Content uploaded by Yolanda Wiersma
Author content
All content in this area was uploaded by Yolanda Wiersma on Aug 28, 2019
Content may be subject to copyright.
Effects of errors in range maps on estimates
of historical species richness of mammals in
Canadian national parks
Lucas D. Habib, Yolanda F. Wiersma* and Thomas D. Nudds Department of Zoology,
University of Guelph, Guelph, Ontario, N1G 2W1, Canada
Abstract
Aim Tests for faunal relaxation in reserves, particularly for mammals, have relied on
comparisons of current species richness with estimates of species richness derived from
historical range maps. However, any range map reflects the extent of occurrence of
species and not necessarily the area of occupancy. Thus, estimates of historical species
richness might be prone to error introduced by Ôfalse positivesÕ, that is, a species might
be considered to have been present in locations where it actually was not. The effect of
such Ôfalse positivesÕcould bias statistical tests of faunal relaxation to type I error, and
result in estimates of the extent of faunal relaxation in reserves greater than was
actually the case. We evaluated the potential for errors in historical range maps to
generate inflated estimates of historical species richness of mammals at sites that are
reserves today.
Location Canadian national parks in the Canadian portion of the Alleghenian–Illinoian
mammal province in south-eastern Canada (the maritime region and parts of southern
Que
´bec, Ontario and Manitoba).
Methods The effect of varying levels of error in range maps on estimates of historical
species richness was tested using geographical information systems (GIS)-based statis-
tical sampling of simulated historical ranges. SpeciesÕareas of occupancy were simulated
to be only 25%, 75% and 95% of published historical species ranges. For each reserve,
estimates of historical species richness from these simulated species ranges were then
compared with similar, previously published estimates of richness based on published
historical species ranges.
Results Previous estimates of historical species richness for reserves were inversely and
linearly related to the degree of inaccuracy of species ranges. If species ranges were, on
average, 5% smaller than the accepted ranges, then estimates of historical species
richness agreed with previous estimates in c. 90% of cases. However, if historical
ranges were, on average, 25% smaller than those used in previous analyses, then
previous historical estimates of species richness may be overestimates in c. 40% of
cases.
Main conclusions Estimates of the extent of faunal relaxation in reserves that use
historical range maps to quantify past species richness appear to be sensitive to even
small errors in the degree to which range maps may overestimate Ôarea of occupancyÕ.
Keywords
Canadian national parks, historical mammal distributions, faunal relaxation, range
maps.
*Correspondence: Department of Zoology, University of Guelph, Guelph, Ontario, N1G 2W1, Canada. E-mail: ywiersma@uoguelph.ca
Journal of Biogeography, 30, 375–380
2003 Blackwell Publishing Ltd
INTRODUCTION
Terrestrial reserves have been shown to undergo faunal
relaxation consistent with the principles of island biogeog-
raphy. Small parks that have become isolated from their
surrounding habitat matrix have been shown to have lost
more of their historical complement of disturbance-sensitive
mammals than large parks that are surrounded by intact
habitat (Glenn & Nudds, 1989; Newmark, 1995; Gurd &
Nudds, 1999; Brashares et al., 2001; Gurd et al., 2001;
Wiersma & Nudds, 2001). Tests for extinctions of mammals
from Canadian reserves have used two approaches to esti-
mate historical baselines of species richness. Species richness
has been estimated from species–area curves generated from
historical range maps assumed to represent species ranges
prior to widespread European settlement (e.g. Glenn &
Nudds, 1989; Gurd & Nudds, 1999). Species richness has
also been directly sampled from these maps using geo-
graphical information systems (GIS) (e.g. Wiersma &
Nudds, 2001). Estimates of the numbers of species that
might have been present in reserves prior to insularization
were compared with current species richness (Glenn &
Nudds, 1989; Gurd & Nudds, 1999; Gurd et al., 2001;
Wiersma & Nudds, 2001) to estimate the extent of faunal
relaxation that each park had undergone.
Estimates of past species richness from historical range
maps may be prone to error because of inaccuracy in those
maps. Range maps are typically derived from a series of point
observations and delineate the Ôextent of occurrenceÕof a
species (van Jaarsveld et al., 1998). Different methods may be
used to extrapolate range boundaries (extent of occurrence)
from observations of species (areas of occupancy), giving very
different impressions of the Ôextent of occurrenceÕ. Authors of
original range maps, for instance, may have had different
confidence in the locations of observations on which they
based their maps, and delineated species historical ranges
accordingly.
For example, the mammal range maps in Banfield (1974)
used previously to test for faunal relaxation in Canadian parks
(Glenn & Nudds, 1989; Gurd & Nudds, 1999; Gurd et al.,
2001; Wiersma & Nudds, 2001) were based on observations
and collected specimens documented in records maintained in
museums. These kinds of data may reflect favoured transport
routes, hunting areas or collection localities (Lawes & Piper,
1998). Further, because historical records are typically fewer,
and historical estimates of species ranges thus cruder, the
discrepancy between the Ôextent of occurrenceÕof a species and
the actual Ôarea of occupancyÕ(van Jaarsveld et al., 1998) may
be even greater for historical data than it is for current distri-
bution data.
Thus, Robinson & Quinn (1992) cautioned that species
might be designated as historically present in an area that is
currently a reserve, when they actually were not (Fig. 1) and
the effect of such Ôfalse positivesÕhas implications for tests of
faunal relaxation. If a reserve was never as species rich as
assumed, then formal statistical tests for faunal relaxation
would be biased to type I error, that is, to detecting relax-
ation when it has not actually happened (Gurd & Nudds,
1999; Wiersma & Nudds, 2001). This criticism has intuitive
appeal, but we are unaware of any formal evaluation of how
poor the match between Ôextent of occurrenceÕand Ôarea of
occupancyÕwould have to be before we might consider sta-
tistical tests for faunal relaxation based on historical data to
be unreliable.
METHODS
We evaluated the effect of potential inaccuracies in
historical ranges of mammals on estimates of historical
species richness for a subset of mammals (see below for
selection criteria) identified as historically present by
Wiersma & Nudds (2001) in Canadian national parks. For
each species, we simulated reductions in Ôarea of
occupancyÕwithin the historical ranges (extents of occur-
rence) by randomly placing ÔholesÕ(simulated unoccupied
areas) within each range. New estimates of historical spe-
cies richness for each park were derived using the
simulated ranges following methods of Wiersma & Nudds
(2001), wherein GIS was used to count directly the number
of species historically present in areas that are currently
national parks. The new estimates of historical species
richness were compared with Wiersma & NuddsÕ(2001)
(a) (b)
Figure 1 Two possible ways to connect the
same set of points to delineate an historical
species range with different implications for
assessment of historical presence of a species
in a reserve. (a) The reserve (stippled)
intersects with the Ôextent of occurrenceÕof
the range; the species is recorded as historic-
ally present in the reserve. (b) The reserve
intersects the Ôextent of occurrenceÕof the
range, but not the actual Ôarea of occupancyÕ
(shaded); the species was not actually present
in the reserve.
2003 Blackwell Publishing Ltd, Journal of Biogeography,30, 375–380
376 L. D. Habib et al.
previous estimates of historical species richness to test how
robust the estimates were to the simulated inaccuracies in
the historical ranges.
We used maps digitized from historical distributions of
native, terrestrial mammals in Canada depicted in Banfield
(1974) to generate simulated historical ranges in a GIS.
Analysis was restricted to disturbance-sensitive mammal
species (defined by Glenn & Nudds, 1989, sensu Humphreys
& Kitchener, 1982) in the Canadian portion of the Alle-
ghenian-Illinoian (AI) mammal province (Fig. 2) (Hagmeier,
1966), the region where faunal relaxation by mammals
consistent with theory has been previously and consistently
detected (Glenn & Nudds, 1989; Gurd & Nudds, 1999;
Gurd et al., 2001; Wiersma & Nudds, 2001). Parks com-
posed of archipelagoes (Georgian Bay Islands National Park,
St Lawrence Islands National Park, and Mingan Archipelago
National Park) were excluded to study the effects of
insularization on terrestrial isolates alone, leaving ten parks
for analysis (Fig. 2).
The six species deemed to be historically present across all
ten national parks by Wiersma & Nudds (2001) were used
in this analysis: Castor canadensis (beaver), Lynx canadensis
(lynx), Microsorex hoyi (pygmy shrew), Tamiasciurus hud-
sonicus (red squirrel), Glaucomys sabrinus (northern flying
squirrel), and Lontra canadensis (river otter). Five of these
six are not currently resident in at least one park; therefore,
if their historical presence in a park is an example of a Ôfalse
positiveÕ, then Wiersma & NuddsÕ(2001) conclusion that
these species were extirpated from the parks may be sensitive
to type I error.
The shape of the AI province was clipped from each
mammal range using ArcView
TM
3.1 (Environmental Sys-
tems Research Institute, Redlands, CA, USA) and used as the
Ôbase rangesÕfrom which to simulate ranges with different
degrees of Ôarea of occupancyÕ. The base historical range of
each species was gridded with 10,000 km
2
cells using Arc-
View
TM
Sample 3.03 extension (Quantitative Decisions,
Merion Station, PA, USA) (Fig. 2). This cell size was chosen
somewhat arbitrarily, but with the intent of compromising
between two potential sources of error. Overly large cells
might have reduced the resolution of the analysis to the point
that effects of a simulated small percentage reduction in Ôarea
of occupancyÕwould be too hard to detect, as it would
reduce the simulated range by only one or two cells. This
might result in false confidence in the use of historical range
maps. On the other hand, overly small cells might have
created an unrealistic level of patchiness in the simulated
Ôarea of occupancyÕ.
A list of cells that comprised the simulated historical ran-
ges, and the parks that intersected each cell, was generated in
Riding Mountain
La Mauricie
Forillon
Kouchibouguac
PEI
Cape Breton
Highlands
Fundy
Kejimkujik
Point Pelee
Bruce Peninsula Location
in Canada
500 km
500 0
Figure 2 The Canadian portion of the Alleghenian–Illinoian mammal province (after Hagmeier, 1966), showing the ten national parks used
in the analysis. Superimposed is the range grid for the mammal province used in the analysis, from which the ranges for each of the six species
used were clipped.
2003 Blackwell Publishing Ltd, Journal of Biogeography,30, 375–380
Errors in range maps and historical species richness 377
ArcView
TM
. This list was exported to SPSS 10.0 (SPSS Inc.,
Chicago, IL, USA) and cells were deleted at random to cre-
ate, for each species, 300 historical ranges – 100 each of 25,
75 and 95% Ôarea of occupancyÕof the historical ranges used
in past studies (Glenn & Nudds, 1989; Gurd & Nudds,
1999; Wiersma & Nudds, 2001) – for a total of 1800 iter-
ations. For each iteration, species were scored as present or
absent in each area that is now a national park based on
whether they intersected with the simulated Ôarea of occu-
pancyÕ. Thus, a simulated ÔgapÕin the interior of the Ôextent
of occurrenceÕrepresented an unoccupied area and a species
would not be considered historically present in a park that
coincided with that gap.
For each park, we calculated the proportion of 100 itera-
tions in which all six species were detected as present histor-
ically, and therefore in agreement with Wiersma & Nudds
(2001). We calculated the mean proportion of agreement
(N ¼10 parks) at each simulated percentage Ôarea of occu-
pancyÕ(N ¼3). The proportion of those simulated results that
agreed with the original estimates by Wiersma & Nudds
(2001) was fitted against the simulated area of occupancy by
linear least squares regression.
RESULTS AND DISCUSSION
The numbers of species estimated to have been historically
present in parks varied with simulated variation in Ôarea of
occupancyÕof historical ranges (Fig. 3). If actual historical
ranges were on average only c. 5% smaller than those depicted
by Banfield (1974), then Wiersma & NuddsÕ(2001) estimates
of historical richness may be accurate in c. 90% of cases
(Fig. 3). That is, if historical range maps overestimated the
Ôextent of occurrenceÕby 5%, then there is a 10% chance that a
species detected at a specific location will be a false positive. If,
however, historical ranges were, on average, 25% smaller
than those used in previous analyses, then Wiersma & NuddsÕ
historical estimates of species richness may have been over-
estimated by one to four species in c. 40% of cases (Figs. 3, 4).
The estimate of the slope suggests that a 10% decrease in the
actual Ôarea of occupancyÕresults in a 13% decrease in ag-
reement with estimates of historical species richness compared
with estimates generated under the assumption that the Ôarea
of occupancyÕis also the Ôextent of occurrenceÕ(i.e. the his-
torical species range).
Figure 3 Data points [•;sindicates Wiersma & Nudds (2001)
baseline value] and linear regression (y ¼95.6)12.9x, r
2
¼0.98,
P¼0.0095) illustrating rate of change of the probability of all six
species being historically present as the Ôarea of occupancyÕis
reduced. Decreasing the Ôarea of occupancyÕof a historical range by
5% results in a 90% agreement with the baseline value which
assumed that Ôarea of occupancyÕ¼Ôextent of occurrenceÕ. Thus, if
historical range maps overestimate Ôarea of occupancyÕby 5%, then
estimates of faunal relaxation may be biased by Ôfalse positivesÕin
10% of cases. Note that standard error for 75% historical range
removed is too small for error bars to be displayed on the scale of
graph.
(a)
(b)
(c)
Iterations (%)Iterations (%)Iterations (%)
Figure 4 The number of species estimated to have been present
historically in a small park (Fundy National Park; solid bars) and a
large park (Riding Mountain National Park; striped bars) at three
simulated Ôareas of occupancyÕ: (a) 25%, (b) 75% and (c) 95% of
historical ranges.
2003 Blackwell Publishing Ltd, Journal of Biogeography,30, 375–380
378 L. D. Habib et al.
At the smallest simulated Ôareas of occupancyÕ(25%),
parks were estimated to have one to six fewer species than
Wiersma & Nudds (2001) reported, and this effect varied
among parks of different sizes. Typically, estimates of spe-
cies richness for larger parks were more robust to variation
in the Ôareas of occupancyÕthan were estimates for smaller
parks (Fig. 4). However, because most reserves are small
[e.g. see Fig. 3 in Gurd et al. (2001)], the problem of Ôfalse
positivesÕfor identification of faunal relaxation could be
particularly acute. These results suggest that studies which
sample historical range maps at a large grain size may be
more robust to errors in the Ôextent of occurrenceÕthan
studies which conduct sampling at a finer resolution or grain
size. In any case, future inventories of biodiversity could be
strategically directed towards sampling to resolve to what
degree Ôextent of occurrenceÕactually differs from Ôarea of
occupancyÕfor species ranges.
Further, accuracy of historical ranges may also vary by
species. Better historical records exist for game species than
for smaller mammals. For example, the beaver was heavily
trapped following European colonization, and therefore has
more point observations. Conversely, the pygmy shrew was
not only ignored by trappers, but was probably rarely sigh-
ted because of its small size. Historical range maps may also
be more accurate for generalist species that can live in a
variety of habitats. For example, the red squirrel is consid-
ered to occupy a variety of habitats; therefore, its historical
range may more nearly reflect its Ôarea of occupancyÕthan
the pygmy shrew, which is considered a habitat specialist,
suggesting a smaller Ôarea of occupancyÕthan indicated by its
historical range. In general, studies that use historical range
maps for abundant, charismatic, or commercially important
species should be more reliable than those using maps of
rare, shy or non-game species. In our analysis, most of the
six species used were generalists, and three of them (river
otter, beaver and lynx) were heavily trapped by Europeans.
Our analysis speaks of the question of the reliability of
estimates of faunal relaxation when historical range maps
overestimate the Ôareas of occupancyÕof species. It is also
possible that historical range maps may underestimate species
Ôareas of occupancyÕ. Thus, a species recorded as historically
absent from a current park would be a false negative, and
would be considered to have appeared in the park over the
interval. In this case, an increase in species richness would be
detected, when, in fact, the correct conclusion would be that
no decrease occurred. While this type of error would bear
directly on estimates of faunal turnover, it does not bear on
assessments of faunal relaxation. In that respect, error from
historical range maps that underestimate Ôareas of occupancyÕ
is conservative when the emphasis is on stemming losses of
species from reserves; a park manager is more concerned with
missing a local extinction than detecting a colonization that
didnot happen.
CONCLUSION
So far as we are aware, this is the first formal assessment
of the implications of Ôfalse positivesÕfor historical data
used to test for faunal relaxation in reserves, but it con-
curs with the intuitive logic of Robinson & Quinn (1992),
who expressed concern that variation in Ôarea of occu-
pancyÕaffects estimates of historical species ranges in
present-day reserves. Nevertheless, although it is generally
accepted that historical range information is often difficult
to obtain and of questionable accuracy, it is difficult to
evaluate how inaccurate historical ranges are likely to be,
and so, how significant this source of error might be for
evaluating the extent of faunal relaxation. In any case,
Aplet & Keeton (1999) argued that historical data ought
not to be simply dismissed or ignored because of the
difficulties it presents. Certainly, historical range maps are
often based on opportunistically collected data, such as
trapping records, and there is no single acceptable stand-
ard against which the maps can be judged (Williams,
1996). Often, they are the only data available with
which to work (van Jaarsveld et al., 1998). Thus, the
outcomes of analyses using historical range data are
somewhat conditional on the accuracy of the historical
data. Where the accuracy of the data is not known, there
is no option other than to accept the assumption that the
majority of the range maps used are accurate, at least
within the range of effects detectable in analyses, and
supply appropriate caveats. At large scales of spatial
analysis, perhaps more error in historical range maps
might be tolerated than at finer scales. Thus, we remain in
somewhat of a quandary, requiring appropriate historical
baseline data for species richness, but perhaps unsure of
the quality of the data. Gurd & Nudds (1999) and
Wiersma & Nudds (2001) may have reported inflated
rates of faunal relaxation, especially in the smallest parks;
however, from a conservation perspective, the effect of
overestimating historical species richness in present-
day parks is a conservative error. Our caveat is this: if
Banfield (1974) depicted historical ranges larger than they
actually were, then faunal relaxation in some Canadian
parks may not be as great as previous work suggested. It
will never be possible to know with certainty, but ex-
tremely large discrepancies between Ôextents of occurrenceÕ
and Ôareas of occupancyÕseem unlikely given that many of
the species that are presumed extirpated from parks are
large, charismatic, game species whose ranges were prob-
ably quite well known. Overall, we concur with others
(van Jaarsveld et al., 1998) who suggest that range maps
are a useful tool for large scale ecological and biogeo-
graphical analyses, so long as their limitations are
acknowledged.
ACKNOWLEDGMENTS
Thanks to W. Yang and A.M. Edwards for assisting with
writing the SPSS script to perform the iterations. This
research was supported by grants from the Natural Sci-
ences Engineering and Research Council of Canada and
Parks Canada to T.D. Nudds. Two anonymous reviewers
provided helpful comments on an earlier draft of the
manuscript.
2003 Blackwell Publishing Ltd, Journal of Biogeography,30, 375–380
Errors in range maps and historical species richness 379
REFERENCES
Aplet, G.H. & Keeton, W.S. (1999) Application of historical
range of variability concepts to biodiversity conservation.
Practical approaches to the conservation of biological
diversity (eds R.K. Baydack, H. Campa III and J. B. Haufler),
pp. 71–86. Island Press, Washington DC.
Banfield, A.W.F. (1974) The mammals of Canada. University of
Toronto Press, Toronto, ON.
Brashares, J.S., Arcese, P. & Sam, M.K. (2001) Human
demography and reserve size predict wildlife extinction in
West Africa. Proceedings of the Royal Society of London B.,
268, 2473–2478.
Glenn, S.M. & Nudds, T.D. (1989) Insular biogeography of
mammals in Canadian parks. Journal of Biogeography,16,
261–268.
Gurd, D.B. & Nudds T.D. (1999) Insular biogeography of
mammals in Canadian parks: a re-analysis. Journal of
Biogeography,26, 973–982.
Gurd, D.B., Nudds, T.D. & Rivard, D.H. (2001) Conservation
of mammals in eastern North American wildlife reserves:
how small is too small? Conservation Biology, 15, 1355–
1363.
Hagmeier, E.M. (1966) A numerical analysis of the distribu-
tional patterns of North American Mammals. II. Re-evalua-
tion of the provinces. Systematic Zoology,15, 279–299.
Humphreys, W.F. & Kitchener, D.J. (1982) The effect of habitat
utilization on species-area curves: implications for optimal
reserve area. Journal of Biogeography,9, 391–396.
Lawes, M.J. & Piper, S.E. (1998) There is less to binary maps
than meets the eye: the use of species distribution data in the
southern African sub-region. South African Journal of
Science,94, 207–210.
Newmark, W.D. (1995) Extinction of mammal populations in
western North American National Parks. Conservation Bio-
logy,9, 512–526.
Robinson, G.R. & Quinn, J.F. (1992) Habitat fragmentation,
species diversity, extinction, and design of nature reserves.
Applied population biology (eds S.K. Jain and L.W.
Botsford), pp. 223–248. Kluwer Academic Publishers, The
Netherlands.
van Jaarsveld, A.S., Gaston, K.J, Chown, S.L. & Freitag, S.
(1998) Throwing biodiversity out with the binary data? South
African Journal of Science, 94, 210–214.
Wiersma, Y.F. & Nudds, T.D. (2001) Comparison of methods
to estimate historical species richness of mammals for tests of
faunal relaxation in Canadian parks. Journal of Biogeogra-
phy,28, 447–452.
Williams, B.K. (1996) Assessment of accuracy in the mapping of
vertebrate biodiversity. Journal of Environmental Manage-
ment,47, 269–282.
BIOSKETCHES
Lucas Habib recently completed B.Sc. at the University
of Guelph, Canada and is currently working as a
freelance biologist. He is interested in spatial issues
related to conservation.
Yolanda Wiersma is a Ph.D. student in the Department
of Zoology, University of Guelph, and a Fulbright
Visiting Scholar at Duke University. She is interested in
landscape ecology and biogeography applications to the
design and management of protected areas.
Thomas Nudds is a professor of Zoology at the
University of Guelph. His research interests include
community ecology, the evolutionary and applied
ecology of waterfowl, and conservation biology. He
recently served on the federally appointed Panel on the
Ecological Integrity of the National Parks.
2003 Blackwell Publishing Ltd, Journal of Biogeography,30, 375–380
380 L. D. Habib et al.
