Journal of Great Lakes Research 38 (2012) 196–205
Contents lists available at SciVerse ScienceDirect
Journal of Great Lakes Research
journal homepage: www.elsevier.com/locate/jglr
Reporting on the condition of stream fish communities in the Canadian tributaries of
Lake Ontario, at various spatial scales
Les W. Stanfield ⁎
Ontario Ministry of Natural Resources, Southern Science and Information Section, Glenora Fisheries Station, 41 Latchery Lane, Picton, Ontario, Canada K0K-2T0
a r t i c l e
i n f o
Article history:
Received 5 January 2011
Accepted 7 February 2012
Available online 13 March 2012
Communicated by Thomas Hrabik
Index words:
Scaling effects
Report card
Restoration prioritization
a b s t r a c t
Assessing stream condition is a foundation of adaptive management. But, summarizing stream condition in
ways that smooth over the variance that is inherent within a watershed may lead to erroneous conclusions
about its condition. This study compared stream condition summarized at a variety of spatial scales from segments to tertiary watersheds in order to evaluate the effect of summarization procedures on map interpretation. Sites were classified as to their degree of impairment from a reference state; a rank sum approach based
on the length of stream sampled and simple scoring criteria (e.g., scores b 50 = F) was used to grade each
reporting area. This study determined that scaling effects, choice of mapping thresholds and the size of
reporting areas all affect the interpretation of results. Larger reporting areas were generally ranked lower;
the source of this tendency is that larger stream segments are generally in a more degraded condition.
There was a good correlation between predicted and observed stream segment condition, which indicates
that the segment scale is the most reliable and informative of the reporting units tested. Providing categorized site conditions and confidence rankings on all larger scale maps offers a partial solution to these challenges. Where multiple sites collected on one segment were available, dramatic changes in condition indicate
potential areas for restoration. This study confirmed the incremental nature of stream degradation associated
with intensive land use, and demonstrated how inclusion of anything but the smallest units of stream area
and confidence rankings in the results can generate biased interpretations of stream condition.
Crown Copyright © 2012 Published by Elsevier B.V. on behalf of International Association for Great Lakes
Research. All rights reserved.
Introduction
Effectively managing the fisheries and other values of the Great
Lakes basin requires an ecosystem approach that incorporates targeted protection, mitigation and restoration of tributary habitats
(GLFC, 2007). Implementation requires a multidisciplinary and hierarchical approach to address issues at the appropriate scale (Fausch
et al., 2002; NRC, 1992). Understanding and communicating the interrelationships of factors that influence the hierarchical nature of
ecosystems and how reporting area boundaries influence ecosystem
indicators have always been a major focus of ecology (Bailey et al.,
2007; Levin, 1992; MacArthur, 1972; Meentemeyer, 1989; Wu and
Li, 2006; Wu and Loucks, 1995). Effective linkages between ecosystem restoration strategies and ecosystem understanding require
studies that use response metrics that are important to the decision
makers, are sensitive to the predictor variables and ecological processes under examination, and capture scaling effects (Bailey et al.,
2007; Lewis et al., 1996; O'Neill et al., 1988; Wu and Hobbs, 2002).
Providing this context to decision makers has been hampered by
⁎ Tel.: + 1 613 476 8777.
E-mail address: Les.Stanfi[email protected].
the complexity of issues relating to the statistical properties of metrics and the implications from the first law of geography: “Everything
is related to everything else, but near things are more related than
distant things” sensu Tobler, 1970, that is the foundation of the laws
of spatial autocorrelation (Wu, 2004).
Ecologists are beginning to understand how patterns in ecosystem
response to human induced disturbances break down as the scale of
measurement changes. Reasons for the break down include, but are
not limited to the following: statistical properties associated with
spatial patterns (Fortin and Dale, 2005); smoothing effects in attribute properties that occur as the habitat unit size changes (Jelinski
and Wu, 1996; Schumm and Lichy, 1965; Wu and Li, 2006); and differences in the variance properties of an attribute that occur over
time and as habitat unit size changes (Dungan et al., 2002; Fortin
and Dale 2005). Wu (1999, 2004) suggests that landscape attributes
that possess hierarchical structure, be it structural or functional,
form multiple hierarchies in the same landscape that may not correspond to each other precisely, in terms of spatial and temporal scales.
Terrestrial ecologists are making tremendous progress in quantifying
and thereby understanding how scaling effects (proximity — study
extent etc.) and hierarchical structure influence the processes that affect terrestrial habitats, (Fortin and Dale 2005; Legendre and
Legendre, 1998; Wu and Li, 2006). The result is that studies now
0380-1330/$ – see front matter. Crown Copyright © 2012 Published by Elsevier B.V. on behalf of International Association for Great Lakes Research. All rights reserved.
doi:10.1016/j.jglr.2012.02.008
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
routinely identify the appropriate scale at which to report terrestrial
processes (e.g., Allen et al., 1984; Fisher et al., 1998; Jelinski and
Wu, 1996; O'Neill et al., 1996; Wu, 1999, 2004). Terrestrial ecologists
are embracing a new paradigm, “hierarchical patch dynamics”, that offers a framework for explicitly incorporating heterogeneity and scale,
and for integrating equilibrium, multiple equilibrium and nonequilibrium perspectives (Wu and Loucks, 1995). Since rivers are a mirror
of their watersheds (Schlosser, 1991), the challenge to riverine ecologists is to sort out the dual effects of the hierarchical influences from
the river network (longitudinal) and the residual horizontal scaling influences of the remaining terrestrial habitats.
A variety of primary variables (climate, geology, glacial history,
etc.) operate in interconnected ways to influence stream habitats in
a hierarchical nature (Dunne and Leopold, 1978; Fausch et al., 2002;
Frissell et al., 1986; Hynes, 1970; Imhof et al., 1996; Schlosser, 1991;
Seelbach et al., 1997; Vannote et al., 1980; Wiley et al., 1997). These
factors operate in similar ways for both terrestrial and aquatic habitats and as such, classification schemes such as those produced by
Maxwell et al. (1995) have been widely applied by both terrestrial
and aquatic ecologists (see Jensen and Bourgeron, 1995). Unfortunately, because of their nested hierarchy and inherent variability,
classification of riverine systems has required another level of complexity to elucidate how scaling effects influence the processes and communities of these ecosystems (Frissell et al., 1986; Seelbach et al., 1997;
Vannote et al., 1980). In summary, geomorphic and climatic processes
influence stream habitat units differentially in a hierarchy from ecoregion to watershed, to reach, to site, to a microhabitat unit. There is an intimate connection between the processes, the landscape and the river
network. Natural (weather and geomorphology) and human disturbances also interact with these processes differentially, depending on
the nature of the disturbance to habitats, the sensitivity of the habitat
unit to each disturbance (see Hynes (1970) for point source impacts
and Dunne and Leopold (1978) for landscape impacts), and the potential
residual hierarchical influences of upland habitats. Given this complexity, it is not surprising that at present there is a poor understanding of
how both the spatial and hierarchical scaling factors combine to influence the expression of stream metrics, yet this is critical to implementation of stream restorations. Wu (2004) refers to this problem as spatial
transmutation, or the tendency for the inappropriate extrapolation of
statistical relationships from one scale to another. This is of concern to
restoration ecologists because, if metrics used to study patterns respond
differently depending on underlying scaling factors, then efforts to direct
restoration to specific habitats based on this analysis could lead to project
failures. Further, conservation organizations routinely use GIS to report
on ecosystem condition at a variety of spatial scales (watershed to ecoregion) for the purposes of engaging/enraging citizens and to help with
priority setting for restoration activities. Without a good understanding
of the implications from spatial transmutation, there is high potential
for inappropriate conclusions being drawn on the condition of various
sized reporting areas from field data. However, it has yet to be demonstrated how habitat unit size and the distribution of data influence measures of, or map interpretations of stream conditions. This has the
potential to influence policy/restoration decisions.
The present study builds on the efforts of Kilgour and Stanfield
(2006) and Stanfield and Kilgour (2006) where a process for categorizing stream site conditions was developed. However, they did not
attempt to summarize the condition of streams in reporting areas
typically used by resource agencies (catchments, watersheds), evaluate the spatial patterns in the data, or test the accuracy of predictions
to the segment data as defined in their landscape datasets.
Therefore, the objectives of this paper are to:
1. conduct a regional assessment of streams draining into the Canadian side of Lake Ontario and use this data to explore how patterns
might be useful for directing efforts to reverse the creeping crisis
of habitat loss in Canada, as recommended by Pearse et al. (1985);
197
2. illustrate how reporting data at varying scales can influence perspective on the overall state of the rivers. The intent is to evaluate
how perspective changes when spatially extensive datasets are
summarized at various reporting scales and to demonstrate how
GIS mapping approaches can facilitate a consistent means of interpreting/reporting data at multiple scales to reveal the inherent
limitations in the data; and
3. evaluate the reliability of landscape model predictions of stream
segment condition by comparing results to field data. Wang et al.
(2006) identified this need as an important step to validate the
segment as the primary reporting unit for streams.
Methods
The study area covers streams located on the north shore of Lake Ontario (Fig. 1). Two physiographic features, the Oak Ridges Moraine and
the Niagara Escarpment, dominate this area and influence the generation of both natural and development spatial patterns in the area. The
extensive deep deposits consisting mostly of sands and gravels of the
moraine, and the fractured and karst nature of the escarpment, provide
large volumes of groundwater discharge to the streams that originate
from them. These features as well as settlement patterns generate
both a south to north and a west to east gradient of development
(Stanfield and Jackson, 2011) in the study area. Most forested areas
are located on top of the moraine and escarpment, along the river valleys and remnant wetlands; low intensity agriculture dominates the tablelands; and urban areas are predominant in proximity to Lake Ontario
and in the west-central part of the study area. The interaction of the
physiographic features and the intertwined land use gradient bring
about a gradient in the stream conditions, whereby streams in northern
areas tend to have lower water temperatures that gradually increase toward the river mouths (see Stanfield and Kilgour, 2006).
Streams in this region have a relatively diverse fish assemblage that
can range from cold-water intolerant communities of brook trout (Salvelinus fontinalis) and slimy sculpin (Cottus cognatus), to tolerant and
warm water assemblages that contain species such as common shiner
(Luxinus cornutus) and fathead minnows (Pimephales promelas). Additionally, these streams now contain a variety of naturalized non-native
salmonids: rainbow trout (Onchorhynchus mykiss), chinook salmon
(Oncorhynchus tshawytscha), coho salmon, (Oncorhynchus kisutch) and
brown trout (Salmo trutta).
Developing a fish assemblage metric
Single pass electrofishing survey data are reported from 704 sites
that were mostly randomly selected. Sites were typically 40–60 m in
length, beginning and ending at a crossover in order to minimize differences in habitats sampled between sites (Stanfield, 2007). A fish
assemblage metric was calculated for each site using canonical correspondence analysis (CA) from biomass data collected between 1995
and 2002 using single pass electrofishing. In two cases, species were
inconsistently identified, mottled sculpin (Cottus bairdi) and slimy
sculpin (Cottus cognatus), and American brook lamprey (Lampetra appendix) and sea lamprey (Petromyzon marinus). These data were combined for the analysis. Sites where no fish were captured and species
found at b5% of sites were not included in the analysis.
The CA axis scores for the log densities of each species generate a
synthetic score for each site that reflects the contrast in fish assemblage
composition across sites for the study area. Calculation of site and taxa
scores on the first ordination axis was performed by iteratively estimating the weighted-average sample scores and the weighted average taxa
scores. Sample scores in CA are usually scaled to a mean of zero and
standard deviation of 1 (ter Braak, 1992). A gradient in the fish assemblages along the CA axis 1 scores emerged by plotting the CA axis scores
for the site of the maximum biomass for each species. Cold-water sensitive taxa were located on the extreme left side of the correlation (high
198
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
Fig. 1. Relationship between landscape and fish assemblage composition as determined through Canonical Correspondence Analysis (CCA). Re-printed with permission and species
acronyms are defined in Table 6 Stanfield and Kilgour (2006). Por = porosity of soils; str = stream; PIC is analogous to LDI; BFI = baseflow index.
negative scores) and sites with warm water tolerant species were found
on the extreme right side (high positive numbers) (Fig. 1). For the remainder of the paper the CA axis 1 fish assemblage is referred to as
the shortened “fish assemblage metric”.
Catchment boundaries, slope, drainage area and proportions of
basin land use and surficial geology for each site's catchment were determined using Arc-hydro (v1.1 for ArcGIS 9.0) (Maidment, 2002) and
ArcGIS. Surficial geology was attributed for each site and converted to
a baseflow index (BFI) using a summed rank percentile approach
(Piggott et al., 2002). Slope for each site was determined over a 200 m
length of stream, centered at each site. Kilgour and Stanfield (2006) created a landscape disturbance index referred to as percent impervious
cover (PIC), to provide a single metric of the overall land cover in a
catchment that reflects forested, agriculture and urban lands. The PIC
was calculated by using a summed ranking of the land cover data
(land cover 28, OMNR, 1999) with the following rankings: urban
(0.2), intensive agriculture (0.1), rangeland (0.05), forest (0.01) and
water (0). Each of the landscape variables represents predictor variables
for subsequent model development designed to evaluate the deviation
from a “reference” condition.
A general linear model to predict the fish assemblage metric was developed from the catchment area, slope, PIC and BFI, and the squared
terms of each predictor variable (to account for potential curvilinear relationships). The model confirmed a threshold response at PIC ratings of
8–10 and a linear response below this threshold (Table 1). Because issues of autocorrelation can confound these types of models, Stanfield
and Kilgour (2006) confirmed the significance of the models by evaluating the PIC effect on the residuals of the model after accounting for the
primary landscape variables (BFI and slope). The absence of catchment
Table 1
Mean sum of the error (MSE) terms and model coefficients for the fish assemblage
metric (CA axis 1) (from Stanfield and Kilgour in, 2006). Numbers in parentheses
indicate the values for 2 and 3 standard deviations used to assess site condition.
MSE (2SD/3SD)
Constant
slope
slope2
BFIa
PICb
PIC2
0.8 (1.8/2.7)
− 1.63
− 0.24
0.03
− 0.02
0.48
− 0.02
a
b
Baseflow Index.
Percent Impervious Cover.
area in the fish assemblage model is likely a reflection of the strong correlation of slope and catchment area in this dataset (see Axis 2 of Fig. 1).
Finally, the relationship between the observed and predicted fish assemblage metric was similar for both the calibration and validation
data sets, thereby providing further support that the model was robust.
Evaluating the state of the fish assemblage metric
The condition (degree of impairment) at each site is determined by
comparing the observed state to a hindcasted predicted reference condition based on the GLM. This technique provides a benchmark reference
state for landscapes where truly reference conditions for all types of
streams do not exist. This approach hindcasts the fish assemblage metric
values for each site by applying the sum of the products of model coefficients and site landscape conditions, after setting forest cover to 100%
(PIC=1). For each of the 704 sites where fish were caught, the difference from hindcasted condition was calculated and compared to the
standard deviations (the square root of the MSE, 0.8 in Table 1) observed
from the model. This analysis was feasible due to the tight relationship
of the GLM predicted verses observed CA scores (MSE =0.804).
Following Kilgour et al. (1998) the condition of each site was determined by subtracting the differences in observed from predicted conditions. With this analysis, sites with observed fish assemblages within 2
SD in either direction of the predicted mean reference condition (i.e.,
b95%) are considered unimpaired, sites within 2–3 SD (>95% but
b99%) are likely impaired and sites >3 SD are considered impaired.
Sites with current conditions that are better than predicted in the reference state by >2 SD are categorized as “better than expected”; no sites
in this dataset met this criteria. In this manner, regardless of catchment
primary conditions, a standardized measure of site condition was calculated. For example, a site with a fish assemblage value of 4.7 and hindcasted community of 1.9 deviates by a score of 2.8. This site is
impaired because the difference is greater than 3 SD (2.7) from the
mean relationship, for sites with similar BFI and slope.
Translating site condition to a landscape perspective
The reporting process is applied to several spatial units typically
used by Canadian managers to report on broader landscape conditions.
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
The Great Lakes biodiversity blueprint project defined the first reporting level in the hierarchy: valley segments (Wichert et al., 2005). This
data (LIO, 2008) was derived using an automated GIS to demarcate
and attribute each valley segment, provided it was classified as Strahler
order 2 or higher. This limitation reflected both time and computer restraints and the scale at which this broad analysis supported decisionmaking. Boundaries for each segment were defined where either
stream order changed; there was a change (line intersection) in hydraulic conductivity (as determined from surficial geology), or there were intersections with water bodies greater than 10 ha in size. Hydraulic
conductivity boundaries reflected changes from poorly (1× 10− 5 to
1 × 10− 2 m/yr), moderately (1× 10− 2 to 1 × 101 m/yr) or well drained
(1× 101 to 1 × 107 m/yr) soil porosities, as described by the Ontario geological survey (1997). Segments were typically much longer than 1 km
in length and contained all minor tributaries (i.e., b2nd order). Upstream catchments are defined from the outlet of each segment and
are attributed with comparable datasets as were used in the fish assemblage GLM of Stanfield and Kilgour (2006) to generate predictions for
each segment. Measures of slope used in the predictions of segment condition differed slightly from those used in the development dataset, with
the former being measured along the entire length of a segment and the
latter being measured 100 m up and downstream of each site. Twentytwo percent of sampled sites were located on b2 Strahler stream order
reaches and have been combined with the larger valley segment summary results.
Summarizing data within a segment
Based on the recommendation of Wu and Li (2006), several approaches for categorizing the condition of each reporting area (e.g.,
mean, median, mode, best, worst, random, and the mean difference of
aggregated scores) were evaluated during a preliminary analysis
phase. This analysis determined that the mean condition was not appropriate for categorical data. All but the median and the modal
methods were subject to the effect of outliers. The modal value (most
frequently occurring) was not an effective means of categorizing segments with b2 sampled sites. Therefore, the median deviation from
expected was used to describe segment condition. Further, variability
in the number of sites available for each reporting area and in site condition combined to affect the confidence in interpretations of landscape
patterns. For each segment, confidence-ratings, based on a combination
of the congruence of ratings and the number of observations in a segment were developed (Table 2).
To compare segment predictions to the observed data, the fish assemblage model is applied to each segment based on the landscape conditions determined from the upstream catchment, as defined from the
outlet of each segment. In effect, this provides a comparison of the reliability of using landscape data alone to predict segment condition. In
conducting the analysis, it became apparent that a number of segments
where multiple observations were available demonstrated a longitudinal gradient in the site condition of the fish assemblage, from unimpaired to impaired. It is likely that this pattern indicates a situation
where local land use activities might be overwhelming larger scale processes. These segments received an additional categorization, as locations of potentially high priority areas for local restoration opportunities.
Summarizing data within larger reporting units
Managers from within the study area identified catchments as a
second level reporting unit. Catchments represent areas that have
specific properties to warrant management plans (e.g., drainage, biodiversity or land use). These moderate sized reporting areas include
all the drainage area upstream of an outlet to a mainstem tributary,
such that a catchment is always a subcomponent of a watershed. In
this study, a watershed is a drainage area upstream of the intersection
of the stream with the Lake Ontario shoreline. The quaternary and
199
Table 2
Decision criteria used to classify confidence ratings in each segment.
Number of
Site classifications
sites/segment (summary of site ranking by reporting areaa)
1
2
2
3
3
3
3
4
4
4
4
4
4
5 or more
5 or more
5 or more
–
2 identical
2 different
3 identical
2 identical, one site differs by 1 category
2 identical, one differs by 2+ categories
All different
4 identical
3 identical, one site class differs by 1 category
3 identical, one site class differs by 2+
categories
Even split (2 site class categories observed)
3 categories observed
4 categories observed
70–100% of site classes identical
50–70% site classes identical
b 50% identical
Proposed
confidence rating
Low
Medium
Low
High
Medium
Low
Low
High
High
Medium
Low
Low
Low
High
Medium
Low
a
Identical infers all sites are categorized similarly, all sites compared relative to the
hierarchy of reporting areas presented above, such that a site that differs by one category is either one higher or lower in the hierarchy.
tertiary watershed delineations represent medium and large scale
reporting units. Quaternary watersheds typically contain one or more
major stream systems and a number of smaller systems, while tertiary
watersheds contain several (3–5) quaternary watersheds that have similar geography and comparable hydrograph conditions (Cox, 1978). Tertiary watersheds combine several adjacent quaternary watersheds that
have similar geology and outlets (Cox, 1978). In this study area, data
were available for analysis of 21 quaternary and 3 tertiary watersheds.
The condition of each larger reporting area was determined using
a weighted sum approach. Area weighting is a standard statistical approach for summarizing condition of a reporting area when the sampling units (segments) vary in size (Cochran, 1977). This technique is
applied to the fish assemblage condition and confidence rankings for
larger reporting units (Eq. (1)). In Eq. (1), UnitScore represents the
score for any reporting unit (j); length indicates the total length of
stream for each category of WF (weighting factor) which is standardized by the total length of streams available for each reporting unit
(j).
UnitScore ¼ ∑ij ðlengthi WF i Þj =∑lengthj 100
ð1Þ
Table 3 contains the weighting factors (WF) for each category (i) of
effect size and confidence rankings. Ratings maximize the contrast between the categories, with a score of 1.0 being optimal (better than
expected or high confidence) and 0 or 0.1 being impaired or low confidence. For example, a reporting area with data for only three segments
of 1000 m of unimpaired, 1500 m likely impaired, and 2000 m impaired
Table 3
Weightings for summary of segment conditions and confidence ratings.
Segment category
Weighting factor
Better than expected
Unimpaired
Likely impaired
Impaired
Confidence rating
High
Medium
Low
1.0
0.7
0.4
0.0
1.0
0.5
0.1
200
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
lengths would have an overall rating of 29 (e.g., (1000*0.7+1500*0.4+
2000*0.0)/4500*100). If, for the same example dataset used above, segment confidence ratings were high for the unimpaired and the impaired
segments and low for the likely impaired segment, the overall confidence
rating for this reporting area would be 0.7 (i.e., (1000*1+1500*0.1+
2000*1)/4500).
With this approach, each reporting level larger than a segment has
the potential of scores from 0 to 100. Illustrating the relative rankings
of each reporting area on maps poses a challenge, since for clarity,
classification is required and like any classification system, selection
of the thresholds can influence interpretation of the results. Since
one of our objectives is to develop an approach that could be easily
understood by the public in support of watershed report cards, academic report card grades are applied: i.e., A >80, B 70–80, C 60–70,
D 50–60 and F b50.
Results
Rating segments
Rating of stream condition generated a similar number of segments in each of the categories of unimpaired, likely impaired and
impaired (Table 4). Categorized segments in closer proximity to
urban lands tended to be impaired and likely impaired; in contrast,
unimpaired sites tended to be located in more northern and eastern
areas (Fig. 2a). Moving in an easterly direction from the Greater Toronto Area (GTA), there was a zone of heterogeneity in the ratings
(see the inset of Fig. 2a). In these areas, segments in close proximity
contrasted greatly. For example, the Wilmot Creek in the eastern
part of the inset had all three categorizes of stream segments within
one small watershed.
The majority (63%) of the confidence ratings for segments were
low; the most common cause (55%) of this classification was due to
only one site record being available for the segment (Fig. 2b). Many
of the remaining low confidence ratings were the result of different
ratings between mainstem and tributary systems within one segment. This observation may erroneously deflate the confidence that
managers might have in the overall rating system used. Only 16% of
segments had high confidence ratings in the classification.
Nine segments (3%) had sites with three classes of conditions, indicating that they are likely under the influence of local land use activities (Fig. 2b). As an example, Salem Creek east of Brighton
(Fig. 2b, inset) had the greatest contrast in classifications of any segment, with seven unimpaired, three likely impaired and one impaired
site. The median classification for this segment was unimpaired. This
segment showed a change in condition that reflected both a longitudinal gradation as well as a clear local effect. Categorization of sites
changed from unimpaired above, impaired immediately below, and
likely impaired in a recovery zone downstream from a failed dam
that washed out the year prior to the field surveys reported here.
These findings supported the conjecture that indeed local land use activities were likely overwhelming landscape influences on this segment of stream. This example also highlights a major limitation of
using a valley segment classification system that does not recognize
Table 4
The number of segments with different impairment and confidence ratings for the fish
assemblage categorization of impairment based on the median class for all sites within
a segment.
Segment
Confidence in rating
Classification
High
Medium
Low
Total
Not impaired
Likely impaired
Impaired
Total
20
10
15
45
16
20
25
61
55
54
60
169
91
84
100
275
headwater systems as being independent from the mainstem reaches.
Salam Creek is considered one segment by the conservative criteria
applied in the ALIS dataset.
Evaluating the accuracy of the models with increasing size of reporting
areas
Regardless of the number of observations per segment, there was
a good correlation between observed and predicted fish assemblage
condition for segments rated as either unimpaired (95%), or impaired
(81%). Rarely (2%) was a segment rated as unimpaired when field
data indicated that it was not in this condition (Table 5). The model
was largely unable to predict sites rated as being in the intermediate
reporting class of likely impaired.
Fish assemblage data was sufficient to assess 48% of the length of
streams located within the 59 example catchments, although the
length of coverage in individual catchments ranged from 3 to 100%
(Fig. 2c). Since this study area had no segments identified as better
than expected, the maximum score attainable within a catchment
was 70 (or B). Only 16 of the catchments received this rating and
these tended to be located in the eastern part of the study area. In
contrast, western catchments tended to receive failing classifications;
this grade was the most common in the study area at 61% of catchments (Table 6). Despite the large dataset, only 20% of catchments
had high confidence ratings, while 47% were low.
Data are available for 27 watersheds (Fig. 2d) and 21 quaternary
watersheds (Fig. 2e; Table 6). As reporting area size increased, the
scores generally decreased. This occurred because in this study area,
it is rare to find large areas of streams in unimpaired condition.
With scores of: 37, 19, and 48 respectively, each of the west, central
and eastern tertiary watersheds are categorized as F.
Discussion
While there are many studies that summarize stream data across
one reporting scale (e.g. Riseng et al., 2010), there do not appear to
be any other studies that compare the effects of summarizing stream
condition at multiple levels in the hierarchy of reporting units. This
study demonstrated that there is good agreement between observed
and predicted stream fish assemblages at the segment scale, especially if the fish assemblage is impaired or unimpaired. This study determined that mapping hindcasted site data provides an effective means
of summarizing fish assemblage data across a broad geographic area.
Providing categorized site conditions and confidence rankings on all
maps offers a tool for interpreting the findings. However, smoothing
effects from summarizing findings to larger areas, the choice of mapping thresholds and scaling effects had a great affect on interpretation
of study findings. Finally, this assessment of the condition of stream
fish assemblages across the north shore tributaries of Lake Ontario
is consistent with the spatial trends in a similar study of Michigan
streams (Riseng et al., 2010).
Correlation between observed and predicted condition of segments
The strong correlation between site observations and segment conditions supports the suggestion by Seelbach et al. (1997) and Wang et
al. (2006) that stream segments represent the lowest level of reporting
unit with reliable predictability from landscape models. These findings
give confidence that segments categorized as undisturbed and disturbed are likely accurate and that segments categorized as likely impaired warrant additional analysis and more sampling in order to
truly ascertain stream condition. The poor agreement in the classification and predictions of likely impaired segments is likely the result of
these sites being located on the cusp of the threshold, where the
model response has the steepest slope. The unknown level of error
that results from combining the ratings from sites located on Strahler
order classification lower than 2 streams, must be acknowledged and
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
201
Fig. 2. Categorized Fish Assemblage conditions and confidence ratings in the classification by study area: a) segments, c) catchments, d) watersheds, e) quaternary watersheds.
Panel b shows confidence ratings of classifications for segments as well as segments identified as being likely under the influence of local impacts (dark blue segments). The
inset in panel b shows a close up of one locally impacted stream with the gradient in site conditions relative to a dam washout shown.
is more likely to be a factor in the middle ratings of stream condition. In
these areas, local conditions could easily have a greater effect on fish
communities than at sites that are above the threshold (>10 LDI) or
are in least disturbed areas (b4 LDI). Use of a more liberal classification
scheme, for example adding another category for 1 SD sites, generated a
greater contrast in classes, but lower agreement (23%) between the
202
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
Table 5
Comparison of observed and predicted fish assemblage condition within a segment
(bolded numbers indicate agreement).
Observed
Predicted
Median category
Unimpaired
Likely impaired
Impaired
Unimpaired
Likely impaired
Impaired
54
58
9
0
55
4
3
68
63
predicted and observed segment condition. Scaling up data to larger
spatial areas, however, poses a series of challenges to understanding
scales of influence.
Scale of reporting influences perception of condition
That stream condition tended to decline as scaling size increased is
indicative of the challenges associated with scaling influences, although
it is difficult to disentangle true effects from perception. When scaling
up the information on stream condition, there are issues associated
with the scales of reporting, homogenization or smoothing effects, classification criteria, aggregation effects (Jelinski and Wu, 1996; Moody
and Woodcock, 1995; Schumm and Lichy, 1965) and the nature of the
river network (Vannote et al., 1980). In this study, there was a strong
tendency for stream condition to degrade as size increased. The longer
stream segments that tend to occur in the lower reaches of watersheds
produce smoothing effects in summary scores of larger reporting areas
because of the “weight” of their greater lengths to the overall score.
There were also instances where data were available from b2nd Strahler order streams that differed in condition compared to the main segments (either better or worse), adding another source of aggregation
error and likely deflating the magnitude of the true smoothing effect
that occurs with increasing reporting area river sizes. Because watershed size and geology have a primary influence on the hydrologic
cycle that governs conditions at a site on a stream, and watersheds
vary in size, it is not feasible to address smoothing effects by standardizing habitat unit size, as is widely used in terrestrial studies (O'Neill et
al., 1996; Townshend and Justice, 1990; Woodcock and Strahler, 1987).
The decision to apply the weighted length correction equation meant
that smoothing effects were inevitable, since catchment sizes varied.
There was also a confounding issue: in this study area, unimpaired
sites tended to be located in smaller streams, so that their contribution
Table 6
Summary of the number of each larger study area for each category of fish assemblage
condition and confidence rating. Criteria to summarize data are defined in Table 3.
Catchment
scores
Catchments
B (70–80)
C (60–70)
D (50–60)
F (b50)
B (70–80)
C (60–70)
D (50–60)
F (b50)
Medium
(33–66)
8
3
2
1
1
3
17
12
Quaternary watersheds
Low
Medium
(b 33)
(33–66)
1
0
1
1
0
2
5
9
Implications of findings to efforts to classify stream conditions across
landscapes
Patterns are most apparent at the smallest reporting units (e.g., segment and catchment); inclusion of confidence rankings helped provide
context for these interpretations by providing a single “score” that helps
users understand the limitations of the data. Hierarchy theory predicts
these findings, as more connected areas are larger and slower to respond than lower levels, particularly for nested hierarchies (Wu and
Li, 2006). These findings have far-reaching implications for efforts to
classify stream habitats into hierarchical reporting units beyond this
habitat unit size. It implies that until these scaling effects are addressed,
summarizing stream data from multiple segments into larger scale ecological units is unlikely to provide a meaningful measure of the condition of all streams within each reporting area. Chu et al. (2010) also
found poor correlations between stream temperatures and attributes
of larger ecological reporting units.
Aggradations and scoring systems matter
Confidence ratings
Low
(b 33)
to larger reporting areas was lower. As reporting area increases, the differences between reporting areas in the number of contributing subareas, and the differences in areas between reporting areas magnify
these scaling effects, making comparisons between the largest reporting areas (e.g., quaternary and tertiary watersheds) less meaningful.
In this study, ratings of each reporting area did not consider Strahler
stream order of each of the contributing segments, as the intention was
to provide an unbiased indicator of system wide condition. In other
words, all segments >2nd order were treated equally. This unbiased
view and the choice of our indicator as a fish assemblage metric represent a different perspective on stream condition than has been reported
in several other studies of these tributaries (Jones and Stockwell, 1995;
Stanfield and Jones, 2003; Stanfield et al., 2006; Stewart et al., 1999). In
these studies, the focus was on migratory salmonids, and in each case
the authors have remarked on the generally high production of salmonids found in many of these systems. For example, Jones and Stockwell
(1995) identified Wilmot Creek as a premiere trout stream in Ontario.
The mainstem supports this classification with a good condition (B) rating, but its tributary catchments are impaired or likely impaired, pulling
the overall watershed condition down from a B to a D. Our more holistic
interpretation highlights the value of the mainstem fishery, while also
identifying the potential concerns in the smaller catchments.
It may be that future studies will identify a statistical process to
address the impact of scaling and smoothing effects and ways to incorporate segment position in the river network. In the meantime, I
agree with Tobler (1989) and Jelinski and Wu (1996) that for tributaries, where area units (catchment sizes) are inherently different,
more emphasis should be placed on mapping approaches that illustrate the summarized data in ways that enable visual interpretations
of scaling effects. For this study area, this means the site categorizations should be included on all maps.
Watersheds
High
(> 66)
Low
(b33)
5
0
0
7
1
2
1
3
1
1
9
6
Tertiary watersheds
Low
Medium
(b33)
(33–66)
0
0
0
0
0
0
3
0
High
(> 66)
0
0
0
2
Medium
(33–66)
High
(>66)
1
0
0
2
High
(>66)
0
0
0
0
1: ratings based on the weighted sum (see Table 3) of the length of categorized
segments in each reporting area.
This study followed the advice of Fotheringham (1989) in the selection of the fish assemblage metric that is both sensitive to changes
in land use/land cover and is objectively based. Use of the coefficients
for each species in the fish assemblage metric provides an opportunity for trend analysis on subsequent datasets. In fact, in Ontario these
approaches have begun to resonate with managers, being recently
referenced in two fisheries management plans (OMNR, 2009; TRCA,
2008) for monitoring change in stream condition over time. Jelinski
and Wu (1996) suggest that given the multitude of spatial and statistical issues associated with reporting on ecosystem condition, measuring changes in condition on one spatial scale over time has the
greatest potential for success in linking cause to effect. However, the
measurement of trends in time and space is still required to define
the reference state from which to measure change.
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
Pearse et al. (1985) warned that our perception of change-inexpected condition “is short sighted at best” and that “the inability
to objectively evaluate the degree of change in habitat is largely due
to a lack of knowledge of the reference condition”. Xu et al. (2005)
faced similar challenges in their attempt to classify the health of Italian
lakes. In their study, field data were divided into five equal categories
with equal numbers in each grouping of low (0) to high (100). This ensured that the final classification would include the full range of scores
of sites, but only for the existing state. The approach used in the present
study assumes the existing state is to some degree degraded and applies
the hindcasting approach to determine the reference condition. This approach provides a benchmark from which to measure future change in
condition.
The scoring system used here is likely to result in biased interpretations of conditions in comparisons between higher levels in the hierarchy of reporting areas (e.g., quaternary and tertiary watersheds),
although the scheme may be appropriate for evaluating the individual condition of lower level reporting areas (e.g., watershed and catchment). The chosen scoring system is somewhat arbitrary and is not as
important as the base data that contributes to the rating in each
reporting area. This study provides a foundation for rating stream condition based on a standard reference condition that is more important
that the actual thresholds for each “grade”. For example, choosing a
more conservative rating scheme for a geographic area that is dominated by urban land use (e.g., scoring F as b40) would result in a better
overall grading of stream condition (less failed categorizes for example), but, provided the same base data are used, would not compromise
the scientific credibility of the data. Rather, such a decision might provide a more realistic evaluation of stream condition for an urbanized
landscape and enable consistent reporting to the public of watershed
health (Briggs et al., 2003).
Finally, providing the base level mapping results regardless of the
level in the hierarchy of reporting area provides an opportunity to
identify potential areas requiring management action that may not
be obvious otherwise.
Other factors to consider and science needs
A core component and therefore limitation of this study is the use of
modeling results that are based on coarse metrics of landscape properties (e.g., PIC and BFI), and analytical approaches that do not consider
proximity and hierarchical relationships of streams. In the future, as
more refined data layers and new approaches for analyzing the data
emerge, model predictions are likely to improve. Future analysis of
the state of the streams in this area will likely require a reanalysis of
the condition of each site, while still using the same metric of fish assemblage. Reporting agencies have already applied the CA coefficients
to recent fish catches as a means of tracking changes in community assemblages over time (Christine Tu, Toronto Region Conservation Authority, pers. comm.).
A conceptual understanding of the effects of hierarchical scaling is
well-established (Fausch et al., 2002; Hynes, 1970; Imhof et al., 1996;
Poff, 1997; Schlosser, 1991 for examples). Several papers demonstrate
the pathways of influence at various spatial scales and levels in the
river network (see Fisher et al., 1998; Stanfield et al., 2006; Wang et
al., 2003; Zorn and Wiley, 2006). However, scaling and autocorrelation
challenges continue to be problematic for riverine ecologists because of
the intertwined issue that watershed areas vary and rivers are organized in a hierarchical network. There continues to be an urgent need
for new analytical methods to tease out these influences on streams
(see Wu, 1999 for an earlier plea). If this approach is not feasible, as suggested by Openshaw et al., (1987; also see Fotheringham, 1989; Jelinski
and Wu, 1996) then the alternative is to develop and apply a framework
around which consistent application of visual interpretations of maps
across the landscape can be made.
203
Practical application of results to stream management and reporting
This study suggests that streams on the Canadian side of Lake Ontario show progressive degradation in response to intensification of
land use. The patterns observed here agree with our knowledge of
the effects of increasing development on fish communities (Leopold,
1968; Stanfield and Kilgour, 2006; Wang et al., 2003) and are consistent with findings of Riseng et al. (2010) from Michigan. This study
suggests that the impacts result from incremental impairment of
streams within a watershed. Watersheds are impaired incrementally
segment by segment and catchment by catchment, until the entire
watershed is affected. Reversal of this creeping crisis requires implementation of a watershed management strategy that recognizes this
incremental process, can identify priority problem areas, and identifies actions that address process level issues regardless of the current status of “significant fisheries”. Addressing this challenge will
require a shift in philosophy and actions by resource managers and
the public to recognize fishless and currently degraded streams as
being important to the overall productive capacity of the entire watershed, and then protect/restore them. This direction is a recommendation of the Department of Fisheries and Oceans fish habitat policy
(Pearse et al., 1985). The presented analysis provides graphic evidence of the implications of not protecting all components of the
stream network. These results support the argument for broader application of the fish habitat policy to all sizes and types of streams
in Canada and for similar policies in other jurisdictions.
Since impacts to streams occur in a nested and incremental way,
so too should restoration efforts. These findings provide guidance as
to which actions are likely to be most beneficial for each stream category type. As suggested by Wang et al. (2006), segments identified as
unimpaired should be the focus for management action that emphasizes protection. In these areas, instream activities such as cover enhancement or bank stabilization would only be beneficial where
local areas are a concern. The nine segments identified as potentially
being under the influence of local land use impacts represent priority
areas of further study in order to determine if and what type of local
stewardship work could reverse the trend of impacts on fish assemblages and habitat. The stream-segments categorized as likely impaired should be the targets of further investigation in order to
confirm status and to identify potential problem areas/types and if
present to initiate stewardship activities to rectify problems. Managers should not expect to restore fish assemblages in segments and
catchments that are in a consistent degraded state by only carrying
out locally based stewardship work. Rather, these areas will require
integrated watershed planning and management to address major
concerns in the watershed before instream habitat works are likely
to result in significant improvements to fish assemblages. These results confirm the need for stronger strategic planning and management for every stream and at all levels of the hierarchy of a river
network, as recommended by Imhof et al., 1996.
Resource managers with large geographic area mandates could incorporate an adaptive monitoring strategy that initially targets all
segments with one sample site. Later, additional sampling could target areas categorized as likely impaired from the first sample. Such
a design takes advantage of the high predictive power of the impaired
and unimpaired condition and optimizes the potential to identify locations where local land use is having its greatest effect (see above).
Conclusions and recommendations
That most watersheds in this study demonstrated at least some
segments in a degraded condition is evidence that the incremental
loss of fish habitat described by Pearse et al. (1985) is still occurring
in this area. This study demonstrates how the results of adaptive sampling of segments can help managers direct limited resources for optimal results. That data from only one site can be sufficient to identify
204
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
whether a segment is categorized as either unimpaired or impaired
will be welcome news to managers struggling with the logistics of
conducting large scale environmental monitoring. Additionally, if a
sample determines a segment as being likely impaired, or if a manager suspects local land use is unduly influencing stream condition, this
study demonstrates that having at least three sites of data that cover a
large longitudinal gradient within a stream segment can help identify
potential areas where local restoration activities are likely to be most
effective. Results suggest that the scale at which the stream data is
collected (e.g., site or segment) is the best scale at which to present
the findings and that as reporting area increases, a variety of statistical and spatial issues combine to reduce the reliability of comparisons
in condition between spatial areas. Therefore, managers and researchers should consider the implications of the scoring systems
used in Watershed Report Cards or state of the ecosystem reports
that are susceptible to smoothing effects. Implementing an adaptive
management framework that classifies stream conditions using unbiased methods will hopefully reverse the stream degradation measured in this area and across the Great Lakes basin during the next
wave of land use intensification.
Acknowledgments
This paper is dedicated to the many field technicians and biologists
who collected the data that made this analysis possible, specifically,
Jeff Anderson, Jon Clayton, Marc Desjardins, Scott Jarvie, Ian Kelsey
and the crews from the Salmonid Ecology Unit. Rebecca Astolfo and
Debbie Irvine-Alger provided GIS and mapping support. Scott Gibson
helped with the data analysis. Al Dextrase, Nick Jones, Nigel Lester,
Bob Metcalfe, and Scott Reid, Bob Bergmann, Doug Dodge, Tim Haxton,
Bruce Kilgour, Danijela Puric-Mladenovic and Christine Tu provided
helpful comments on an earlier draft of this report. Two anonymous reviewers greatly improved the manuscript and Anne Dumbrille provided
editorial comments on the final draft. The Ministry of Natural Resources,
Water Resources Information Project and Environment Canada through
the Water Use and Supply project provided funding for this project. The
Southern Ontario Stream Monitoring and Research Team provided data
and support for the project. Cheryl Lewis, Scott Christilaw, Wendy Leger
and Silvia Strobl provided administrative support.
References
Allen, R.F.H., O'Neill, R.V., Hoekstra, T.W., 1984. Interlevel relations in ecological research and management. USDA Forest Service Gen. Tech. Rep. RM-110. Rocky
Mountain Forest and Range Experiment Station.
Bailey, D., Billeter, R., Aviron, S., Schweiger, O., Herzog, F., 2007. The influence of thematic resolution on metric selection for biodiversity monitoring in agricultural
landscapes. Landsc. Ecol. 22, 461–473.
Briggs, T., Craig, C., Larson, P., Maaskant, K., Quinlan, C., Wilcox, I., 2003. Watershed
reporting: improving public access to information. Available from: http:/www.
conservation-ontario.on.ca/projects/pdf/reports (Accessed May 2010).
Chu, C., Jones, N.E., Allin, L., 2010. Linking the thermal regimes of streams in the Great
Lakes Basin, Ontario, to landscape and climate variables. River Res. Appl. 26,
221–241.
Cochran, W.G., 1977. Sampling Techniques. John Wiley and Sons, New York.
Cox, E.T., 1978. Counts and Measurements of Ontario Lakes: Watershed Unit Summaries Based on Maps of Various Scales by Watershed Unit. Ontario Ministry of Natural Resources.
Dungan, J.L., Perry, J.N., Dale, M.R.T., Legendre, P., Citron-Pousty, S., Fortin, M.-J.,
Jakomulska, A., Miriti, M., Rosenberg, M.S., 2002. A balanced view of scale in spatial
statistical analysis. Ecography 25, 626–640.
Dunne, T., Leopold, L.B., 1978. Water in Environmental Planning. Freeman, New York.
Fausch, K.D., Torgensen, C.E., Baxter, C.V., Li, H.W., 2002. Landscapes to riverscapes:
bridging the gap between research and conservation of stream fishes. BioScience
52, 483–498.
Fisher, S.G., Grimm, N.B., Marti, E., Gomez, R., 1998. Hierarchy, spatial configuration,
and nutrient cycling in a desert stream. Aust. J. Ecol. 23, 41–52.
Fortin, M.J., Dale, M., 2005. Spatial Analysis: A Guide to Ecologists. Cambridge University Press, Cambridge, UK. 365 pp.
Fotheringham, A.S., 1989. Scale-independent spatial analysis. In: Goodchild, M.E., Gopal,
S. (Eds.), Accuracy of Spatial Databases. Taylor and Francis, London, pp. 221–228.
Frissell, C.A., Liss, W.J., Warren, C.E., Hurley, M.D., 1986. A hierarchical framework for
streams habitat classification: viewing streams in a watershed context. Environ.
Manage. 10, 199–214.
GLFC (Great Lakes Fishery Commission, Editor). 2007. A joint strategic plan for management of Great Lakes fisheries (adopted in 1997 and supersedes 1981 original).
Great Lakes Fish. Comm. Misc. Publ. 2007-01. Available at http://www.glfc.org/
fishmgmt/jsp97.pdf [accessed - Dec 21, 2010].
Hynes, H.B.N., 1970. The Ecology of Running Waters. University of Toronto Press, Toronto, Can.
Imhof, J.G., Fitzgibbon, J., Annable, W.K., 1996. A hierarchical evaluation system for
characterizing watershed ecosystems for fish habitat. Can. J. Fish. Aquat. Sci.
(Suppl 1), 312–326.
Jelinski, D.E., Wu, J., 1996. The modifiable areal unit problem and implications for landscape ecology. Landsc. Ecol. 11, 129–140.
Jensen, M.E., Bourgeron, P.S., 1995. A Guidebook for Integrated Ecological Assessments.
Springer-Verlag, New York, NY.
Jones, M.L., Stockwell, J.D., 1995. A rapid assessment procedure for the enumeration of
salmonine populations in streams. N. Am. J. Fish. Manage. 15, 551–562.
Kilgour, B.W., Stanfield, L.W., 2006. Hindcasting reference conditions in streams. In:
Hughes, R.M., Wang, L., Seelbach, P.W. (Eds.), Landscape Influences on Stream Habitats and Biological Assemblages: American Fisheries Society, Symposium, 48. Bethesda Maryland, pp. 623–639.
Kilgour, B.W., Somers, K.M., Matthews, D.E., 1998. Using the normal range as a criterion
for ecological significance in environmental monitoring and assessment. Ecoscience
5, 542–555.
Legendre, P., Legendre, L., 1998. Numerical Ecology. Elsevier Science Amsterdam, The
Netherlands.
Leopold, L.B., 1968. Hydrology for urban land planning — a guidebook on the hydrologic effects of urban land use. U.S. Geological Survey, Circular 554, Washington,
DC. 18 pp.
Levin, S.A., 1992. The problem of pattern and scale in ecology. Ecology 73, 1943–1967.
Lewis, C.A., Lester, N.P., Bradshaw, A.D., Fitzgibbon, J.E., Fuller, K., Hakanson, L., 1996.
Considerations of scale in habitat conservation and restoration. Can. J. Fish.
Aquat. Sci. (Suppl. 1), 440–445.
LIO, 2008. Aquatic Landscape Information System Dataset. Land Information Ontario,
managed by Ontario Ministry of Natural Resources, Peterborough, ON.
MacArthur, R.H., 1972. Geographical Ecology. Princeton University Press, Princeton, NJ.
Maidment, D.R., 2002. Arc Hydro GIS for Water Resources. ESRI Press, Redlands CA.
Maxwell, J.R., Edwards, C.J., Jensen, M.E., Paustian, S.J., Parrott, H., Hill, D.M., 1995. A hierarchical framework of aquatic ecological units in North America (Nearctic Zone).
United States Department of Agriculture, Forest Service, North Central Forest Experimental Station, General Technical Report NC-176.
Meentemeyer, V., 1989. Geographical perspectives of space, time, and scale. Landsc.
Ecol. 3, 163–173.
Moody, A., Woodcock, C.E., 1995. The influence of scale and the spatial characteristics
of landscapes on land-cover mapping using remote sensing. Landsc. Ecol. 10,
363–379.
National Research Council (NRC), 1992. Restoration of Aquatic Ecosystems. National
Academy Press, Washington, DC.
O'Neill, R.Y., Krummel, J.R., Gardner, R.H., Sugihara, G., Jackson, B., DeAngelis, D.L.,
Milne, B.T., Turner, M.G., Zygmunt, B., Christensen, S.W., Dale, Y.H., Graham, R.L.,
1988. Indices of landscape pattern. Landsc. Ecol. 1, 153–162.
O'Neill, R.V., Hunsaker, C.T., Timmins, S.P., Jackson, B.L., Jones, K.B., Riitters, K.H.,
Wickham, J.D., 1996. Scale problems in reporting landscape pattern at the regional
scale. Landsc. Ecol. 11, 169–180.
Ontario Ministry of Natural Resources (OMNR), 1999. LANDSAT-1 MSS. Inventory,
Monitoring and Assessment Section, Peterborough, ON.
Ontario Ministry of Natural Resources (OMNR), 2009. Wilmot Creek Fisheries Management Plan. Peterborough District, Peterborough, ON.
Openshaw, S., Charlton, M., Wymer, C., Craft, A., Mark, I., 1987. Geographical analysis
machine for the automated analysis of point data sets. Int. J. Geogr. Inf. Syst. 1,
35–51.
Pearse, P.H., Bertrand, F., MacLaren, J.W., 1985. Currents of Change; Final Report: Inquiry on Federal Water Policy. Environment Canada. 066214189x.
Piggott, A., Brown, D., Moin, S., 2002. Calculating a groundwater legend for existing
geological mapping data. Proceedings of the 55th Canadian Geotechnical and 3rd
Joint IAH-CNC and CGS Groundwater Specialty Conference. Southern Ontario Section of the Canadian Geotechnical Society, pp. 863–871.
Poff, N.L., 1997. Landscape filters and species traits: towards mechanistic understanding and prediction in stream. J. N. Am. Benthol.l Soc. 16, 391–409.
Riseng, C.M., Wiley, M.J., Seelbach, P.W., Stevenson, R.J., 2010. An ecological assessment of
Great Lakes tributaries in the Michigan Peninsulas. J. Great Lakes Res. 36, 505–519.
Schlosser, I.J., 1991. Stream fish ecology: a landscape perspective. BioScience 41,
704–712.
Schumm, S.A., Lichy, R.W., 1965. Time, space, and causality in geomorphology. Am. J.
Sci. 263, 110–119.
Seelbach, P., Wiley, W.M.J., Kotanchik, J.C., Baker, M.E., 1997. A landscape based ecological classification system for river valley segments in Lower Michigan. Michigan
Department of Natural Resources Fisheries Technical Report No.2036.
Stanfield, L.W., 2007. Ontario Stream Assessment Protocol. Ontario Ministry of Natural
Resources, internal publication, Peterborough, Ont v7.0.
Stanfield, L.W., Jackson, D.A., 2011. Understanding the factors that influence headwater
stream flows in response to storm events. J. Amer. Water Res. 47, 315–336.
Stanfield, L.W., Jones, M.L., 2003. Factors influencing rearing success of Atlantic salmon
stocked as fry and parr in Lake Ontario tributaries. N. Am. J. Fish. Manage. 23,
1175–1183.
L.W. Stanfield / Journal of Great Lakes Research 38 (2012) 196–205
Stanfield, L.W., Kilgour, B.W., 2006. Effects of percent impervious cover on fish and benthos
assemblages and in-stream habitats in Lake Ontario Tributaries. In: Hughes, R.M., Wang,
L., Seelbach, P.W. (Eds.), Landscape Influences on Stream Habitats and Biological Assemblages: American Fisheries Society, Symposium, 48. Bethesda Maryland, pp. 577–599.
Stanfield, L.W., Gibson, S.F., Borwick, J.A., 2006. Using a landscape approach to identify
the distribution and production of tributary habitats for juvenile salmonids in Lake
Ontario. Tributaries. In: Hughes, R.M., Wang, L., Seelbach, P.W. (Eds.), Landscape
Influences on Stream Habitats and Biological Assemblages: American Fisheries Society, Symposium, 48. Bethesda Maryland, pp. 607–621.
Stewart, T.J., Lange, R.E., Orsatti, S.D., Schneider, C.P., Mathers, A., Daniels, M.E., 1999.
Fish-community objectives for Lake Ontario. Great Lakes Fishery Commission Special Publication 99-1. Ann Arbor, Michigan.
ter Braak, C.J.F., 1992. CANOCO — a FORTRAN program for canonical assemblage ordination. Microcomputer Power, Ithaca, NY.
Tobler, W.R., 1970. A computer movie simulating urban growth in the Detroit region.
Econ. Geogr. 46, 234–240 (Suppl).
Tobler, W.R., 1989. Frame independent spatial analysis. In: Goodchild, M.F., Gopal, S.
(Eds.), Accuracy of Spatial Databases. Taylor and Francis, London, pp. 115–122.
Townshend, J.R.G., Justice, C.O., 1990. The spatial variation of vegetation changes at
very coarse scales. Int. J. Remote. Sens. 11, 149–157.
TRCA, 2008. Rouge River Watershed Based Fisheries Management Plan Toronto Region
Conservation Authority, Downsview, ON.
Vannote, R.L., Minshall, G.W., Cummins, K.W., Sedell, J.R., Cushing, C.E., 1980. The river
continuum concept. Can. J. Fish. Aquat. Sci. 37, 130–137.
Wang, L., Lyons, J., Rasmussen, P., Seelbach, P., Simon, T., Wiley, M., Kanehl, P., Baker, E.,
Niemela, S., Stewart, P.M., 2003. Watershed, reach and riparian influences on
stream fish assemblages in the Northern Lakes and Forest Ecoregion, U.S.A. Can.
J. Fish. Aquat. Sci. 60, 491–505.
Wang, L., Seelbach, P.W., Hughes, R.M., 2006. Introduction to landscape influences
on stream habitats and biological assemblages. In: Hughes, R.M., Wang, L.,
205
Seelbach, P.W. (Eds.), Landscape Influences on Stream Habitats and
Biological Assemblages: American Fisheries Society, Symposium, 48. Bethesda
Maryland, pp. 1–24.
Wichert, G.A., Brodribb, K.E., Henson, B.L., Phair, C., 2005. Great Lakes Conservation
Blueprint for Aquatic Biodiversity, Volume 1. Nature Conservancy of Canada and
Ontario Ministry of Natural Resources. 86 pp.
Wiley, M.J., Kohler, S.L., Seelbach, P.W., 1997. Reconciling landscape and local views of
aquatic communities: lessons from Michigan trout streams. Freshw. Biol. 37,
133–148.
Woodcock, C.E., Strahler, A.H., 1987. The factor of scale in remote sensing. Remote.
Sens. Environ. 21, 311–332.
Wu, J., 1999. Hierarchy and scaling: extrapolating information along a scaling ladder.
Can. J. Remote. Sens. 25, 367–380.
Wu, J., 2004. Effects of changing scale on landscape pattern analysis: scaling relations.
Landsc. Ecol. 19, 125–138.
Wu, J., Hobbs, R., 2002. Key issues and research priorities in landscape ecology: an idiosyncratic synthesis. Landsc. Ecol. 17, 355–365.
Wu, J., Li, H., 2006. Concepts of scale and scaling. In: Wu, J., Jones, K., Li, H., Loucks, O.L.
(Eds.), Scaling and Uncertainty Analysis in Ecology: Methods and Applications.
NLD, Dondrecht, pp. 3–16.
Wu, J., Loucks, O.L., 1995. From balance-of-nature to hierarchical patch dynamics: a
paradigm shift in ecology. Q. Rev. Biol. 70, 439–466.
Xu, F.L., Zhao, Z.Y., Zhan, W., Zhao, S.S., Dawson, R.W., Shu, T., 2005. An ecosystem index
methodology (EHIM) for lake ecosystem health assessment. Ecol. Model. 188,
327–339.
Zorn, T.G., Wiley, M.J., 2006. Influence of landscape characteristics on local habitat and
fish biomass in stream of Michigan's Lower Peninsula. In: Hughes, R.M., Wang, L.,
Seelbach, P.W. (Eds.), Landscape Influences on Stream Habitats and Biological
Assemblages: American Fisheries Society, Symposium, 48. Bethesda Maryland,
pp. 375–393.