choice of variables may reduce map accuracy and produce maps that do
not adequately represent habitats and species distributions, thus having important implica-
tions when these maps are used for decision-making.
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 1 / 18
Citation: Lecours V, Brown CJ, Devillers R, Lucieer
VL, Edinger EN (2016) Comparing Selections of
Environmental Variables for Ecological Studies: A
Focus on Terrain Attributes. PLoS ONE 11(12):
e0167128. doi:10.1371/journal.pone.0167128
Editor: Stefan Lo¨tters, Universitat Trier, GERMANY
Received: June 22, 2016
Accepted: November 9, 2016
Published: December 21, 2016
Copyright: ©2016 Lecours et al. This is an open
access article distributed under the terms of the
Creative Commons Attribution License, which
permits unrestricted use, distribution, and
reproduction in any medium, provided the original
author and source are credited.
Data Availability Statement: Data used in this
study are third party data belonging to the
Department of Fisheries and Oceans, Canada. 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 The authors confirm
that they requested these data in this manner and
received no special access privileges others would
not have.
Funding: VL received funding from the Natural
Sciences and Engineering Research Council of
Canada. The funders had no role in study design,
Due to the difficulty in sampling ecological data at sufficient spatial and temporal resolutions,
many ecological studies rely on the use of surrogates to understand species distribution and
ecological processes. Amongst commonly used surrogates, terrain attributes (e.g. slope, rugos-
ity, aspect) derived from digital elevation (DEM) or bathymetric (DBM) models have proven
their value in a broad range of terrestrial and marine ecological studies [1]. Such attributes can
now be derived easily using tools available in most Geographic Information Systems (GIS).
While tools are increasingly user-friendly, a lack of transparency in most software on the actual
algorithms used [2] can prevent users from making an informed decision on which tools to
use. Also, terrain attributes sharing the same name but generated using different algorithms
(e.g. slope) have been shown to produce different derivative surfaces [2,3]. Software developers
and authors of published work are often not explicit on the methods they use to derive terrain
attributes (e.g. algorithm or tool). This lack of information can possibly influence the analysis
and interpretation of the resulting terrain attribute surfaces, and consequently the ecological
application for which they are being used.
Choosing an appropriate selection of terrain attributes for specific ecological applications
can be challenging, and users will often simply use the terrain attributes made available by the
software they have access to or are familiar with, without further questioning if those attributes
are the most appropriate ones for their study. In a related study, Lecours et al. [4] showed that
many terrain attributes covary, which may cause potential problems for many statistical analy-
ses. In a seabed classification context, Diesing et al. [5] recommended integrating the reduc-
tion of covariation within practices. In an attempt to identify an optimal combination of
terrain attributes to use in ecology that would reduce covariation while extracting as much
information as possible on the terrain, Lecours et al. [4] recommended using a combination of
six easily computable terrain attributes for ecological studies that consider topography or
bathymetry: (1) relative deviation from mean value, which is a measure of relative position
that can identify local peaks and valleys, (2) standard deviation, which is a measure of rugosity,
(3) local mean, (4) slope, and (5–6) easterness and northerness, which together provide infor-
mation on the orientation of the slope (i.e. aspect).
This article aims to describe the effects of subjectively selecting input variables for ecological
applications, with a particular focus on terrain attributes. The specific objectives are (1) to
compare the performance of Lecours et al. [4] recommended selection of terrain attributes to
other selections in a real ecological context, (2) to demonstrate the relative importance of ter-
rain morphology in aiding our understanding of ecological questions compared to other envi-
ronmental variables, and (3) to report on the consequences of selecting different input
variables on both the accuracy of habitat maps and the spatial distribution of the outputs.
Materials and Methods
Benthic habitat mapping is the act of mapping significantly distinct areas of the seafloor based
on their physical, chemical and biological characteristics at particular spatial and temporal
scales [6]. The marine environment presents particular challenges in observing and sampling
seafloor characteristics. However, developments in acoustic remote sensing technologies, spe-
cifically multibeam echosounders (MBES), now allow the collection of high-resolution
remotely sensed data of the seafloor. Bathymetric measurements from MBES can be used to
generate DBMs, from which terrain attributes can be derived [7]. Additionally, MBES systems
can also record acoustic reflectance (backscatter) data that provide information on seafloor
properties (e.g. surficial geology, porosity). In combination, these attributes are commonly
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 2 / 18
data collection and analysis, decision to publish, or
preparation of the manuscript.
Competing Interests: The authors have declared
that no competing interests exist.
used for the production of benthic habitat maps. For the purpose of this study, two common
approaches to habitat mapping were used: unsupervised and supervised classifications.
Datasets from Brown et al. [8], covering 3,650 km
of German Bank, an area of the Canadian
continental shelf off Nova Scotia (Fig 1), were used to address the objectives of this study.
These data comprised a 50 m resolution DBM, 3,190 geo-referenced underwater images of the
seabed visually classified into five bottom types (glacial till, silt and mud, rippled silt, rippled
sand, reef), 4,816 geo-referenced sea scallop observations, and three backscatter data deriva-
tives (Q1, Q2, Q3; Fig 1). Details on how the data were collected and processed can be found
in Brown et al. [8]. For comparison with surfaces used in Lecours et al. [4], the fractal dimen-
sion, which is a quantitative representation of surface complexity, was measured over 10,000
areas of German Bank. Values ranged from 2.09 to 2.93, thus including regions of low
(towards 2.00), moderate and high complexities (towards 3.00).
A total of 24 different terrain attributes were derived from the DBM and grouped into
seven selections of six terrain attributes each (Table 1). The terrain attributes were selected
from groups of variables that exhibited various behaviours during the statistical analyses per-
formed in Lecours et al. [4] (see caption of Table 1 and S1 Appendix for more details). Selec-
tion 1 corresponds to our recommended selection of six terrain attributes. These terrain
attributes were computed using the TASSE (Terrain Attribute Selection for Spatial Ecology)
toolbox for ArcGIS [9]. Selections 2 to 7 were built to maximize variability and resemblance to
Selection 1 ( avoid having two measures of slope or curvatures within one selection). Par-
ticular focus was also given to terrain attributes that were identified as potentially important
by Lecours et al. [4].
Unsupervised Classifications of Potential Habitat Types. Biophysical classifications of
the area were performed to create benthoscape maps [10], which are produced by following a
landscape style approach like when landscape features are delineated from terrestrial datasets.
This top-down, unsupervised approach to habitat mapping is often used to map features that
can only be resolved within the remotely sensed data, without attempting to delineate features
beyond what the remote sensing techniques are capable of resolving. A total of 29 benthoscape
maps were built using the Modified k-Means unsupervised classification tool in
Whitebox GAT v.3.2 “Iguazu”. Algorithms such as k-means are commonly used in both terres-
trial and marine ecological applications [5], but this particular algorithm is different from the
regular k-means ones as it does not require a subjective input from the user to define the num-
ber of classes. The algorithm first segments the MBES derived data layers into a liberal, overes-
timated number of units, and then iteratively merges classes based on a pre-defined distance
threshold between their cluster centres, eventually reaching an optimal, objective number of
units. These units are then compared and subsequently recombined based on best match
against independently classified in situ photographic data, classified into broad biophysical
benthoscape classes. Using this approach, biophysical features can be delineated at a broader
scale over the study area to generate a benthoscape map.
To assess the relative importance of the different environmental variables and the conse-
quences of using different input variables in habitat mapping, four scenarios were tested with
each of the seven selections, resulting in 28 habitat maps. Maps were first created using each
selection alone (six input layers), then adding the bathymetry (seven layers), the three back-
scatter derivatives (nine layers), and finally both the bathymetry and the backscatter derivatives
(ten layers). In order to quantify the relative influence of terrain morphology in potential habi-
tat characterization of German Bank, an additional habitat map was produced using only the
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 3 / 18
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 4 / 18
backscatter derivatives and the bathymetry (four layers, not accounting for terrain
Following the method outlined in Brown et al. [8], the resulting clusters for each classifica-
tion were spatially compared to the 3,190 photographs of the seabed. Clusters corresponding
to the same habitat types were grouped together and mapped as the corresponding habitat
types. Confusion matrices, summarizing agreement and disagreement between the ground-
truth data and the results from the classified bottom types [11], were built to compute the over-
all accuracy and kappa coefficient of agreement of each habitat map. The two measures are
commonly used in ecology [12] and in remote sensing [11,13]. The success of the discrimina-
tion of each individual bottom type by the 29 classifications was assessed using the producer’s
accuracy [11], and a spatial comparison of the outputs was made to assess the amplitude of
change caused by selecting different variables. This was quantified using the percentage of pix-
els that were classified as the same bottom type by different classifications.
Fig 1. German Bank study area with some of the input variables used in this study: the ground-truth data for the bottom
types, the sea scallops observations, the bathymetry, the three backscatter derivatives and the six terrain attributes from
Selection 1.
Table 1. Selections of terrain attributes used to build the habitat maps and models. The ID numbers refer to Lecours et al. [4] and allow finding the soft-
ware and parameters with which the attributes were generated (see also S1 Appendix). Marker variables correspondto important variables; whether they
were found on strong components (Sel. 1) or weak components (Sel. 4) is linked to the amount of topographic structure they accounted for. Variables with low
cardinality (Sel. 2) did not have many different values, thus limiting their ability to explain slight variations in terrain morphology. Complex variables (Sel. 3) cor-
respond to redundant variables. The terrain attributes identified by an asterisk were previously identified as potentially important [4]. The underlined attributes
were recommended in [4].
Selection 1 Selection 2 Selection 3 Selection 4
Marker Variables on Strong Components Variables with Low Cardinality Complex Variables Marker Variables on Weak
ID31 Easterness ID1 Bathymetric Position Index ID70 Mean of Residuals*ID132 Plan Curvature
ID67 Local Mean ID2 Center vs Neighbor
ID116 Plan Curvature ID153 Profile Curvature
ID90 Northerness ID42 Easterness*ID136 Profile Curvature ID158 Representativeness*
ID157 Relative Deviation from Mean
ID101 Northerness*ID178 Slope ID188 Slope Variability
ID166 Slope ID111 Percentile ID201 Surface Roughness ID219 Total Curvature
ID190 Standard Deviation ID143 Profile Curvature ID221 Value Range ID227 Vector Ruggedness
Selection 5 Selection 6 Selection 7
Mix of Selections 1 and 2 Mix of Selections 1 and 3 Mix of Selections 1 and 4
ID1 Bathymetric Position Index ID70 Mean of Residuals*ID158 Representativeness*
ID2 Center vs. Neighbor variability*ID178 Slope ID188 Slope Variability
ID42 Easterness*ID221 Value Range ID227 Vector Ruggedness
ID67 Local Mean ID31 Easterness ID67 Local Mean
ID90 Northerness ID90 Northerness ID31 Easterness
ID166 Slope ID190 Standard Deviation ID90 Northerness
Scenario A: Each Selection used Alone (6 layers)
Scenario B: Each Selection used with Depth (7 layers)
Scenario C: Each Selection used with the Three Backscatter Derivatives (9 layers)
Selection D: Each Selection used with Depth and the Three Backscatter Derivatives (10 layers)
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 5 / 18
Supervised Classifications of Sea Scallop Habitats. In addition to the unsupervised classi-
fications, a bottom-up supervised approach to habitat mapping was used in which the in situ data
were used to segment the environmental data to predict sea scallop (Placopecten magellanicus)
habitat on German Bank. Maximum entropy (MaxEnt) [14,15], a presence-background method,
was used to perform these supervised classifications of scallops habitat. We recognize that there
are benefits and drawbacks associated with all modelling techniques and that there is still a debate
surrounding which one works best [8]. For the purpose of this study, a technique that could be
kept consistent across the methodology was required in order to enable comparisons of out-
comes. While any technique could have been used, MaxEnt was selected because it was shown to
perform better than other species distribution models (SDM) in both terrestrial [16] and marine
realms [17]. Following the method of Brown et al. [8], the classifier was run in the MaxEnt soft-
ware v.3.3.3k with the default settings, except that the number of background points was
increased to 50,000 to account for background conditions in full measure in such a large area.
The 3,813 scallop observations selected by Brown et al. [8] were used to train the model, while the
remaining 1,003 observations were kept for validation. A total of 29 MaxEnt models were run:
for each of the seven selections, four models were run according to the scenarios previously men-
tioned resulting in 28 models, and one model was run without terrain attributes.
The MaxEnt software was also used to perform jackknife tests and to calculate the area
under the curve (AUC) derived from threshold independent receiver operating characteristic
(ROC) curves; the former quantify the percentage contribution of each input variable to the
models while the latter serves to assess the performance of SDMs [16]. We acknowledge that
there is currently a debate in the literature surrounding the use of AUC as a measure of model
evaluation [18]; some authors argue that AUC can be inappropriate when different modeling
techniques are used [19] or if two different species or areas are compared [20]. However, AUC
often performs better than other measures [21,22] and is appropriate when the species, study
area, and the training and test samples are the same across the compared models [23,24], like
in the current study.
Model outputs were evaluated in terms of their statistical fit to the validation data (AUC
[25]. A 95% confidence interval based on the standard deviate (1.96 standard deviations of the
value) was used to identify the significant differences in performances [26]. The good-
ness-of-fit of the models to the training data (AUC
) was used to assess models’ generaliz-
ability (i.e. transportability, transferability). Generalizability is described by Vaughan &
Ormerod (p.720 [22]) as “a basic requirement for predictive models” that describes the ability
of a model to produce accurate predictions with data other than the training dataset. Gener-
alizability was measured using the difference (AUC
) between AUC
and AUC
A model that over-fits the training data will have a high AUC
but a low AUC
as it per-
forms poorly on the test dataset, thus resulting in a high AUC
. Such a model is too specific
to the training data and less generalizable. A diagnostic of the input variables contribution to
the different models was also performed based on the results from the jackknife procedure, in
order to identify the loss or gain in explanatory power as each variable is removed from the
models or used alone [28]. Finally, a spatial comparison of the models was performed to evalu-
ate the consequences of variable selection on the model outputs.
Unsupervised Classifications
Performance of Classifications. The overall accuracies and kappa coefficients of the 29
habitat maps are presented in Fig 2. Selection 1 (i.e. the proposed attribute selection) outper-
formed the others with the highest overall accuracy and kappa coefficient in three of the four
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 6 / 18
Fig 2. Map accuracies measured with (A) a kappa coefficient of agreement and (B) the overall accuracy.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 7 / 18
scenarios. The highest kappa coefficient was obtained when combining Selection 1 with
bathymetry and the backscatter derivatives. The highest overall accuracy, 68.3%, was reached
when combining Selection 5 with bathymetry (Fig 2B). Selection 1 combined with bathymetry
had the second highest overall accuracy (67.1%). Selections with only three attributes from
Selection 1 (i.e. Selections 5, 6 and 7) usually outperformed their related selection with none of
the proposed attribute (i.e. Selections 2, 3 and 4). Selection 4 resulted in poor classifications, and
Selections 2, 3 and 6 performed generally poorly except when bathymetry was added. Com-
pared to the classification that only used bathymetry and the backscatter derivatives (i.e. no
topography), eight classifications had a higher overall accuracy: the four classifications that used
Selection 1 as input, Selections 5 and 6 combined with bathymetry, and Selections 5 and 6 com-
bined with both bathymetry and the backscatter derivatives. In terms of kappa coefficients, only
four classifications performed better than the one with no topography: Selection 1 with the
backscatter derivatives, Selection 1 with both bathymetry and the backscatter derivatives, Selec-
tion 5 with bathymetry, and Selection 6 with both bathymetry and the backscatter derivatives.
Differences up to 45.5% were observed between the overall accuracy values and the kappa
coefficients for a same selection and scenario. Differences were substantial with an average of
28.5% and a standard deviation of 11.9%. The average difference between the two measures of
accuracy for the four maps using Selection 1 was the lowest, followed by the average difference
for the four maps of Selections 5, 7, 6, 2, 3 and 4.
Discrimination of Benthoscape Classes. Selection 1 performed on average better than the
others when discriminating between the five bottom types (Fig 3). When looking at the individ-
ual habitat types, 25 of the 28 other classifications discriminated glacial till better than the classi-
fication with only bathymetry and the backscatter derivatives (producer’s accuracy of 77.1%),
indicating that terrain morphology is not a good surrogate of the presence of glacial till. The
“silt and mud” class seemed driven primarily by bathymetry and sediment properties (i.e. back-
scatter derivatives), with a producer’s accuracy of 87.4% for the classification that did not
account for terrain morphology. Only two of the 28 remaining classifications discriminated that
habitat type better, although several other classifications were very close to achieving that accu-
racy. Reefs were generally poorly discriminated. The classification with no topography reached
a producer’s accuracy of 19.8%, and only six of the remaining classifications performed better,
including three of the classifications using Selection 1. Rippled silt seemed to be better explained
by the bathymetry and the backscatter derivatives, with the corresponding classification reach-
ing an accuracy of 57.6%. Only four other classifications did better, including two classifications
that included Selection 1. Finally, rippled sand was very poorly discriminated by all the classifi-
cations, which may be due to its small sample size (only 49 photographs).
In terms of mean producer’s accuracy for the five habitat types, only three classifications
did better than the one with no topography (48.4%): Selection 1 with the backscatter deriva-
tives (48.5%), Selection 1 with bathymetry and the backscatter derivatives (51.6%), and Selec-
tion 6 with bathymetry and the backscatter derivatives (51.3%). When averaging the mean
producer’s accuracies from the four scenarios for each selection, Selection 1 ranked first, fol-
lowed by Selections 5, 7, 6, 3, 2 and 4.
Spatial Variations of Outputs from Different Selections. The most accurate map
according to the kappa coefficients of agreement was made from Selection 1 combined with
bathymetry and the backscatter derivatives. The spatial similarity indices of that map with the
other habitat maps built with ten layers are presented in Table 2. Compared to Selection 1,
Selection 6 produced the most spatially similar map with 90.0% similarity. Selection 4 is the
least similar with only about 41.7% identically classified pixels. The other maps were between
73.3% and 79.4% similar to the map with Selection 1, except for the map with no topography
(i.e. only bathymetry and the backscatter derivatives) with 82%.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 8 / 18
Supervised Classifications
Predictive Capacity and Robustness. Fig 4 shows the performance of the 29 MaxEnt
models. All models performed significantly better than random (i.e. AUC
±95% confi-
dence interval >0.500). Models with higher AUC
and lower standard deviations are more
robust and present the highest predictive capacity [29]. In general, adding bathymetry, the
Table 2. Spatial similarity of the habitat maps and SDMs generated from Selections 2 to 7, compared to the map and modelbuilt from Selection 1.
A similarity of 90% indicates that 90% of the pixels were classified as the same habitat type in the two compared maps, or that 90% of the pixels were within
±5% of probability distribution in the two compared models.
Spatial Similarity with Selection 1 (%)
Scenario with 10 Layers
Unsupervised Classifications Supervised Classifications (within ±5% probability)
Selection 2 73.3 72.9
Selection 3 77.0 64.6
Selection 4 41.7 65.3
Selection 5 79.4 81.4
Selection 6 90.0 71.5
Selection 7 78.4 70.9
No topography 82.1 66.9
Fig 3. Comparison of the discrimination ability of the computed classifications with that of the classification computed using only
bathymetry and the backscatter derivatives, based on the number of bottom types (maximum possible of 5) that were better
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 9 / 18
backscatter derivatives, or all of them to the terrain attributes improved the models predictive
capacity. However, Selection 1 and other selections that include terrain attributes from Selec-
tion 1 did not always follow that trend. For instance, Selection 1 used alone (only six terrain
attributes; black diamond in Fig 4) performed better than other selections combined with
bathymetry or the backscatter derivatives (e.g. blue and green squares and triangles in Fig 4).
Selection 1 combined with the backscatter derivatives (black triangles in Fig 4) performed bet-
ter than other selections that were combined with both bathymetry and the backscatter deriva-
tives (i.e. most circles in Fig 4).
In the scenario where only terrain attributes are used (diamonds in Fig 4), Selection 1 per-
formed the best, followed by the three selections that include three terrain attributes from
Selection 1 (Selections 5, 6, and 7). The same pattern was observed when combining the selec-
tions with the three backscatter derivatives (triangles in Fig 4). A different pattern arose when
adding bathymetry to the selections, one in which Selection 1 performed second best behind
Selection 6. However, the 95% confidence intervals measured around the AUC values show
that the difference in performances between Selection 1 and 6 are not significant for the two
scenarios where Selection 6 performed better than Selection 1.
Generalizability. Fig 5 shows the generalizability of the 29 SDMs. Models with higher
fitted better the training data while models with lower AUC
predicted more
Fig 4. Performance and robustness of the 29 MaxEnt models. Models in the top-left corner of the graph performed better and are more robust.
Colour legend: Selection 1 (black), Selection 2 (blue), Selection 3 (red), Selection 4 (green), Selection 5 (purple), Selection 6 (orange), Selection 7
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 10 / 18
efficiently the validation data. Models with high AUC
and low AUC
are therefore the
most generalizable, as they do not over-fit the training data [29]. Fig 5 shows that the models
that included bathymetry (scenarios with seven and ten layers; squares and circles in Fig 5) are
more similar than the other models, especially for the models that combined ten input layers.
Models that used only terrain attributes or combined them with backscatter derivatives
(diamonds or triangles in Fig 5) showed similar patterns, where the best models in terms of
also had a higher AUC
, an indication that the best models were also the ones that
over-fitted the data the most. In those two scenarios, Selection 1 clearly stands out as a good
trade-off between predictive ability and over-fitting of data, making it the most likely to be
generalizable and to perform well. When considering bathymetry (squares and circles in Fig
5), a similar pattern emerged whether or not the backscatter derivatives were added: Selections
1, 3 and 6 stand out as being more generalizable. Selections 7 and 4 have the highest AUC
but also the highest AUC
, therefore having a tendency to over-fit the training data.
Variables Contribution. The percentage of contribution of each variable used as input in
the 29 models can be found in Fig 6. When used, bathymetry and two of the backscatter deriv-
atives (Q1 and Q2) contributed the most to the models, with a respective average of 39.2%,
25.4% and 19.6% for the 15 models that used them. Bathymetry contributed less to the models
that include local mean as input, resulting from the high collinearity between these two
Fig 5. Generalizability of the 29 MaxEnt models. Models closer to the top-left corner are more generalizable as they performed well on the training
data and replicated well to the validation data. See Fig 4 for colour legend.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 11 / 18
variables; when two variables are correlated, MaxEnt is known to assign a more important per-
centage contribution to one of the two and a lower one to the other [28]. Consequently, local
mean is a surrogate of bathymetry and appears as an important variable, with an average con-
tribution of 51.2% for the 12 models that include it. In general, measures of rugosity like stan-
dard deviation and vector ruggedness measure also contributed to the models.
The analysis of changes in model gain based on the jackknife procedure described the
impact on model gain of removing each variable from the models, in addition to provide what
would be the model gain if each variable would be used alone. This analysis provided addi-
tional information on the variables contribution and the performance of models. In MaxEnt, a
variable with a high gain when used alone in a model contributes useful information to the
model [28]. On the other hand, a variable that contributes unique information to a model
makes the gain decrease when it is excluded from the model [28]. In this study, all variables in
all models provided unique information in training the models, except for the four models
that included Selection 2. In terms of transferability of this uniqueness to the training data (i.e.
if the variables still provide unique information when applied to the validation data), Selection
1 performed better than the others in three of the four scenarios. It only failed to outperform
the other selections when ten layers were used, likely due to spatial correlations between local
mean and bathymetry, and slope and local standard deviation. Regarding usefulness, Selection
1 generally did not provide as many useful variables to the models being trained as the other
selections. However, these useful variables were generally also useful for the validation data,
thus transferable, which was not the case for the other selections. For instance, Selection 1 in
Fig 6. Percentage of variable contribution for the 29 MaxEnt models. Only contributions greater than 5% are labeled.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 12 / 18
combination with bathymetry and the backscatter derivatives had six variables providing use-
ful information to the trained model, and these six variables were all useful for the validation
data, an indication of robustness and generalizability. Finally, only two models would not have
reached a higher AUC
if any one of their inputs were removed: Selection 1 combined with
the backscatter derivatives and Selection 1 combined with bathymetry.
Spatial Variations of Predictions from Different Selections. The model computed from
the combination of Selection 1 with bathymetry and the backscatter derivatives showed the
best trade-off between robustness, uniqueness and generalizability. It was therefore used as a
reference to spatially compare the outputs of comparable models, i.e. those computed with ten
layers (Table 2). The most similar model to the reference one, based on a ±5% margin in prob-
ability distribution, was the model computed with Selection 5 (81.4% similar). The lowest simi-
larity was 64.6% (Selection 3). In average, the six other models were 71.1% similar to the one
made from Selection 1. The map produced without terrain morphology had a spatial similarity
index of 66.9% with the map from Selection 1 combined with bathymetry and the backscatter
Selections of Terrain Attributes
Results suggest that Selection 1 of terrain attributes, which corresponds to the combination of
a measure of relative position, a measure of rugosity, two measures of aspect (easterness and
northerness), topographic mean and slope, is more appropriate than the other selections
tested. First, the proposed selection of terrain attributes performed better than the other selec-
tions tested, both in the application of top-down and bottom-up approaches to habitat map-
ping: they generally (1) produced more accurate habitat maps, (2) better discriminated
individual habitat types, (3) produced SDMs with higher AUC values, (4) produced more
robust and generalizable SDMs, (5) provided SDMs with the most variables carrying unique
information, and (6) had the highest number of variables carrying useful information that rep-
licated well to the validation data. Using real data, these results confirm that the six recom-
mended terrain attributes best describe the topographic structure of the terrain by capturing
different and unique characteristics of the terrain. Results also indicate robustness and gener-
alizability of the proposed framework. Many aspects of this study highlighted better perfor-
mances of Selection 1 compared to Selections 2, 3 and 4, thus confirming the limited ability of
these three selections to adequately and fully describe terrain geomorphology.
The findings from this study, utilizing MBES-derived surfaces from German Bank, support
many of the findings presented by Lecours et al. [4] based on terrain attributes generated from
artificial surfaces. Their proposed operational framework was based on two literature-
grounded assumptions: fractal-based surfaces created with spectral synthesis are appropriate
representations of natural surfaces [30,31], and the scale-invariance property of fractals allows
results to be generalized to other spatial scales (i.e. different resolution and/or extent) [32,33].
Artificial surfaces proved their value in ecology [34] and geomorphometry [3]. DTMs of real
terrains are actually geographic “representations” of real terrains, thus in theory no different
than DTMs representing artificial terrains with characteristics found in real terrains. However,
a number of authors argue that fractal-based surfaces should be limited to the development of
null hypotheses [35,36]. The debate is still unsettled; while some claim that “it is heuristically
clear that seafloor or landscape topography is best described by fractal geometry” (p.981 [37]),
others prefer to argue that despite demonstrating fractal-like properties [38], real terrains are
not perfectly fractal [39]. Without necessarily contributing to this debate, the current study
confirmed that results gained from the artificial fractal surfaces in Lecours et al. [4] hold when
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 13 / 18
using a DTM representation of a real terrain (i.e. German Bank) at another spatial scale (i.e. an
extent of 3650 km
represented at 50 m resolution). Consequently, it confirmed the appropri-
ateness of the proposed framework for selecting terrain attributes and its application to any
terrestrial and marine ecological application, regardless of the scale of the environmental data.
Terrain Morphology as an Environmental Factor
Results of both types of classifications indicate that bathymetry and substrate characteristics
(for which the backscatter derivatives were a proxy) had a positive, and sometimes more
important impact on the performance of the classifications than terrain morphology (quanti-
fied through terrain attributes); adding bathymetry and the backscatter derivatives to terrain
attribute variables often increased map accuracy for both benthoscape and sea scallop suitabil-
ity distributions on German Bank. For the unsupervised classifications, only two of the five
bottom types (glacial till and reefs) seemed to be driven to a certain level by local geomorphol-
ogy. In addition, reefs and rippled silt were poorly discriminated by a majority of classifica-
tions, likely because their distribution is influenced by other environmental factors or that the
variables tested were measured and analyzed at a scale that did not match the scale of the rele-
vant geomorphological features [6]. In agreement with results from Brown et al. [8], the Max-
Ent analysis showed that bathymetry, sediment properties and rugosity are important
variables in predicting sea scallops distribution, but that aspect, slope and relative position are
not. Only four SDMs out of 28 performed better than the model with no topography (only
bathymetry and the three backscatter derivatives).
Other variables (e.g. physical, oceanographic, ecological) may drive particular species or
assemblage distributions more than terrain geomorphology. However, they were not used in
this study as they were not available at the same spatial scale as the MBES data. When includ-
ing more variables, users need to keep in mind that covariation may influence models like
MaxEnt. If an oceanographic variable is correlated with a terrain characteristic, the user needs
to keep only one of them. This is also true of the proposed selection of terrain attributes; as
demonstrated in Lecours et al. [4] and confirmed in the current study, each of the six proposed
terrain attributes captures a unique characteristic of the terrain, but some of these characteris-
tics may be spatially correlated in a certain area.
The framework for selecting terrain attributes for ecological studies proposed by Lecours
et al. [4], and supported by the findings of this study, aims at helping the end-users select a
robust combination of terrain attributes that best captures the different characteristics of ter-
rain geomorphology. The recommended selection of six terrain attributes serves as a guide as
to which set of attributes should be tested in order to achieve the best outcome. It provides
end-users with an optimal set of attributes, from which a subset combined with other environ-
mental variables can result in a high accuracy map or model output. The best results will not
necessarily come from the use of all six terrain attributes, but may only come from some of
them. For instance, if particular terrain characteristics have no ecological meaning in an appli-
cation, using the terrain attributes that capture these characteristics will not yield the best out-
come. It is therefore highly site and case specific as to which variables should be included [6].
Nonetheless, the recommended approach provides the optimal starting point from which ter-
rain attributes can be selected.
Consequences of Variable Selection
Results highlight the importance of appropriately selecting input variables in both unsuper-
vised and supervised classifications, and consequently the inappropriateness of making such
selection arbitrarily. For instance, the benthoscape map generated from the combination of
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 14 / 18
Selection 1 with bathymetry and the backscatter derivatives yielded an overall accuracy
and a kappa coefficient of agreement that are respectively only 0.2% and 0.6% different
than the map built from Selection 6, bathymetry and the backscatter derivatives. It would
be quite intuitive to interpret the difference in map outputs as insignificant based only on
these measures of accuracy. However, 10.0% of the study area was classified differently by
these two classifications, an area corresponding to about 362 km
. In addition, the differ-
ences occurred in all regions of the study area and across all the habitat types. In the worst
case scenario (i.e. the difference between Selection 4 and Selection 1, Table 2), the total
area that was mapped differently covers over 2,115 km
. The results of this study indicate
that a subjective selection of terrain attributes could potentially provide a map that is in
average 26.7% different in terms of the location and boundaries of benthoscape classes,
which has serious implications for ecological applications that use these maps and models
for decision-making.
Comparisons with Other Studies: Terrestrial and Marine
Many different terrain attribute selections have been used in terrestrial and marine ecology
([1,7]; references therein). In a meta-analysis of ecological studies using geomorphometry,
Bouchet et al. [1] found that about a third of the studies only used one terrain attribute and
that very few authors used more than four. While focusing on forest ecosystems, Sharaya &
Sharyi [40] wrote that in general, one to three basic terrain attributes are used to study land-
scape phenomena and that the “insufficient representativeness” (ibid, p. 2) of terrain attributes
makes for an inefficient use of topography as a variable in ecology. In a management context
and using the same dataset as in the current study, Brown et al. [8] selected six terrain attri-
butes based on previous use in marine ecology studies and “iterative testing of a large number
of different layers by the authors” (ibid, p. 3). This relatively subjective way of selecting terrain
attributes is the most common one in ecology. However, it provides many significant and
valid insights for many applications; most of the common terrain attributes found in the eco-
logical literature (e.g. local mean, slope, aspect) [1] are part of our proposed selection, or are
related to one of the proposed attributes. For instance, different types of curvature are com-
monly used, which Lecours et al. [4] found to be correlated to the recommended relative devia-
tion from mean value, although more ambiguously defined and thus not included in the
recommended selection. Despite using a subjective selection of terrain attributes, Brown et al.
[8] yielded valid results. Their MaxEnt model had a high predictive capacity, although it had
some level of over-fitting and was less robust than some of the best models of the current
study. If implemented in the current study using the same method, an unsupervised classifica-
tion made from their selection of variables would rank amongst the best benthoscape maps
and be 90.8% similar to the map built with Selection 1 and the four other environmental vari-
ables. This demonstrates that despite potentially resulting in huge differences (c.f. “Conse-
quences of Variable Selection” above), subjective selection of terrain attributes can sometimes
produce relevant and valid results. As a matter of fact, the model by Brown et al. [8] has been
used in subsequent studies and to inform fisheries stock assessment process and management
Finally, the observed differences between the overall accuracy measures and the kappa coef-
ficients of agreement confirm that the overall accuracy might be a poor guide of the value of a
classification, something that has been already argued in the literature [45,46]. Based on our
results, we would recommend the kappa coefficient as a more appropriate measure than the
overall accuracy for ecological mapping, but as recently highlighted by Diesing et al. [5], there
is a need to move towards spatial representations of accuracy.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 15 / 18
Selecting the most appropriate environmental variables to use in a specific study can be very
challenging. This study demonstrated the importance of carefully selecting variables for eco-
logical work; maps and models that perform similarly can still produce very different spatial
outcomes, which can have important implications when these maps and models are used in
decision-making for conservation and management. Using two different approaches to habitat
mapping, this paper also confirmed that the selection of terrain attributes recommended in
Lecours et al. [4] performs better than other selections, thus serving as a guide to make better
use of geomorphometry in ecology. Results also showed that while this selection of terrain
attributes ensures that most of the local topographic structure is captured when performing
terrestrial or marine ecological studies, and while terrain morphology can help improve maps
and models, it is not always the most important environmental factor for all ecological applica-
tions. The relationship between terrain morphology and ecological phenomena is species, area
and scale-dependent [6]. The use of the proposed selection of terrain attributes, in combina-
tion with other environmental variables (e.g. precipitations, climate, currents), will help ecolo-
gists produce more robust analyses and generate maps and models with a higher degree of
confidence. In order to get the best representation of the environment as possible and to best
inform policy, conservation and management efforts, we recommend (1) that stakeholders
prepare more than a single map using different combinations of environmental variables,
and (2) that they select the best outcome based on map accuracy or model performance
Supporting Information
S1 Appendix. Additional information on the selections of variables.
Many thanks are due to Dr. Jessica A. Sameoto, Dr. Stephen J. Smith, and the Department of
Fisheries and Oceans Canada for sharing and allowing us to use the German Bank dataset.
Authors are listed by order of contribution.
Author Contributions
Conceptualization: VL RD CJB ENE VLL.
Formal analysis: VL CJB.
Funding acquisition: VL RD.
Investigation: VL.
Methodology: VL CJB RD.
Project administration: VL RD ENE.
Resources: CJB RD ENE.
Software: VL.
Supervision: CJB RD VLL ENE.
Writing – original draft: VL.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 16 / 18
Writing – review & editing: VL CJB RD VLL ENE.
1. Bouchet PJ, Meeuwig JJ, Kent CPS, Letessier TB, Jenner CK. Topographic determinants of mobile ver-
tebrate predator hotspots: current knowledge and future directions. Biol Rev. 2015; 90:699–728. doi:
10.1111/brv.12130 PMID: 25125200
2. Dolan MFJ, Lucieer VL. Variation and uncertainty in bathymetric slope calculations using geographic
information systems. Mar Geod. 2014; 37:187–219.
3. Jones KH. A comparison of two approaches to ranking algorithms used to compute hill slopes. GeoIn-
formatica. 1998; 2:235–256.
4. Lecours V, Devillers R, Simms AE, Lucieer VL, Brown CJ (under review). Towards a framework for ter-
rain attribute selection in environmental studies.
5. Diesing M, Mitchell P, Stephens D. Image-based seabed classification: what can we learn from terres-
trial remote sensing? ICES J Mar Sci. 2016;
6. Lecours V, Devillers R, Schneider DC, Lucieer VL, Brown CJ, Edinger EN. Spatial scale and geographic
context in benthic habitat mapping: review and future directions. Mar Ecol Prog Ser. 2015; 535:259–
7. Lecours V, Dolan MFJ, Micallef A, Lucieer VL. Characterising the ocean frontier: a review of marine
geomorphometry. Hydrol Earth Syst Sc Disc. 2016,
8. Brown CJ, Sameoto JA, Smith SJ. Multiple methods, maps, and management applications: purpose
made seafloor maps in support of ocean management. J Sea Res. 2012; 72:1–13.
9. Lecours V. Terrain Attribute Selection for Spatial Ecology (TASSE), v. 1.0. 2015. Available from: www.
10. Zajac RN. Challenges in marine, soft-sediment benthoscape ecology. Landscape Ecol. 2008; 23:7–18.
11. Jensen JR. Introductory digital image processing, a remote sensing perspective. 3rd ed. Upper Saddle
River: Pearson Prentice Hall; 2005.
12. Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA. Evaluating resource selection functions. Ecol
Model. 2002; 157:281–300.
13. Congalton RG. A review of assessing the accuracy of classifications of remotely sensed data. Remote
Sens Environ. 1991; 37:35–46.
14. Jaynes ET. Information theory and statistical mechanics. Phys Rev. 1957; 106:620–630.
15. Phillips SJ, Dudı
´k M, Schapire RE. A maximum entropy approach to species distribution modeling. Pro-
ceedings of the 21st International Conference on Machine Learning. New Yord: ACM Press; 2004.
p. 655–662.
16. Phillips SJ, Anderson RP, Schapire RE. Maximum entropy modeling of species geographicdistribu-
tions. Ecol Model. 2006; 190:231–259.
17. Monk J, Ierodiaconou D, Versace VL, Bellgrove A, Harvey E, Rattray A, et al. Habitat suitability for
marine fishes using presence-only modelling and multibeam sonar. Mar Ecol Prog Ser. 2010: 420;157–
18. Jime
´nez-Valverde A. Insights into the area under the receiver operating characteristic curve (AUC) as a
discrimination measure in species distribution modelling. Global Ecol Biogeogr. 2012; 21:498–507.
19. Peterson AT, PapeşM, Sobero
´n J. Rethinking receiver operating characteristic analysis applications in
ecological niche modeling. Ecol Model. 2008; 213:63–72.
20. Lobo JM, Jime
´nez-Valverde A, Real R. AUC: a misleading measure of the performance of predictive
distribution models. Global Ecol Biogeogr. 2008; 17:145–151.
21. McPherson JM, Jetz W, Rogers DJ. The effects of species’ range sizes on the accuracy of distribution
models: ecological phenomenon or statistical artefact? J Appl Ecol. 2004; 41:811–823.
22. Vaughan IP, Ormerod SJ. The continuing challenges of testing species distribution models. J Appl Ecol.
2005; 42:720–730.
23. Elith J, Phillips SJ, Hastie T, Dudı
´k M, Chee YE, Yates CJ. A statistical explanation of MaxEnt for ecolo-
gists. Divers Distrib. 2011; 17:43–57.
24. Merow C, Smith MJ, Silander JA Jr. A practical guide to MaxEnt for modeling species’ distributions:
what it does, and why inputs and settings matter. Ecography. 2013; 36:1058–1069.
25. Fitzpatrick MC, Gotelli NJ, Ellison AM. MaxEnt versus MaxLike: empirical comparisons with ant species
distributions. Ecosphere. 2013; 4:1–15.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 17 / 18
26. Zweig MH, Campbell G. Receiver-operating characteristics (ROC) plots: a fundamental evaluation tool
in clinical medicine. Clin Chem. 1993; 39:561–577. PMID: 8472349
27. Warren DL, Seifert SN. Ecological niche modeling in Maxent: the importance of model complexity and
the performance of model selection criteria. Ecol Appl. 2011; 21:335–342. PMID: 21563566
28. Khatchikian C, Sangermano F, Kendell D, Livdahl T. Evaluation of species distribution model algorithms
for fine-scale container breeding mosquito risk prediction. Med Vet Entomol. 2011; 25:268–275. doi: 10.
1111/j.1365-2915.2010.00935.x PMID: 21198711
29. Palialexis A, Georgakarakos S, Karakassis I, Lika K, Valavanis VD. Prediction of marine species distri-
bution from presence-absence acoustic data: comparing the fitting efficiency and predictive capacity of
conventional and novel distribution models. Hydrobiologia. 2011; 670; 241–266.
30. Peitgen HO, Saupe D. The science of fractal images. New York: Springer-Verlag; 1988.
31. Chipperfield JD, Dytham C, Hovestadt T. An updated algorithm for the generation of neutral landscapes
by spectral synthesis. PLOS ONE. 2011; 6:e17040. doi: 10.1371/journal.pone.0017040 PMID:
32. Pfeifer P. Fractal dimension as working tool for surface-roughness problems. Appl Surf Sc. 1984;
33. Keitt TH. Spectral representation of neutral landscapes. Landscape Ecol. 2000; 15:479–493.
34. With KA, King AW. The use and misuse of neutral landscape models in ecology. Oikos. 1997; 79:219–
35. Tate NJ. Estimating the fractal dimension of synthetic topographic surfaces. Comput Geosci. 1998;
36. Evans IS, McClean CJ. The land surface is not unifractal: variograms, cirque scale and allometry. Z
Geomorphol. 1995; 101:127–147.
37. Herzfeld UC, Overbeck C. Analysis and simulation of scale-dependent fractal surfaces with application
to seafloor morphology. Comput Geosci. 1999; 25:979–1007.
38. Milne BT. Spatial aggregation and neutral models in fractal landscapes. Am Nat. 1992; 139:32–57.
39. Halley JM, Hartley S, Kallimanis AS, Kunin WE, Lennon JJ, Sgardelis SP. Uses and abuses of fractal
methodology in ecology. Ecol Lett. 2004; 7:254–271.
40. Sharaya LS, Sharyi PA. Geomorphometric study of the spatial organization of forest ecosystems. Russ
J Ecol+. 2011; 42:1–8.
41. Sameoto JA, Smith SJ, Glass A, Hubley B, Denton C. Scallop Fishing Area 29: stock status and update
for 2014. DFO Can Sci Advis Sec Res Doc 2014/ 64. 2014: 67 p.
42. Sameoto JA, Smith SJ, Nasmith LE, Glass A, Denton C. Scallop Fishing Area 29: stock status and
update for 2015. DFO Can Sci Advis Sec Res Doc 2015/ 067. 2015: 69 p.
43. Smith SJ, Hubley PB. Impact of survey design changes on stock assessment advice: sea scallops.
ICES J Mar Sci. 2014; 71:320–327.
44. Smith SJ, Sameoto JA, Brown CJ. Setting biological reference points for sea scallops Placopecten
magellinacus allowing for the spatial distribution of productivity and fishing effort. Can J Fish Aquat Sci.
45. Felix NA, Binney DL. Accuracy assessment of a Landsat-assisted vegetation map of the coastal plain of
the Arctic National Wildlife Refuge. Photogramm Eng Rem S. 1989; 55:475–478.
46. Fielding AH, Bell JF. A review of methods for the assessment of prediction errors in conservation pres-
ence/absence models. Environ Conserv. 1997; 24:38–49.
Comparing Environmental Variables for Ecological Studies
PLOS ONE | DOI:10.1371/journal.pone.0167128 December 21, 2016 18 / 18

Supplementary resource (1)

... 9258-107) consisting of species different from Acacia seyal represent its absences. (Lecours et al., 2016). The environmental variables relevant for each species were thus selected by performing ecological niche factor analysis (ENFA; Basille et al., 2008;Hirzel et al., 2002). ...
... An important aspect that has a pivotal role in the accuracy of the prediction for a given species is the selection of relevant environmental variables used as predictors during the modeling flow. Making a subjective choice or inappropriate selection may reduce prediction accuracy (Lecours et al., 2016). The right method of variable selection is closely related to the aims and questions of the study. ...
Full-text available
Around the world, SDMs have been widely used to support forest management planning and biodiversity conservation. Beyond the prediction of species distribution provided by the SDMs, this study aimed to analyze the spatial distribution of tree species diversity using SDMs. The study area is a Faidherbia albida parkland in Central Senegal. It is characterized by a tree-based farming system dominated by Faidherbia albida. Using a robust and representative dataset of 9258 tree species occurrence, we first determined by an SDM the current potential spatial distribution of the 16 main tree species forming the parkland. Specifically, using 6 SDM algorithms and applying several modeling techniques with different categories of predictor variables (e.g., climate, topography, soil properties and human impact) we benchmarked 576 SDMs to achieve best model predictions for tree species. Then, tree species diversity maps were created on the basis of the resulting SDM predictions. Finally, the spatial dynamics of tree species diversity were discussed in relation to landscape characteristics, including heterogeneity, composition and human impact. The results showed that there is no single ‘best’ SDM algorithm (among the 6 algorithms tested) or modeling approach for all species. Benchmarking several modeling techniques allowed strengthening SDM performance, achieving AUC values that ranged from 0.64 (intermediate accuracy) to 0.87 (very good accuracy). The spatial dynamics of tree species diversity is related to the landscape heterogeneity and composition. In the Sahelian agroforestry systems (AFS), tree diversity is sustained by anthropization. A significant negative correlation with the distance to the village was found, i.e. the closer you get to the village, the greater the diversity of trees. This study could be crucial for analyzing tree species diversity when abundance information is not available.
... ARA attempts to use the angular backscatter intensity information to estimate type of sediment as well as other sediment properties, such as the sediment grain size, index of impedance and volume heterogeneity (Mulhearn, 2000;Fonseca et al., 2009). These inversion parameters are useful for identifying differences in the seafloor that can be attributed to variations in features and biological habitats (Brown and Blondel, 2009;Che Hasan et al., 2014;Lecours et al., 2016a). However, there are no studies that assess the importance of combined AR and ARA mosaics on classification accuracy. ...
... The lower coverage of high complexity substrates likely caused the lower importance of those variables associated with complexity compared to ARA derivatives. Previous studies that use MBES data to classify seafloor habitat have typically found bathymetry (depth) to be the highest ranking variable in terms of importance (Che Hasan et al., 2014;Rattray et al., 2015;Lecours et al., 2016a;Ierodiaconou et al., 2018). Since our site is dominated by granitic reefs it is important to note that the transition of the granitic reefs is relatively steep and can cause issues with lack of observations across the depth range. ...
Full-text available
Accurate maps of biological communities are essential for monitoring and managing marine protected areas but more information on the most effective methods for developing these maps is needed. In this study, we use Wilsons Promontory Marine National Park in southeast Australia as a case study to determine the best combination of variables and scales for producing accurate habitat maps across the site. Wilsons Promontory has full multibeam echosounder (MBES) coverage coupled with towed video, remotely operated underwater vehicle (ROV) and drop video observations. Our study used an image segmentation approach incorporating MBES backscatter angular response curve and bathymetry derivatives to identify benthic community types using a hierarchical habitat classification scheme. The angular response curve data were extracted from MBES data using two different methods: 1) angular range analysis (ARA) and 2) backscatter angular response (AR). Habitat distributions were predicted using a supervised Random Forest approach combining bathymetry, ARA, and AR derivatives. Variable importance metrics indicated that ARA derivatives, such as grain size, impedance and volume heterogeneity were more important to model performance than AR derivatives mean, skewness, and kurtosis. Additionally, this study investigated the impact of segmentation software settings when creating segmented surfaces and their impact on overall model accuracy. We found using fine scale segmentation resulted in the best model performance. These results indicate the importance of incorporating backscatter derivatives into biological habitat maps and the need to consider scale to increase the accuracy of the outputs to help improve the spatial management of marine environments.
... Depending on the interest of the research, different layers of information are needed. However, there is a consensus in the literature that mapping the seafloor is the basis for knowing the distribution of geodiversity and its relationship with marine biodiversity (Lundblad et al. 2006;Brown et al. 2011Brown et al. , 2012Diesing et al. 2016;Lecours et al. 2016;Baker and Harris 2020;Harris and Baker 2020). It is, therefore, necessary to link the mapping of the seabed to the concept of habitat, that is, to map the nature, distribution, and extent of physical environments as a way of seeking to predict or have a forecast basis for the occurrence of associated biological communities (Quaresma et al. 2020). ...
Continental shelves are areas of high heterogeneity that are poorly understood and increasingly influenced by anthropic activities. The tropical passive continental shelf of northeastern Brazil is the narrowest in the country and in some stretches is among the world’s narrowest. Narrow continental shelves are unusual features on passive continental margins. This is clearly reflected in the small number of studies in the international literature. In this chapter, we present a synthesis of the current knowledge about seafloor morphology, its associated benthic ecosystems, and the role of eustatic variations in the evolution of this tropical shelf. Major human uses are also discussed.KeywordsContinental shelfShelf sedimentationSiliciclastic–carbonate sedimentation
... A map of benthic structures was prepared to complement and validate the seascape geomorphology data available in the scientific literature and the data produced by participatory mapping [61][62][63] with the extension (BTM), for ArcGIS™ 10.5. A bathymetric grid was used, following the methodology and classification criteria proposed for the Pernambuco Continental Shelf, in the Northeast of Brazil [64,65]. ...
Full-text available
Seascape mapping is critical to understanding ecosystem services and managing areas with potential for fishing, power generation, mining, and tourism. Despite advances in marine geophysics, the necessary equipment to make underwater cartography can be expensive and requires a certain degree of specialization. In areas with scarce data, ethnomapping can be used for the elaboration or complementation of marine cartography. In addition, it provides information about the nature, concepts, phenomena, and nomenclatures attributed by the local population. The aim of this study was to integrate the knowledge of artisanal fishermen from NE Brazil into the mapping of seascapes, validating the obtained cartography with scientific sampling. Focus groups were used to promote an open discussion of local users' knowledge about seascapes and their importance for local fisheries. After analyzing, it was possible to correlate the products of participatory mapping with the scientific data available in the literature, resulting in the seascape ethnomap. Nine seascapes relevant to subsistence fisheries were identified. The mapping of seascapes and fauna captured by indigenous fishermen was similar to that produced from preexisting geology, geomorphology, and fishing data. This validated the methodological protocol and the importance of the participation of local populations in coastal conservation and management activities.
... Harris and Baker (2012) describe benthic marine habitats as geographically distinct areas of the seafloor where physical attributes are linked to species or communities that occur in clusters. Other researchers have examined the relationship between the seafloor's physical, chemical, and biological features to geographically designate habitats with similar characteristics using (geo) statistical approaches (Lecours et al. 2016). With some benthic habitat mapping studies adopting a landscape-scale strategy similar to mapping biophysical patterns in the terrestrial environment, spatial scale in defining habitat is often problematic (Lecours et al. 2015). ...
Full-text available
Setiawan A, Siregar VP, Susilo SB, Mardiastuti A, Agus SB. 2022. Geomorphological classification of benthic structures of Kaledupa Atoll Wakatobi National Park, Indonesia. Biodiversitas 23: 3784-3792. Geomorphological mapping is one solution to understanding the complexity of the seabed pattern. This research aims to map the benthic structure of Kaledupa Atoll using the Remote Sensing and Geographic Information System with the Benthic Terrain Modeler (BTM) extension. The data used is in situ bathymetry data using Mapsounder 585 and Sentinel 2A satellite image data downloaded for free. BTM is a technique for examining benthic habitat and shallow water geomorphology. BTM was used to analyze integrated depth data to determine the bathymetric position index (BPI), slope, and classification of reef geomorphological structures. Thirteen structural classes of habitat geomorphology resulting from this research are Flat plain with the largest area of ??17142.35 ha, and the class with the smallest area is Crevices class, with narrow gullies rock outcrops 1.64 ha. While the other classes in a row are Broad slope 9104.53 ha; Flat ridge tops, upper slopes 4262.67 ha; Current scoured depressions on a slope 3420.98 ha; Narrow depressions at the base of rock outcrops 2348.20 ha; Rock outcrop highs, narrow ridges 1792.45 ha; Local ridges, boulders, pinnacles on slopes 879.29 ha; Scarp, cliff or small local depressions on a slope 703.02 ha; Local ridges, boulders, pinnacles on broad flat 490.17 ha; Local depressions, current scoured on flat 416.39 ha; Local ridges, boulders, pinnacles on broad depressions 16.63 ha; and Steep slopes 2.88 ha.
... Spatial pattern metrics have been developed to quantify seascape composition (the abundance and variety of patch types), configuration (the spatial arrangement of patch types) and terrain morphology (e.g. slope, structural complexity) from habitat maps or digital bathymetric models (Lecours et al., 2016;Wedding et al., 2011). In shallow water reefs, these have, amongst others, provided new insights into seascape connectivity (McMahon et al., 2012), species-specific responses to environmental structure (Hitt et al., 2011), the importance of terrain complexity (Wedding et al., 2019) and scale-dependent responses (Kendall et al., 2011). ...
Full-text available
Benthic components of tropical mesophotic coral ecosystems (MCEs) are home to diverse fish assemblages, but the effect of multiscale spatial benthic characteristics on MCE fish is not well understood. To investigate the influence of fine‐scale benthic seascape structure and broad‐scale environmental characteristics on MCE fish, we surveyed fish assemblages in Seychelles at 30, 60 and 120 m depth using submersible video transects. Spatial pattern metrics from seascape ecology were applied to quantify fine‐scale benthic seascape composition, configuration and terrain morphology from structure‐from‐motion photogrammetry and multibeam echosounder bathymetry and to explore seascape–fish associations. Hierarchical clustering using fish abundance and biomass data identified four distinct assemblages separated by the depth and geographic location, but also significantly influenced by variations in fine‐scale seascape structure. Results further revealed variable responses of assemblage characteristics (fish biomass, abundance, trophic group richness, Shannon diversity) to seascape heterogeneity at different depths. Sites with steep slopes and high terrain complexity hosted higher fish abundance and biomass, with shallower fish assemblages (30–60 m) positively associated with aggregated patch mixtures of coral, rubble, sediment and macroalgae with variable patch shapes. Deeper fish assemblages (120 m) were positively associated with relief and structural complexity and local variability in the substratum and benthic cover. Our study demonstrates the potential of spatial pattern metrics quantifying benthic composition, configuration and terrain structure to delineate mesophotic fish–habitat associations. Furthermore, incorporating a finer‐scale perspective proved valuable to explain the compositional patterns of MCE fish assemblages. As developments in marine surveying and monitoring of MCEs continue, we suggest that future studies incorporating spatial pattern metrics with multiscale remotely sensed data can provide insights will that are both ecologically meaningful to fish and operationally relevant to conservation strategies. To investigate the influence of benthic seascape structure on MCE fish assemblages in Seychelles, this study surveyed benthic structure and fish assemblages using submersible video transects at mesophotic depths. Spatial pattern metrics measuring benthic habitat composition, configuration and terrain structure were extracted from Structure‐from‐Motion photogrammetry models to quantify fish‐habitat associations. The results revealed depth‐ and site driven grouping of mesophotic fish assemblages that show significant associations with fine‐scale (cm‐m) terrain structure, seascape composition and configuration.
... The TF was surrogated by the surface elevation (SE) and landscape position (LP). Given that topography plays a critical role in explaining the spatial variability of soil properties, selecting appropriate topographic conditions is crucial, but it can be particularly challenging for ecosystem studies (Lecours et al., 2016). Pilesjő et al. (2005) demonstrated that topographic elevation and drainage area are two suitable attributes for explaining variabilities in organic matter, clay content, and pH, while Wang et al. (2001) illustrated that LP controls soil properties at the hillslope catchment scale. ...
Identifying environmental factors that influence the vegetation community formation of various habitats in semiarid regions is imperative for managing these fragile ecosystems. The objective of this study was to quantify the relationships between selected environmental factors and vegetation community composition (VCC) in meadow and sand dune habitats using a structural equation model (SEM). Ecological data from 48 sample sites in semiarid sandy land were analyzed. The selected latent environmental variables included soil water (SW), topsoil salinity and sodicity, topsoil nutrients, topsoil texture (TT) and topography feature (TF). SW was the most influential factor for the Phragmites communis, Agriophyllum squarrosum, and Caragana microphylla communities composition. TT was the most influential factor for the Artemisia halodendron and Artemisia frigida - Agropyron desertorum communities composition. TF was the most influential factor for the Carex duriuscula and Phragmites communis - Bolboschoenus yagara communities composition. Considering the direct and indirect impact relationship, for the meadow habitat, TF was the primary factor causing the community distribution. For the sand dune habitat, TT was the primary factor causing the community distribution. As expected, the quantitative relationships between VCC and influencing factors differ depending on the ecological scales of interest.
Full-text available
The benthic structure of seamounts is critical for understanding the ecological environment and for assessing the influence of resource exploitation. However, the characteristics of the benthic structure of the seamount, especially for guyot, are still far from being clearly understood. For the first time, we carried out detailed hydroacoustic mapping in conjunction with surficial sediment sampling and underwater video recording to investigate the geomorphical and biological characteristics of Pallada Guyot in the Western Pacific Ocean (WPO). We utilized the Benthic Terrain Modeler (BTM) as an initial step to describe the detailed benthic structures and then classify the textural seabed according to backscatter images and sediment samples. We further discussed the relationship between geomorphology and the occurrence of benthic megafauna from video images. The results revealed that 13 classes of benthic structural zones were differentiated, and the dominant zones were flat abyssal plains, where the number and size of megafauna were smaller than those on the flank and flat-topped areas. The second most notable feature is the flat top, where sea cucumbers, starfish, fish, and shrimp have higher biomass and diversity. In the flank region, which is characterized by complex and extensive current-scoured ridges and depressions, sponges and corals are distributed in high-relief bedrocks. We also found that the maximum water depth where cold-water corals develop is 2,250 m. The sponge grounds appear in a marked bathymetric belt (1,800–2,150 m), which is shallower than that on a tropical seamount (2,500–3,000 m) located in the northeastern Atlantic. The findings of this study contribute to understanding the geomorphological drivers and biogeography of WPO seamounts and provide a reference for identifying priority areas for improved marine mineral planning in WPO areas.
Full-text available
Coastal habitats have the potential to be biodiversity hotspots that provide important ecosystem services, but also hotspots for human development and exploitation. Continued use of coastal ecosystem services requires establishing baselines that capture the present state of the benthos. This study employs habitat mapping to establish a baseline describing the spatial distribution of benthic organisms along the western coast of Placentia Bay, an Ecologically and Biologically Significant Area (EBSA) in Newfoundland, Canada. The influence of seafloor characteristics on the distribution of four dominant epifaunal assemblages and two macrophyte species were modelled using two machine learning techniques: the well-established Random Forest and the newer Light Gradient Boosting Machine. When investigating model performance, the inclusion of fine-scale (<1 m) substrate information from the benthic videos was found to consistently improve model accuracy. Predictive maps developed here suggest that the majority of the surveyed areas consisted of a species-rich epifaunal assemblage dominated by ophiuroids, porifera, and hydrozoans, as well as prominent coverage by Agarum clathratum and non-geniculate crustose coralline algae. These maps establish a baseline that enables future monitoring of Placentia Bay’s coastal ecosystem, helping to conserve the biodiversity and ecosystem services this area provides.
Full-text available
Seafloor characteristics can help in the prediction of fish distribution, which is required for fisheries and conservation management. Despite this, only 5%–10% of the world's seafloor has been mapped at high resolution, as it is a time‐consuming and expensive process. Multibeam echo‐sounders (MBES) can produce high‐resolution bathymetry and a broad swath coverage of the seafloor, but require greater financial and technical resources for operation and data analysis than singlebeam echo‐sounders (SBES). In contrast, SBES provide comparatively limited spatial coverage, as only a single measurement is made from directly under the vessel. Thus, producing a continuous map requires interpolation to fill gaps between transects. This study assesses the performance of demersal fish species distribution models by comparing those derived from interpolated SBES data with full‐coverage MBES distribution models. A Random Forest classifier was used to model the distribution of Abalistes stellatus, Gymnocranius grandoculis, Lagocephalus sceleratus, Loxodon macrorhinus, Pristipomoides multidens, and Pristipomoides typus, with depth and depth derivatives (slope, aspect, standard deviation of depth, terrain ruggedness index, mean curvature, and topographic position index) as explanatory variables. The results indicated that distribution models for A. stellatus, G. grandoculis, L. sceleratus, and L. macrorhinus performed poorly for MBES and SBES data with area under the receiver operator curves (AUC) below 0.7. Consequently, the distribution of these species could not be predicted by seafloor characteristics produced from either echo‐sounder type. Distribution models for P. multidens and P. typus performed well for MBES and the SBES data with an AUC above 0.8. Depth was the most important variable explaining the distribution of P. multidens and P. typus in both MBES and SBES models. While further research is needed, this study shows that in resource‐limited scenarios, SBES can produce comparable results to MBES for use in demersal fish management and conservation. This study assesses the performance of demersal fish species distribution models by comparing those derived from interpolated singlebeam (SBES) data with full‐coverage multibeam (MBES) distribution models. The results indicated that the performance of the SBES distribution models was not significantly different from that of the MBES models. While further research is needed, this study shows that in resource‐limited scenarios, SBES can produce comparable results to MBES for use in demersal fish management and conservation.
Full-text available
Toolbox for ArcGIS built from results of Lecours et al. (2017) and validated in Lecours et al. (2016). This toolbox generates six independent terrain attributes that together summarize topographic or bathymetric variability. References: Lecours, V., Devillers, R., Simms, A.E., Lucieer, V.L., and Brown, C.J. (2017) Towards a framework for terrain attribute selection in environmental studies. Environmental Modelling & Software, 89:19-30. Lecours, V., Brown, C.J., Devillers, R., Lucieer, V.L., and Edinger, E.N. (2016) Comparing selections of environmental variables for ecological studies: a focus on terrain attributes. PLoS ONE, 11:e0167128. TO CITE: v. 1.1: Lecours, V. (2017) Terrain Attribute Selection for Spatial Ecology (TASSE) v. 1.0: Lecours, V. (2015) Terrain Attribute Selection for Spatial Ecology (TASSE)
Full-text available
Toolbox for ArcGIS built from results of Lecours et al. (2017) and validated in Lecours et al. (2016). This toolbox generates six independent terrain attributes that together summarize topographic or bathymetric variability. References: Lecours, V., Devillers, R., Simms, A.E., Lucieer, V.L., and Brown, C.J. (2017) Towards a framework for terrain attribute selection in environmental studies. Environmental Modelling & Software, 89:19-30. Lecours, V., Brown, C.J., Devillers, R., Lucieer, V.L., and Edinger, E.N. (2016) Comparing selections of environmental variables for ecological studies: a focus on terrain attributes. PLoS ONE, 11:e0167128. TO CITE: v. 1.1: Lecours, V. (2017) Terrain Attribute Selection for Spatial Ecology (TASSE) v. 1.0: Lecours, V. (2015) Terrain Attribute Selection for Spatial Ecology (TASSE
Full-text available
Maps that depict the distribution of substrate, habitat or biotope types on the seabed are in increasing demand by marine ecologists and spatial planners, underpinning decision making in relation to marine spatial planning and marine protected area network design. Yet, the science discipline of image-based seabed mapping has not fully matured and rapid progress is needed to improve the reliability and accuracy of maps. To speed up the process we have conducted a literature review of common practices in terrestrial image classification based on remote sensing data, a related discipline, albeit with a larger scientific community and longer history. We identified the following key elements of a mapping workflow: (i) Data pre-processing, (ii) Feature extraction, (iii) Feature selection, (iv) Classification, (v) Post-classification enhancements, and (vi) Evaluation of classification performance. Insights gained from the review served as a baseline against which recent seabed mapping studies were compared. In this way we identified knowledge gaps and propose modifications to the mapping workflow. A main concern in current seabed mapping practice is that a large amount of often correlated predictor features is extracted, creating a multidimensional feature space. To effectively fill this space with an appropriate amount of training samples is likely to be impossible. Hence, it is necessary to reduce the dimensionality of the feature space via data transformation [e.g. principal component analysis (PCA)] or feature selection and remove correlated features. We propose to make dimensionality reduction an integral part of any mapping workflow. We also suggest to adopt recommendations for accuracy assessment originally drawn up for terrestrial land cover mapping. These include the publication of two or more measures of accuracy including overall and class-specific metrics, publication of associated confidence limits and the provision of the error matrix.
Full-text available
Geomorphometry, the science that quantitatively describes terrains, has traditionally focused on the investigation of terrestrial landscapes. However, the dramatic increase in the availability of digital bathymetric data and the increasing ease by which geomorphometry can be investigated using Geographic Information Systems (GIS) has prompted interest in employing geomorphometric techniques to investigate the marine environment. Over the last decade, a suite of geomorphometric techniques have been applied (e.g. terrain attributes, feature extraction, automated classification) to investigate the characterisation of seabed terrain from the coastal zone to the deep sea. Geomorphometric techniques are, however, not as varied, nor as extensively applied, in marine as they are in terrestrial environments. This is at least partly due to difficulties associated with capturing, classifying, and validating terrain characteristics underwater. There is nevertheless much common ground between terrestrial and marine geomorphology applications and it is important that, in developing the science and application of marine geomorphometry, we build on the lessons learned from terrestrial studies. We note, however, that not all terrestrial solutions can be adopted by marine geomorphometric studies since the dynamic, four-dimensional nature of the marine environment causes its own issues, boosting the need for a dedicated scientific effort in marine geomorphometry. This contribution offers the first comprehensive review of marine geomorphometry to date. It addresses all the five main steps of geomorphometry, from data collection to the application of terrain attributes and features. We focus on how these steps are relevant to marine geomorphometry and also highlight differences from terrestrial geomorphometry. We conclude with recommendations and reflections on the future of marine geomorphometry.
Full-text available
The area under the receiver operating characteristic (ROC) curve, known as the AUC, is currently considered to be the standard method to assess the accuracy of predictive distribution models. It avoids the supposed subjectivity in the threshold selection process, when continuous probability derived scores are converted to a binary presence–absence variable, by summarizing overall model performance over all possible thresholds. In this manuscript we review some of the features of this measure and bring into question its reliability as a comparative measure of accuracy between model results. We do not recommend using AUC for five reasons: (1) it ignores the predicted probability values and the goodness-of-fit of the model; (2) it summarises the test performance over regions of the ROC space in which one would rarely operate; (3) it weights omission and commission errors equally; (4) it does not give information about the spatial distribution of model errors; and, most importantly, (5) the total extent to which models are carried out highly influences the rate of well-predicted absences and the AUC scores.
Full-text available
Understanding the effects of scale is essential to the understanding of natural eco - systems, particularly in marine environments where sampling is more limited and sporadic than in terrestrial environments. Despite its recognized importance, scale is rarely considered in benthic habitat mapping studies. Lack of explicit statement of scale in the literature is an impediment to better characterization of seafloor pattern and process. This review paper highlights the importance of incorporating ecological scaling and geographical theories in benthic habitat mapping. It reviews notions of ecological scale and benthic habitat mapping, in addition to the way spatial scale influences patterns and processes in benthic habitats. We address how scale is represented in geographic data, how it influences their analysis, and consequently how it influences our understanding of seafloor ecosystems. We conclude that quantification of ecological processes at multiple scales using spatial statistics is needed to gain a better characterization of species−habitat relationships. We offer recommendations on more effective practices in benthic habitat mapping, including sampling that covers multiple spatial scales and that includes as many environmental variables as possible, adopting continuum-based habitat characterization approaches, using statistical analyses that consider the spatial nature of data, and explicit statement of the scale at which the research was conducted. We recommend a set of improved standards for defining benthic habitat. With these standards benthic habitats can be defined as ‘areas of seabed that are (geo)statistically significantly different from their surroundings in terms of physical, chemical and biological characteristics, when observed at particular spatial and temporal scales’.
Management for the major sea scallop (Placopecten magellanicus) fisheries in Canada is based on maximum sustainable yield (MSY) biomass and fishing mortality reference points applied to the whole stock, under the assumption that fishing mortality is uniformly distributed in space. However, scallop fishing vessels concentrate fishing in areas that consistently exhibit high densities resulting in a nonuniform spatial distribution of fishing effort. This study applies a spatial model for fishing effort derived from satellite vessel monitoring system data, scallop habitat suitability maps, and relative scallop density from a spatial stock assessment model to evaluate precautionary approach reference points in support of sustainable management. Target harvest rates were evaluated in terms of MSY for the higher habitat suitability areas. The results indicated that although MSY for the spatial model were similar to those when assuming a uniform distribution of effort, the biomass and catch rates over all areas were higher. The spatial model predicted that the MSY would be taken with less fishing effort, potentially lessening the benthic impacts from the scallop fishery.
Terrain attributes (e.g. slope, rugosity) derived from digital terrain models are commonly used in environmental studies. The increasing availability of GIS tools that generate those attributes can lead users to select a sub-optimal combination of terrain attributes for their applications. Our objectives were to identify sets of terrain attributes that best capture terrain properties and to assess how they vary with surface complexity. 230 tools from 11 software packages were used to derive terrain attributes from nine surfaces of different topographic complexity levels. Covariation and independence of terrain attributes were explored using three multivariate statistical methods. Distinct groups of correlated terrain attributes were identified, and their importance in describing a surface varied with surface complexity. Terrain attributes were highly covarying and sometimes ambiguously defined within software documentation. We found that a combination of six to seven particular terrain attributes always captures more than 70% of the topographic structure of surfaces.