Content uploaded by Travis Longcore
Author content
All content in this area was uploaded by Travis Longcore on Dec 18, 2018
Content may be subject to copyright.
Western North American Naturalist 78(4), © 2018, pp. 617–632
Landscape modeling of the potential natural
vegetation of Santa Catalina Island, California
TRAVIS LONGCORE1,*, NINA NOUJDINA1, AND PETER J. DIXON2
1University of Southern California, Los Angeles, CA
2Catalina Island Conservancy, 330 Golden Shore, #170, Long Beach, CA 90802
ABSTRACT.—The vegetation of Santa Catalina Island has been significantly transformed through a history of intro-
duction of exotic plant species and disturbance by large introduced herbivores. Many of these disturbances have been
reduced in recent decades, using measures such as carefully controlling the number of bison and removing cattle,
sheep, feral pigs, and goats. The success of subsequent vegetation restoration actions depends on the choice of the right
plant community for a location, which may not be obvious for an island with extensive areas dominated by exotic
species. Environmental niche modeling is an approach to re-create the spatial distribution of habitat types for such a
purpose. Such models, however, often require both presence and absence data to be meaningful, while in this scenario
absence is misleading because it may reflect a long history of disturbance. Maximum entropy modeling is a technique
to model species distributions with presence-only data that has been shown to produce accurate results. We used this
modeling tool to model the environmental niche for distinct vegetation types, conceptualized as potential natural vege-
tation, on Catalina Island as a means to predict locations where restoration actions would be most successful and to
predict potential natural vegetation prior to anthropogenic disturbance. Using an existing vegetation map, we extracted
random points from within the polygons defining each native vegetation type. We then modeled the habitat suitability
for each habitat using high-resolution environmental data that included elevation, aspect, hillshade, northeastness,
slope, solar radiation, and topographic wetness index. The resulting models were combined to produce a map of poten-
tial natural vegetation. A 1977 map of potential natural vegetation included 4 vegetation types (woodland, chaparral,
scrub, and grassland) to which we compared our results. Our new model of potential natural vegetation has high
spatial complexity and high resolution. It also shows naturalistic responses to topography that are consistent with the
broad patterns mapped in 1977 while providing fine-scale resolution to inform restoration efforts.
RESUMEN.—La vegetación de la isla Santa Catalina se transformó de manera significativa debido al historial de
especies de plantas exóticas introducidas y a la perturbación generada por la introducción de grandes herbívoros. Gran
parte de esta perturbación disminuyó en las últimas décadas, tales como el control cuidadoso del número de bisontes,
la eliminación del ganado vacuno, las ovejas, los jabalíes y las cabras. El éxito de futuras medidas de restauración de la
vegetación dependerá de la correcta elección de la comunidad vegetal para la localidad, lo que puede no ser evidente en
una isla con grandes áreas dominadas por especies exóticas. El modelado de nichos ambientales es un enfoque que
recrea la distribución espacial de los tipos de hábitats. Sin embargo, estos modelos a menudo requieren que los datos de
presencia y de ausencia sean significativos, mientras que en este escenario la ausencia es confusa, ya que puede reflejar
el largo historial de perturbación. El modelado de máxima entropía es una técnica que recrea la distribución de las
especies únicamente con datos de presencia que demuestren resultados precisos. Utilizamos esta herramienta de mod-
elación para recrear el nicho ambiental de los distintos tipos de vegetación de la isla Santa Catalina (conceptualizadas
como vegetación natural potencial) como un medio para predecir los lugares donde las medidas de restauración serían
más exitosas y para predecir la vegetación natural potencial antes de la perturbación antropogénica. Usando un mapa de
vegetación vigente, seleccionamos aleatoriamente, puntos dentro de los polígonos que definen cada tipo de vegetación
nativa. Luego modelamos el hábitat adecuado para cada hábitat, mediante datos ambientales de alta resolución que
incluyen elevación, aspecto, sombreado, disposición orientada al noreste, pendiente, radiación solar e índice de humedad
topográfica. Los modelos resultantes se combinaron para crear un mapa de vegetación potencial natural. Comparamos
nuestros resultados con un mapa de vegetación potencial natural de 1977 que incluía cuatro tipos de vegetación
(bosques, chaparrales, matorrales y pastizales). Nuestro nuevo modelo de vegetación potencial natural presenta alta
complejidad espacial, alta resolución y muestra respuestas naturalistas a la topografía, todo esto es consistente con los
amplios patrones mapeados en 1977, al mismo tiempo que proporcionan una propuesta detallada para informar las
medidas de restauración.
*Corresponding author: longcore@usc.edu
617
TL orcid.org/0000-0002-1039-2613 PJD orcid.org/0000-0001-9222-0649
The natural vegetation of the California
Channel Islands is of significant interest to
ecologists and island managers in light of the
severe disturbance and degradation caused by
histories of agriculture and ranching on this
biodiverse archipelago. Interpretation of the
effects of current and prehistoric human
activities depends on a view of what vegeta-
tion might be supported by the biophysical
patterns and processes present at a given
point in time (Rick et al. 2014). Similarly, the
goals of current management strategies de -
pend on a realistic understanding of which
vegetation types might be supported by the
geophysical patterns of the island landscapes.
Now that significant progress has been made
toward removal or management of exotic
mammals (Keitt et al. 2002, Donlan et al.
2003, Nogales et al. 2004, Kindsvater 2010),
natural resource managers on the islands are
focusing even more on which vegetation types
to restore, and where.
The concept of potential natural vegetation
(PNV) offers an avenue to evaluate landscapes
that have been impacted by humans relative
to both their past condition and possible
futures. The PNV concept was introduced by
Faber (1937) and elaborated and promoted
by Tüxen (1956). It offers a framework to
describe what vegetation might exist in the
absence of human disturbance, being defined
as “the vegetation that would develop in a
particular ecological zone or environment,
assuming the conditions of flora and fauna to
be natural, if the action of man on the vegeta-
tion mantle stopped and in the absence of
substantial alteration in present climatic con-
ditions” (Tüxen 1956, translated in Gallizia
Vuerich et al. 2001). The concept has been
the subject of debate and redefinition over
time (Gallizia Vuerich et al. 2001) to determine
whether the PNV is that which would occur
should human disturbance cease (Westoff and
van der Maarel 1978), or whether it is the final
stage of succession (Moravec 1969, Härdtle
1995) synonymous with the Clementsian con-
cept of “climax” (Küchler and Zonneveld 1988).
Current understanding suggests incorporation
of irreversible alterations in environmental
conditions (that is, a focus on potential rather
than reconstructed vegetation) (Härdtle 1995)
and inclusion of more edaphic and topo-
graphic variables than those included in the
original focus on climate in creating PNV
maps (Gallizia Vuerich et al. 2001). Recent
criticisms of the PNV concept focus on the
difficulties with conceiving of vegetation
types as coherent entities in light of the
known individualistic character of species
movements over ecological time (Carrión
and Fernández 2009). The concept retains its
utility in environmental planning for the
assessment of vegetation (Fischer et al. 2013)
and in investigating the potential impacts of
changed environments, such as those from
changed climates (del Río and Penas 2006,
Bryn 2008, Lapola et al. 2008), or the removal
of stressors such as exotic herbivores.
The PNV concept may be a “provisionally
useful fiction” (Jackson 2013), in that the real
world never matches the ideal, because climatic
variation, disturbance patterns, and accidents
of history affect the particular pattern of actual
vegetation at any particular point in time. Yet
it offers the ability to fill in gaps in historical
knowledge and to generate hypotheses for
management actions. For example, conserva-
tion goals may not be consistent with the
capacity of the landscape to support them
because of misunderstandings of historical
distributions (Szabó et al. 2017). Two major
critiques of PNV as historically practiced
remain. First, PNV maps were often created
with expert knowledge (Fischer et al. 2013)
and therefore subject to biases and lacking in
granularity. Second, PNV was mapped at the
regional level (Tüxen 1956), and resulting maps
tended not to have sufficient spatial resolution
for site-level analysis (Zerbe 1998).
Spatially explicit numerical modeling offers
a remedy to the coarse resolution and biases of
regional-scale PNV maps. Numerical model-
ing with GIS has become more common and
has been used to model PNV, including in
Germany (see Fischer et al. 2013), Switzerland
(Brzeziecki et al. 1993), the Czech Republic
(Tichý 1999), Norway (Hemsing and Bryn 2012)
and China (Liu et al. 2009). With sufficient
input data resolution, landscape modeling can
offer a highly detailed assessment of potential
environmental niches across the landscape;
and indeed, this practice is commonplace for
the production of individual species models.
If a machine-learning approach is used, the
biases associated with expert production of
maps can be avoided.
Maxent is a predictive statistical GIS mod-
eling method that uses machine learning
618 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
based on a maximum entropy algorithm
(Phillips et al. 2006, Phillips and Dudík
2008, Elith et al. 2011). The program pro -
cesses environmental variables and evaluates
the com binations and interactions between
the variables to predict the probability of
encountering the modeled species across the
landscape, based on the similarity of the envi-
ronmental conditions (Phillips et al. 2006, Raes
and ter Steege 2007, Wollan et al. 2008, New-
bold et al. 2009, Kruijer et al. 2010). Maxent
has been used to assess past, potential, real-
ized, and future species distributions (Tingley
et al. 2014, Soto-Berelov et al. 2015, Bose et
al. 2016, Taylor et al. 2016); model vegetation
dynamics (Rodríguez-Sánchez and Arroyo 2008,
Papeş et al. 2012, Soto-Berelov et al. 2015);
assess shifts for a niche shared by species
communities (Velásquez-Tibatá et al. 2013,
Bertram and Dewar 2015, Nazeri et al. 2015);
model niches for endangered species (Babar
et al. 2012), prioritize areas for conservation
(Gaucherel et al. 2016, Akhter et al. 2017);
define niche constraints (Marino et al. 2011,
Camps et al. 2016); and assess risk of invasive
species (Bromberg et al. 2011, Duursma et al.
2013, Simpson and Prots 2013, McDowell et
al. 2014, Tingley et al. 2014, Collette and Pither
2015, Choudhury et al. 2016).
The advantage of Maxent over other ecolog-
ical models is that it requires presence data
only and produces results of high accuracy
(Elith et al. 2011, Collette and Pither 2015,
Nazeri et al. 2015, Choudhury et al. 2016). In
addition, Maxent evaluates each explanatory
variable in its relative importance for the pre-
diction and assigns a rank to it.
We therefore use Maxent as a model to
develop a PNV map for Santa Catalina Island,
a California Channel Island, following similar
approaches used elsewhere (Hemsing and
Bryn 2012, Zhang et al. 2013). Hemsing and
Bryn (2012) compared 3 PNV models for a
mountainous region in Norway, with one model
created from expert opinion, one from a rule-
based GIS model, and one from Maxent. They
found desirable attributes of the Maxent
approach, including its production of relative
probability maps, its objective approach, and
the ease of repeat model runs. Maxent or a
rule-based approach was preferable to an
expert map in all but the most disturbed
localities (Hemsing and Bryn 2012). Similarly,
Zhang et al. (2013) used Maxent to model
distributions with functional groups of plant
species as the unit of analysis.
A challenge to the use of Maxent for this
purpose is the wide extent of exotic vegetation
types on Catalina Island (Knapp 2005). Euro-
pean grasses predominate across many areas
of the island, especially on disturbed soils and
on shallow, rocky ridges and slopes (Thorne
1967). These grasses include species of the
genera Bromus, Avena, and Hordeum, and some
previous observations suggested that native
grasses (Stipa pulchra, S. cernua, and S. lepida)
returned to such locations when grazing
pressure was removed (Thorne 1967). Exotic
annual grasslands have replaced areas that
may once have been occupied by chaparral or
shrublands, or they may have replaced native
grasslands in other locations, making the in -
ter pretation of their presence challenging.
Our goal was to develop a map of PNV for
Santa Catalina Island as a means both to
understand the historical distribution of vege-
tation types across the island and to provide
testable hypotheses to guide ongoing plant
res toration efforts. Suppression of woody
plant regeneration through herbivory by in -
troduced ungulates is a long-term concern of
island managers (Minnich 1980, Stratton 2009,
Knapp 2010a, 2010b, Ramirez et al. 2012), and
a detailed and high-resolution understanding
of the environmental niche occupied by woody
vegetation types would be a key output of a
PNV map. This effort builds on previous
species modeling efforts (Franklin and Knapp
2010) and introduces higher-resolution envi-
ronmental data and a vegetation community
approach. We also compare the results to an
existing low-resolution map of Catalina Island
PNV, created as part of a statewide effort by
Küchler (1977), which predicts that 4 vegeta-
tion types would have dominated the predistur-
bance landscape (coastal sage scrub, grassland,
chaparral, and woodland).
METHODS
Study Area
Santa Catalina Island is located 32 km off
the shore of Los Angeles County, California, in
the Pacific Ocean (Schoenherr et al. 1999)
(Fig. 1). It is part of the California Channel
Islands, which are all under varying degrees
of conservation ownership and support many
rare and endemic plant and animal species
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 619
(Rick et al. 2014). The island is 194 km2and
ranges from sea level to 640 m elevation. It
is a geologically young island, a little more
than 100 million years old with 3 major geo-
logical formations: Catalina schist, Catalina
pluton, and various andesite rocks (Rowland
1984). The climate is a marine-influenced
Mediterranean-type climate with an average
annual temperature of 15.8 °C and an aver-
age annual precipitation of 29.2 cm over the
past 41 years (ncdc.noaa.gov).
Environmental Niche Modeling
We used Maxent to model PNV of Santa
Catalina Island, California, in the absence of
grazing disturbance. Following Moravec
(1998), we relied on an actual vegetation
map (Knapp 2005) to model potential vegeta-
tion and a high-resolution digital elevation
model and its derivatives to serve as envi-
ronmental layers (LAR-IAC 2006). Both data
sets had been produced within a 1-year time
period, which minimized the odds of change
in vegetation communities.
VEGETATION MAP OF CATALINA ISLAND.—The
actual vegetation map of Catalina Island was
developed from aerial imagery obtained in
2000 and comprised 50 classes, each describ-
ing a specific land cover, vegetation type, or
vegetation community (Knapp 2005). We chose
this map as a baseline because of its exhaustive
nature, fine spatial resolution of 4-ha mapping
units, and homogeneity of polygons; each class
represented a thoroughly described, uniform
vegetation community characterized by 1 or 2
dominant species or land cover types.
Our main interest was to model potential
vegetation cover of the island with native
species. Thus, from the available 50 classes,
we selected 11 classes that represented pure
stands of native vegetation: Island Woodland
(IW), Coastal Sage Scrub (CSS), Island Chap-
arral (IC), Coastal Bluff Scrub (CBS), Coastal
Marsh (CM), Maritime Cactus Scrub (MCS),
Mulefat Scrub (MFS), Riparian Herbaceous
(RH), Southern Beach and Dune (SBD), and
Southern Riparian Woodland (SRW). We also
included Grassland (GR) (even though cur-
rent grassland areas are dominated by exotic
species) as indicators of areas where native
grasses and forbs might have been distributed
as posited by Thorne (1967). The actual vege-
tation map was provided in a Esri shapefile
format and was projected to UTM Zone 11
620 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
Fig. 1. Location and topography of Santa Catalina Island, Los Angeles County, California.
with the NAD83 datum. It was reprojected to
the California State Plane Coordinate System
with NAD 1983 to match the projection of the
environmental layers.
ENVIRONMENTAL VARIABLES.—A 1.5-m-
resolution digital elevation model (LAR-IAC
2006) served as a source for other explanatory
variables that are frequently used in environ-
mental studies. We derived slope, hillshade,
solar radiation index, aspect, northeast–south-
west gradient, and topographic wetness index
from the digital elevation model. Slope is a
measure of the steepness of a surface, usually
expressed in degrees or as a percentage. We
converted its values to radians to fit Maxent
input requirements. Hillshade simulates the
shadow cast by the sun onto the terrain and
thus emphasizes the 3-dimensional aspect of
the area. Solar radiation index is a total
amount of incoming solar energy calculated
for each pixel. This variable depends on
slope, aspect, and relative position of the sun.
Aspect is another topographic element that
describes the compass direction the slope is
facing and is measured in degrees from north.
Even though aspect is a circular variable, we
in cluded it in the input data set, relying on
Maxent’s ability to omit variables that have
little or no effect on the model. Aspect, how-
ever, is commonly transformed by trigonomet-
ric functions, such as cosine and sine, thereby
retaining the continuity of the variable (Roberts
1986, Palmer 1993). Northeast–southwest gra-
dient is a generalized proxy for the “wetness”
characteristic and fits the environment of the
Channel Islands (Paudel et al. 2016). This vari-
able was calculated in ArcGIS 10.3.1 (Esri,
Redlands, CA) using Map Algebra expression
Con ([Aspect] == −1, 0, Sin (([Aspect] + 45)
* 3.14159 / 180)) .
Topographic wetness index (TWI), a wetness
proxy, is derived from slope, flow accumula-
tion, and flow direction. It was calculated in
ArcGIS using Map Algebra expression
Ln (“Flow Accumulation” /
(Tan (“Slope-in-Radians”)) + 0.01) .
The variables described above were selected
from a larger set of environmental layers that
included geology and soils. The resolution of
the geology and soils layers was much coarser
than the rest of the data, which would have
caused significant reduction of spatial resolu-
tion of the input data as a whole, adding vague-
ness to the final results. To improve the model
in the future, we highly recommend adding
these environmental variables at a resolution
similar to the base data set. Before including
the input layers, we tested them for multi-
collinearity using the Exploratory Regression
tool in ArcMap to confirm that no layer had a
variance inflation factor that exceeded 2.5 (cor-
responding to a correlation with other vari-
ables of 0.6). The tool was run on a subset of
5000 points randomly selected from each layer.
All layers were converted to Esri ASCII grid
format to meet Maxent input requirements.
TRAINING SAMPLES.—Polygons of the actual
vegetation map that corresponded to target
vegetation classes were selected and buffered
with a negative distance of 1 m to omit poten-
tial transitional zones between classes. Training
points for each class were built in ArcGIS as
random samples, stratified by corresponding
polygon area, with a 3-m minimum allowed
distance between points. A pool of random
points for each vegetation class was then used
as the Maxent input according to the area
covered by the vegetation class (Table 1).
MODEL DEVELOPMENT.—We developed a
model for each vegetation class with 40 repli-
cations using subsampling with 20 random test
points withheld in each model run and a maxi-
mum model iteration of 5000. In subsampling
replication, the presence points are repeatedly
split into random training and testing subsets,
in accordance with the set proportion. Spatial
resolution of modelled species distribution
corresponded to the resolution of input layers
and was equal to 1.5 ×1.5 m. Model perfor-
mance was assessed using the area under the
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 621
TABLE 1. Training sample sizes for vegetation classes.
Area Sample
Vegetation class (m2) size
Coastal bluff scrub 313,843 272
Coastal marsh 9639 50
Coastal sage scrub 3,9183,246 2418
Grassland 2,3349,156 1446
Island chaparral 3,7252,699 2029
Island woodland 725,927 554
Maritime cactus scrub 11,512 50
Mule fat scrub 4836 10
Riparian herbaceous 98,116 199
Southern beach and dune 520,001 276
Southern riparian woodland 651,417 450
receiver operating characteristic curve (AUC)
statistic. The receiver operating characteristic
(ROC) curve is a plot of the values of sensitiv-
ity against 1 −Specificity. In short, a model
with good discriminating ability has high sen-
sitivity and high specificity, and the line of
such a curve tends to be closer to the upper
left corner of the plot area; thus, the area
under it will increase. Poor-fit models, on the
other hand, will have ROC curves approximat-
ing a 45-degree diagonal line on the plot area.
The model also produced jackknife tests for
regularized training gain, test gain, and AUC
for each vegetation class.
Each model was thresholded to identify
whether a vegetation class could be expected at
each pixel. We then overlaid the predicted dis -
tributions such that models that performed best
by AUC were layered below models with lower
performance so that the high-performing mod-
els would not swamp the lower-performing
models. We adjusted this order by hand so
that the highest values of each model were
mapped in the composite. We also developed
a composite map based on a strict rule-based
approach, where the model suitability values
were standardized to a scale of 0 to 1, and each
pixel was assigned the vegetation classification
with the highest standardized value at that
point (Supplementary Material 1).
Comparison of Composite Model
to 2005 and 1977 Maps
Our composite model of vegetation differed
from the actual vegetation map on which it
was based in that we assigned each pixel to
the class having the highest suitability value,
even if that class was not the vegetation present
at that location. We therefore compared our
classifications with the 2005 actual vegetation
map to see how it differed from our idealized
environmental niche models for those locations.
For comparison, we selected the 4 most domi-
nant vegetation types: Coastal Sage Scrub,
Island Woodland, Grassland, and Island Chap -
arral. The raw raster maps modeled by Maxent
for each of the 4 vegetation classes repre-
sented a probability surface of that species
being encountered in the area, other things
being equal. We set a threshold of 0.51 for
these layers to set a reasonable feasibility
extent for each of them. We then applied a
generalization technique to remove isolated
pixels from the maps. After testing several
smoothing methods, including filtering and
focal statistics, we chose the expand-and-shrink
method. A sequence of expanding by 2 fol-
lowed by shrinking by 3 and finishing with
expanding by 1 not only eliminated isolated
single pixels, but also preserved the original
clusters of pixels and thus resulted in a smooth
map suitable for further comparative analysis
without any single-pixel inclusions. The same
vegetation classes were selected from the actual
vegetation map with the addition of mixed
classes; for example, the Coastal Sage Scrub
class was extended by adding Grassland to
make Coastal Sage Scrub/Grassland.
The classic book Terrestrial Vegetation of
California (Barbour and Major 1977) contains
an insert map of the PNV of California (Küch-
ler 1977). Developed at the scale of the state,
the map contains 54 plant communities, 4 of
which are found on Catalina Island (Küchler
1977). This is the only known map of the PNV
of Catalina Island, and so despite the differ-
ences in spatial resolution, we compared our
composite map with it by selecting 4 vegeta-
tion classes that corresponded to the 4 types
mapped by Küchler. We then took a random
sample of 10,000 points, created contingency
tables comparing the 2 maps, and tested sig-
nificance using a chi-square test implemented
in JMP Pro 12 software (SAS Institute, Inc.,
Cary, NC).
RESULTS
A model was created for each vegetation
class, with performance ranging from fair
(AUC > 0.7) to good (AUC > 0.8) and excellent
(AUC > 0.9) (Table 2, Fig. 2). Less common
vegetation classes generally had better model
performance, with common vegetation classes
having a lower AUC (e.g., Coastal Sage Scrub,
Grassland). Environmental variables played
differing roles in predicting the distribution
of vegetation types (Fig. 3). The relative im -
portance of the environmental variables for
modeling each class was assessed by intercom-
parison of percent contribution, permutation
importance, and jackknife importance. Eleva-
tion was most important for Coastal Bluff
Scrub, Coastal Marsh, Maritime Cactus Scrub,
Riparian Herbaceous, and Southern Beach
and Dune; solar radiation was most important
for Island Chaparral, Island Woodland, and
Southern Riparian Woodland; aspect was most
622 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
important for Coastal Sage Scrub; and slope
was most important for Mulefat Scrub. Distri -
bution of some vegetation types could be
explained almost entirely by topographic attri -
butes, such as Coastal Marsh (extreme low ele-
vation), Southern Beach and Dune (low eleva-
tion), and Mulefat Scrub (flat and gentle slopes).
The Maxent results for each class model-
ing were thresholded and overlaid in ArcGIS
to present a map of potential vegetation
cover (Fig. 4). We carefully adjusted a thresh-
old value for each class based on the perfor-
mance of the model. In general, classes with
lower AUC values have a higher probability
threshold. The rule-based composite map
(Supplementary Ma te rial 1) showed similar
patterns, as expected, because it was based
on the same underlying distribution models,
but with a sub stantially reduced distribution
of Island Woodland, which was replaced by
Island Chaparral.
Comparison of Model with
2005 Vegetation Map
The modeled potential vegetation corre-
sponds at the broadest levels with the 2005
vegetation map from which it was derived
(Fig. 5). The environmental niche models sug-
gest that Island Woodland is more suitable in
some areas that are currently Island Chaparral,
Coastal Sage Scrub, and Grassland. In addition,
some areas that were mapped as Grassland in
2005 have higher suitability values in the model
for Coastal Sage Scrub. Overall, the modeled
map tracks the topographic layers used as
environmental inputs closely and much more
distinctly than the actual vegetation. In doing
so, it highlights the prevalence of Island Chap-
arral on north-facing slopes and Coastal Sage
Scrub on south-facing slopes.
Comparison of Model with
1977 Potential Natural Vegetation Map
The patterns of modeled vegetation match
the overall patterns proposed by Küchler in
his 1977 map. Coastal Sage Scrub dominates
on south-facing slopes, Island Chaparral on
north-facing slopes, Grassland in the interior,
and Island Woodland (which Küchler defined
as Southern Oak Woodland) on the southern
part of the island (Fig. 5). Each of these rela-
tionships is significant in a contingency table
analysis (chi-square test: P< 0.05), with the
exception of the correspondence between our
grassland locations and the Küchler map. Our
map predicts more Island Woodland than did
Küchler, and more Coastal Sage Scrub in place
of Grassland.
DISCUSSION
The models produced for vegetation classes
on Catalina Island are abstractions that do
not represent the actual vegetation at any
particular moment in history. They do, however,
provide quantifiable, replicable representations
of the relationship between the environment
and vegetation classes that are in many ways
consistent with previously observed environ-
mental niches (e.g., Westman 1983). For exam-
ple, with no a priori requirement to do so, the
models confirmed the importance of aspect
to the distribution of Island Chaparral and
Coastal Sage Scrub. They also indicate that
solar radiation is the most important factor
determining the distribution of a set of vegeta-
tion types that require lower solar radiation
and higher moisture to thrive (e.g., Southern
Riparian Woodland, Island Woodland, and
Island Chaparral).
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 623
TABLE 2. Maxent model performance (AUC and SD) and mapping thresholds for each vegetation class.
Vegetation class AUC SD Threshold
CBS (Coastal Bluff Scrub) 0.979 0.003 0.37
CM (Coastal Marsh) 0.997 0.001 0.33
CSS (Coastal Sage Scrub) 0.662 0.011 0.51
GR (Grassland) 0.727 0.011 0.45
IC (Island Chaparral) 0.722 0.010 0.51
IW (Island Woodland) 0.892 0.012 0.37
MCS (Maritime Cactus Scrub) 0.990 0.005 0.42
MFS (Mulefat Scrub) 0.988 0.006 0.40
RH (Riparian Herbaceous) 0.967 0.009 0.46
SBD (Southern Beach and Dune) 0.984 0.001 0.35
SRW (Southern Riparian Woodland) 0.904 0.012 0.36
624 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
Fig. 2. Habitat suitability models for vegetation types on Santa Catalina Island. See Table 2 for vegetation class codes.
The models are fraught with all the difficul-
ties of treating vegetation classes as coherent
phytosociological units when all evidence in -
dicates that plant species respond individual-
istically to changes in the environment. The
models also represent a problematic space-
for-time substitution by predicting future or
past vegetation based on existing conditions
that themselves are the result of complex his-
tories not necessarily encapsulated in their
topographic positions. Nevertheless, it is an
accepted practice to construct PNV maps by
extrapolating actual vegetation to areas where
native vegetation is absent (Moravec 1998),
and as such, the results appear realistic, or at
least appear to provide reasonable hypotheses
that can be tested.
The question of what is “natural” vegetation
for the Channel Islands is not a simple one.
The Channel Islands have undergone a series
of changes that would have affected vegetation
distribution over the past tens of thousands
of years. These changes include management of
vegetation with fire by indigenous human pop -
ulations perhaps starting as early as 14–12.5 kya
(Hardiman et al. 2016). We are unable to ad -
dress the many changes in flora and fauna of
the late Quaternary but rather have to interpret
the modeled vegetation distributions as incor-
porating disturbance regimes embodied in the
2005 vegetation distribution used as training
data. With knowledge of the distribution of
human populations and hypotheses about their
role in encouraging or discouraging certain
keystone plant species (e.g., oaks), distribution
modeling could be used to test for that human
influence (Tulowiecki and Larsen 2015).
The long-time presence of pigs, sheep, and
goats on Catalina Island is likely also to have
affected soil depth, character, and erosion
(Coblentz 1977, Brumbaugh and Leishman
1982, Minnich 1982, Van Vuren and Coblentz
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 625
Fig. 3. Permutation importance of each environmental variable to each vegetation type model. See Table 2 for vegeta-
tion class codes. DEM = digital elevation model, TWI = topographic wetness index.
1987). We did not include soils in our models
because preliminary models showed that soils
did not contribute significantly and because
the resolution of soils maps is generally lower
than the resolution of the environmental data
that we did use. Although soils are likely to
have changed during the period when pigs,
sheep, and goats were present, it is not clear
that these changes are as important at the
10-m scale at which we modeled vegetation
distributions. A future effort might investigate
the degree to which island topography has been
altered in the modern era through erosion and
the possible effect that alteration would have
at scales relevant to vegetation distribution.
The importance of topography and especially
aspect in many models may be attributed to
associated environmental variation, such as fog
days. Early botanists recognized that fogs
were more common on the eastern ridges and
at higher elevations—areas that were described
as “luxuriously vegetated” (Millspaugh and
Nutall 1923)—and modern research confirms
the importance of fog for the vegetation of the
Channel Islands (Fischer et al. 2009, Fischer
et al. 2016). This pattern is evident in our
mapping of extensive Island Woodland on the
eastern end of the island at higher elevations.
It is generally construed that the PNV
represents the state of an environment without
disturbance (Jackson 2013). We are not certain
this has to be the case, because one can model
vegetation that is dependent on disturbance as
long as it is defined that the outputs represent
vegetation that will be somewhere along a set
of successional pathways associated with the
overall vegetation type, if underlying ecologi-
cal processes and conditions (e.g., climate)
are stable. The role of natural disturbance on
vegetation type distribution on Catalina Island
may be smaller than expected. Fire caused by
lightning rather than by human ignition is
626 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
Fig. 4. Composite map of modeled distribution of vegetation on Santa Catalina Island.
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 627
Fig. 5. Comparison of (a) potential natural vegetation (Küchler 1977), (b) 2005 actual vegetation (Knapp 2005), and
(c) modeled potential vegetation on Santa Catalina Island, using 4 vegetation units shared in each map: Coastal Sage
Scrub, Grassland/California Prairie, Island Chaparral, and Island Woodland.
extremely uncommon (Minnich 1982, Carroll et
al. 1993), likely a result of the maritime condi-
tions (although 3 lightning fires were recorded
between 2001 and 2005 in Catalina Island
Conservancy records). Fires have occurred in
the absence of grazing as exotic grasses
increase, but it is not clear that fires naturally
(without human influence) would have been
common. The remaining disturbances would
be relatively limited (e.g., landslides).
The question that then arises is whether
our modeled vegetation actually represents
the potential distribution of vegetation types
on the landscape in the absence of disturbance
and whether it is therefore similar to the his-
toric condition before cattle and sheep were
introduced in the 1850s. Two lines of evidence
speak to these questions.
First, removal of grazing on the current
landscape leads to vegetation change. In a
series of exclosure experiments undertaken by
the Catalina Island Conservancy, postfire Island
Chaparral and Island Woodland recovered in
exclosures where all exotic grazers were re -
moved but were transformed to grasslands
where grazers were present (Fig. 6) (Knapp
2005). Our environmental niche models pro-
vide a set of hypotheses about what vegetation
types are most likely to reestablish in exclo-
sures or whether exotic herbivore numbers
and distributions are otherwise controlled.
Second, a thorough investigation into the
historical ecology of Catalina Island could
provide evidence of the performance of our
models. Historical ecology can refer to both a
multi-thousand-year history of a place, spanning
the Pleistocene and into the Holocene (Rick et
al. 2014), or a description of ecological condi-
tions at some given point prior to a particular
intensity of human disturbance (Van Dyke
and Wasson 2005, Grossinger et al. 2007). For
southern California, historical ecology efforts
have focused on defining vegetation and asso-
ciated species distributions and landscape
processes in the late 1800s, prior to wide-
spread urbanization (Stein et al. 2010, Beller
et al. 2011, 2014). This type of investigation
could use our modeled distributions as a
starting point with which to compare docu-
mentary evidence. As a preliminary view, we
are optimistic that the evidence would support
the patterns we describe.
Minnich (1982) described the replacement
of Island Chaparral and Coastal Sage Scrub by
grasslands in response to grazing pressure on
Catalina Island, as have others. Our results
suggesting a greater distribution of Island
Woodland and Island Chaparral are generally
consistent with those of other authors who
have concluded that these communities have
been reduced in extent by grazing (Kindsvater
2010). We do, however, model a considerable
area of grassland and lack the detail to specify
whether it would indeed be grassland in the
absence of the many invasive exotic grass
species that characterize the California flora.
The grasslands are modeled most along ridge-
lines where soils are likely to be shallow, and
this pattern may indeed hold true in an envi-
ronment where exotic grasses are aggressively
controlled, but this question would certainly
be a productive avenue for future research.
Future research efforts might, for example,
georeference the locations and plant species
described by Millspaugh and Nutall (1923) and
other early accounts such as Stehman Forney’s
handwritten journals associated with the 1877
U.S. Coast Survey (Smith 1897, Cockerell 1939).
Early photographs, such as those reviewed by
Minnich (1982), could also be reviewed and
compared with model outputs. Use of such
historical sources would help to refine the
models and composite maps, especially where
they can confirm presence of the vegetation
types inhibited by exotic herbivores.
It has been suggested that numerical mod-
eling misses some of the detail found in expert
knowledge (Fischer et al. 2013) in the creation
of PNV maps. Our results do, however, provide
useful hypotheses about past vegetation that
would be testable with a detailed historical
ecology investigation, and about future vege-
tation trajectories in the event that grazing
pressure is removed. At the least, the quantita-
tive approach used here is replicable and could
be modified with a new actual vegetation map
or environmental inputs. Furthermore, it dem -
onstrates the further utility of such an approach
in creating high-resolution output of PNV that
provides spatially explicit hypotheses at the
management-relevant scale of meters, which
compares favorably with the only other avail-
able effort (Küchler 1977).
SUPPLEMENTARY MATERIAL
One online-only supplementary file accom-
panies this article (scholarsarchive.byu.edu/
wnan/vol78/iss4/12).
628 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 629
Fig. 6. Example of vegetation recovery in exclosures on Santa Catalina Island. Photo by Amy Catalano.
SUPPLEMENTARY MATERIAL 1. Composite map
of modeled vegetation distribution of Santa
Catalina Island. Top: Rule-based composite with
suitability values normalized to a scale of 0 to 1
and highest value given precedence. Bottom:
Models ordered with lower AUC models given
precedence and adjusted by hand to avoid exclu-
sion of lower-performing models from the map.
ACKNOWLEDGMENTS
The University of Southern California
Dornsife College of Letters, Arts and Sciences
provided seed funding for this research. The
Catalina Island Conservancy provided access
to key data sets. We thank 2 reviewers for
constructive and insightful comments.
LITERATURE CITED
AKHTER, S., M.A. MCDONALD, P. VAN BREUGEL, S. SOHEL,
E.D. KJAER, AND R. MARIOTT. 2017. Habitat distribu-
tion modelling to identify areas of high conservation
value under climate change for Mangifera sylvatica
Roxb. of Bangladesh. Land Use Policy 60:223–232.
BABAR, S., G. AMARNATH, C.S. REDDY, A. JENTSCH, AND S.
SUDHAKAR. 2012. Species distribution models: eco-
logical explanation and prediction of an endemic and
endangered plant species (Pterocarpus santalinus
L.f.). Current Science 102:1157–1165.
BARBOUR, M.G., AND J. MAJOR. 1977. Terrestrial vegetation
of California. Wiley-Interscience, New York, NY.
BELLER, E., S. BAUMGARTEN, R. GROSSINGER, T. LONG -
CORE, E. STEIN, S. DARK, AND S. DUSTERHOFF. 2014.
Northern San Diego County lagoons historical ecol-
ogy investigation: regional patterns, local diversity,
and landscape trajectories. San Francisco Estuary
Institute, Richmond, CA.
BELLER, E.E., R.M. GROSSINGER, M.N. SALOMON, S.J.
DARK, E.D. STEIN, B.K. ORR, P.W. DOWNS, T.R.
LONG CORE, G.C. COFFMAN, A.A. WHIPPLE, ET AL.
2011. Historical ecology of the Lower Santa Clara
River, Ventura River, and Oxnard Plain: an analysis
of terrestrial, riverine, and coastal habitats. San
Francisco Estuary Institute, Oakland, CA.
BERTRAM, J., AND R.C. DEWAR. 2015. Combining mecha-
nism and drift in community ecology: a novel statis-
tical mechanics approach. Theoretical Ecology 8:
419–435.
BOSE, R., F. MUNOZ, B.R. RAMESH, AND R. PELISSIER.
2016. Past potential habitats shed light on the bio-
geography of endemic tree species of the Western
Ghats biodiversity hotspot, South India. Journal of
Biogeography 43:899–910.
BROMBERG, J.E., S. KUMAR, C.S. BROWN, AND T.J. STOHL -
GREN. 2011. Distributional changes and range pre-
dictions of downy brome (Bromus tectorum) in Rocky
Mountain National Park. Invasive Plant Science and
Management 4:173–182.
BRUMBAUGH, R., AND N. LEISHMAN. 1982. Vegetation
change on Santa Cruz Island, California: the effect of
feral animals [sheep grazing]. USDA Forest Service
General Technical Report PSW-58, Pacific Southwest
Forest and Range Experiment Station (USA).
BRYN, A. 2008. Recent forest limit changes in south-east
Norway: effects of climate change or regrowth after
abandoned utilisation? Norsk Geografisk Tidsskrift–
Norwegian Journal of Geography 62:251–270.
BRZEZIECKI, B., F. KIENAST, AND O. WILDI. 1993. A simu-
lated map of the potential natural forest vegetation
of Switzerland. Journal of Vegetation Science 4:
499–508.
CAMPS, D., D. VILLERO, J. RUIZ-OLMO, AND L. BROTONS.
2016. Niche constraints to the northwards expansion
of the common genet (Genetta genetta, Linnaeus
1758) in Europe. Mammalian Biology 81:399–409.
CARRIÓN, J.S., AND S. FERNÁNDEZ. 2009. The survival of
the ‘natural potential vegetation’ concept (or the
power of tradition). Journal of Biogeography 36:
2202–2203.
CARROLL, M.C., L.L. LAUGHRIN, AND A.C. BROMFIELD.
1993. Fire on the California islands: does it play a
role in chaparral and closed cone pine forest habi-
tats? Pages 73–88 in F.G. Hochberg, editor, Third
California Islands Symposium: recent advances in
research on the California islands.
CHOUDHURY, M.R., P. DEB, H. SINGHA, B. CHAKDAR, AND
M. MEDHI. 2016. Predicting the probable distribu-
tion and threat of invasive Mimosa diplotricha
Suavalle and Mikania micrantha Kunth in a pro-
tected tropical grassland. Ecological Engineering
97:23–31.
COBLENTZ, B.E. 1977. Some range relationships of feral
goats on Santa Catalina Island, California. Journal of
Range Management 30:415–419.
COCKERELL, T.D.A. 1939. Natural history of Santa
Catalina Island. Scientific Monthly 48:308–318.
COLLETTE, L.K.D., AND J. PITHER. 2015. Modeling the
potential North American distribution of Russian
olive, an invader of riparian ecosystems. Plant Ecol-
ogy 216:1371–1383.
DEL RÍO, S., AND A. PENAS. 2006. Potential areas of ever-
green forests in Castile and Leon (Spain) accord-
ing to future climate change. Phytocoenologia 36:
45–66.
DONLAN, C.J., D.A. CROLL, AND B.R. TERSHY. 2003.
Islands, exotic herbivores, and invasive plants: their
roles in coastal California restoration. Restoration
Ecology 11:524–530.
DUURSMA, D.E., R.V. GALLAGHER, E. ROGER, L. HUGHES,
P.O. DOWNEY, AND M.R. LEISHMAN. 2013. Next-
generation invaders? Hotspots for naturalised sleeper
weeds in Australia under future climates. PLOS
ONE 8:e84222.
ELITH, J., S.J. PHILLIPS, T. HASTIE, M. DUDÍK, Y.E. CHEE,
AND C.J. YATES. 2011. A statistical explanation of
MaxEnt for ecologists. Diversity and Distributions
17:43–57.
FABER, A. 1937. Erläuterungen zum pflanzensoziolo -
gischen Kartenblatt des mittleren Neckar- und des
Ammertalgebietes. Forstdirektion und Württemberg
Naturaliensammlung, Stuttgart.
FISCHER, D.T., C.J. STILL, C.M. EBERT, S.A. BAGUSKAS,
AND A. PARK WILLIAMS. 2016. Fog drip maintains
dry season ecological function in a California coastal
pine forest. Ecosphere 7(6):e01364.
FISCHER, D.T., C.J. STILL, AND A.P. WILLIAMS. 2009. Sig-
nificance of summer fog and overcast for drought
stress and ecological functioning of coastal California
endemic plant species. Journal of Biogeography
36:783–799.
630 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
FISCHER, H.S., S. WINTER, E. LOHBERGER, H. JEHL, AND
A. FISCHER. 2013. Improving transboundary maps of
potential natural vegetation using statistical model-
ing based on environmental predictors. Folia Geo -
botanica 48:115–135.
FRANKLIN, J., AND D.A. KNAPP. 2010. Habitat relationships
and potential restoration sties for Quercus pacifica
and Q. tomentella on Catalina Island. Pages 69–94 in
D.A. Knapp, editor, Oak ecosystem restoration on
Santa Catalina Island, California: proceedings of an
on-island workshop, February 2–4, 2007. Catalina
Island Conservancy, Avalon, CA.
GALLIZIA VUERICH, L., L. POLDINI, AND E. FEOLI. 2001.
Model for the potential natural vegetation mapping
of Friuli Venezia-Giulia (NE Italy) and its applica-
tion for a biogeographic classification of the region.
Plant Biosystems 135:319–335.
GAUCHEREL, C., R. VEZY, F. GONTRAND, D. BOUCHET, AND
B.R. RAMESH. 2016. Spatial analysis of endemism
to redefine conservation areas in Western Ghats
(India). Journal for Nature Conservation 34:33–41.
GROSSINGER, R., C. STRIPLEN, R. ASKEVOLD, E. BREW-
STER, AND E. BELLER. 2007. Historical landscape
ecology of an urbanized California valley: wetlands
and woodlands in the Santa Clara Valley. Landscape
Ecology 22:S103–S120.
HARDIMAN, M., A.C. SCOTT, N. PINTER, R.S. ANDERSON,
A. EJARQUE, A. CARTER-CHAMPION, AND R.A. STAFF.
2016. Fire history on the California Channel Islands
spanning human arrival in the Americas. Philosophi-
cal Transactions of the Royal Society B–Biological
Sciences 371:20150167.
HÄRDTLE, W. 1995. On the theoretical concept of the
potential natural vegetation and proposal for an
up-to-date modification. Folia Geobotanica et Phyto-
taxonomica 30:263–276.
HEMSING, L.Ø., AND A. BRYN. 2012. Three methods for
modelling potential natural vegetation (PNV) com-
pared: a methodological case study from south-central
Norway. Norsk Geografisk Tidsskrift–Norwegian
Journal of Geography 66:11–29.
JACKSON, S.T. 2013. Natural, potential and actual vegeta-
tion in North America. Journal of Vegetation Science
24:772–776.
KEITT, B.S., C. WILCOX, B.R. TERSHY, D.A. CROLL, AND
C.J. DONLAN. 2002. The effect of feral cats on the
population viability of Black-vented Shearwaters
(Puffinus opisthomelas) on Natividad Island, Mexico.
Animal Conservation 5:217–223.
KINDSVATER, L. 2010. Plant communiites associated with
the rare, paleoendemic oak, Quercus tomentella on
Santa Cruz and Santa Rosa Islands, California. Pages
53–68 in D.A. Knapp, editor, Oak ecosystem restora-
tion on Santa Catalina Island, California: proceedings
of an on-island workshop, February 2–4, 2007.
Catalina Island Conservancy, Avalon, CA.
KNAPP, D.A. 2005. Vegetation community mapping on
Santa Catalina Island using orthorectification and
GIS. Pages 193–203 in D.K. Garcelon and C.A.
Schwemm, editors, Proceedings of the Sixth Califor-
nia Islands Symposium. National Park Service Tech-
nical Publication CHIS-05-01. Institute for Wildlife
Studies, Arcata, CA.
KNAPP, D.A. 2010a. Changes in oak distribution and density
by decade on Santa Catalina Island, 1943 to 2005.
Pages 47–52 in D.A. Knapp, editor, Oak eco system
restoration on Santa Catalina Island, California:
proceedings of an on-island workshop, February 2–4,
2007. Catalina Island Conservancy, Avalon, CA.
KNAPP, D.A. 2010b. Ecosystem restoration on Santa Cata -
lina Island: a synthesis of resources and threats.
Pages 135–216 in D.A. Knapp, editor, Oak ecosystem
restoration on Santa Catalina Island, California: pro-
ceedings of an on-island workshop, February 2–4,
2007. Catalina Island Conservancy, Avalon, CA.
KRUIJER, H., N. RAES, AND M. STECH. 2010. Modelling
the distribution of the moss species Hypopterygium
tamarisci (Hypopterygiaceae, Bryophyta) in Central
and South America. Nova Hedwigia 91:399–420.
KÜCHLER, A.W. 1977. Natural vegetation of California.
Map in M.G. Barbour and J. Major, editors, Terres-
trial vegetation of California. Wiley-Interscience, New
York, NY.
KÜCHLER, A.W., AND I.S. ZONNEVELD, EDITORS. 1988. Veg -
etation mapping. Kluwer Academic Publishers,
Dordrecht.
LAPOLA, D.M., M.D. OYAMA, C.A. NOBRE, AND G. SAMPAIO.
2008. A new world natural vegetation map for global
change studies. Annals of the Brazilian Academy of
Sciences 80:397–408.
LIU, H., L. WANG, J. YANG, N. NAKAGOSHI, C. LIANG, W.
WANG, AND Y. LV. 2009. Predictive modeling of the
potential natural vegetation pattern in northeast
China. Ecological Research 24:1313–1321.
[LAR-IAC] LOS ANGELES REGION IMAGERY ACQUISITION
CONSORTIUM. 2006. LARIAC1 Archive. Los Angeles
County GIS Data Portal. https://egis3.lacounty.gov/
dataportal/lariac1-archive
MARINO, J., M. BENNETT, D. COSSIOS, A. IRIARTE, M.
LUCHERINI, P. PLISCOFF, C. SILLERO-ZUBIRI, L. VIL-
LALBA, AND S. WALKER. 2011. Bioclimatic constraints
to Andean cat distribution: a modelling application for
rare species. Diversity and Distributions 17:311–322.
MCDOWELL, W.G., A.J. BENSON, AND J.E. BYERS. 2014.
Climate controls the distribution of a widespread
invasive species: implications for future range
expansion. Freshwater Biology 59:847–857.
MILLSPAUGH, C.F., AND L.W. NUTALL. 1923. Flora of Santa
Catalina Island (California). Field Museum of Natural
History, Botanical Series 5:1–413.
MINNICH, R.A. 1980. Vegetation of Santa Cruz and Santa
Catalina Islands. Pages 123–138 in D.M. Power, editor,
The California islands: proceedings of a multidis -
ciplinary symposium. Santa Barbara Museum of
Natural History Santa Barbara, CA.
MINNICH, R.A. 1982. Grazing, fire, and the management
of vegetation on Santa Catalina Island, California.
Pages 444–449 in Dynamics and management of
Mediterranean-type ecosystems. Pacific Southwest
Forest and Range Experiment Station, Berkeley, CA.
MORAVEC, J. 1969. Succession of plant communities and
soil development. Folia Geobotanica et Phytotaxo-
nomica 4:133–164.
MORAVEC, J. 1998. Reconstructed natural versus potential
natural vegetation in vegetation mapping: a discussion
of concepts. Applied Vegetation Science 1:173–176.
NAZERI, M., N. MADANI, L. KUMAR, A.S. MAHINY, AND
B.H. KIABI. 2015. A geo-statistical approach to model
Asiatic cheetah, onager, gazelle and wild sheep
shared niche and distribution in Turan Biosphere
Reserve–Iran. Ecological Informatics 29:25–32.
NEWBOLD, T., T. READER, S. ZALAT, A. EL-GABBAS, AND
F. GILBERT. 2009. Effect of characteristics of butter-
fly species on the accuracy of distribution models in
LONGCORE ET AL.♦CATALINA ISLAND POTENTIAL NATURAL VEGETATION 631
632 WESTERN NORTH AMERICAN NATURALIST (2018), VOL. 78 NO. 4, PAGES 617–632
an arid environment. Biodiversity and Conservation
18:3629–3641.
NOGALES, M., A. MARTÍN, B.R. TERSHY, C.J. DONLAN, D.
VEITCH, N. PUERTA, B. WOOD, AND J. ALONSO. 2004.
A review of feral cat eradication on islands. Conser-
vation Biology 18:310–319.
PALMER, M.W. 1993. Putting things in even better order:
the advantages of canonical correspondence analy-
sis. Ecology 74:2215–2230.
PAPEŞ, M., A.T. PETERSON, AND G.V.N. POWELL. 2012. Vege-
tation dynamics and avian seasonal migration: clues
from remotely sensed vegetation indices and ecolog-
ical niche modelling. Journal of Biogeography 39:
652–664.
PAUDEL, S., G.W.T. WILSON, B. MACDONALD, T. LONG-
CORE, AND S.R. LOSS. 2016. Predicting spatial extent
of invasive earthworms on an oceanic island. Diver-
sity and Distributions 22:1013–1023.
PHILLIPS, S.J., R.P. ANDERSON, AND R.E. SCHAPIRE. 2006.
Maximum entropy modeling of species geographic
distributions. Ecological Modelling 190:231–259.
PHILLIPS, S.J., AND M. DUDÍK. 2008. Modeling of species
distributions with Maxent: new extensions and a
comprehensive evaluation. Ecography 31:161–175.
RAES, N., AND H. TER STEEGE. 2007. A null model for
significance testing of presence-only species distri -
bution models. Ecography 30:727–736.
RAMIREZ, A.R., R.B. PRATT, A.L. JACOBSEN, AND S.D. DAVIS.
2012. Exotic deer diminish post-fire resilience of
native shrub communities on Santa Catalina Island,
southern California. Plant Ecology 213:1037–1047.
RICK, T.C., T.S. SILLETT, C.K. GHALAMBOR, C.A. HOFMAN,
K. RALLS, R.S. ANDERSON, C.L. BOSER, T.J. BRAJE,
D.R. CAYAN, R.T. CHESSER, ET AL. 2014. Ecological
change on California’s Channel Islands from the
Pleistocene to the Anthropocene. BioScience 64:
680–692.
ROBERTS, D.W. 1986. Ordination on the basis of fuzzy set
theory. Vegetatio 66:123–131.
RODRÍGUEZ-SÁNCHEZ, F., AND J. ARROYO. 2008. Recon-
structing the demise of Tethyan plants: climate-driven
range dynamics of Laurus since the Pliocene. Global
Ecology and Biogeography 17:685–695.
ROWLAND, S.M. 1984. Geology of Santa Catalina Island.
California Geology 37:239–251.
SCHOENHERR, A.A., C.R. FELDMETH, AND M.J. EMERSON.
1999. Natural history of the islands of California.
University of California Press, Berkeley, CA.
SIMPSON, M., AND B. PROTS. 2013. Predicting the distri -
bution of invasive plants in the Ukrainian Carpathi-
ans under climatic change and intensification of
anthropogenic disturbances: implications for biodi-
versity conservation. Environmental Conservation
40:167–181.
SMITH, W.S.T. 1897. The geology of Santa Catalina Island.
California Academy of Sciences.
SOTO-BERELOV, M., P.L. FALL, S.E. FALCONER, AND E.
RIDDER. 2015. Modeling vegetation dynamics in the
Southern Levant through the Bronze Age. Journal of
Archaeological Science 53:94–109.
STEIN, E.D., S. DARK, T. LONGCORE, R. GROSSINGER, N.
HALL, AND M. BELAND. 2010. Historical ecology as a
tool for assessing landscape change and informing
wetland restoration priorities. Wetlands 30:589–601.
STRATTON, L. 2009. Restoration strategies for overcoming
limitations to scrub oak regeneration on Catalina
Island. Pages 185–200 in Proceedings of the 7th
California Islands Symposium. Institute for Wildlife
Studies, Arcata, CA.
SZABÓ, P., P. KUNEŠ, H. SVOBODOVÁ-SVITAVSKÁ, M.G. ŠVAR-
COVÁ, L. KŘÍŽOVÁ, S. SUCHÁNKOVÁ, J. MÜLLEROVÁ,
AND R. HÉDL. 2017. Using historical ecology to
reassess the conservation status of coniferous forests
in Central Europe. Conservation Biology 31:150–160.
TAYLOR, P.J., A. NENGOVHELA, J. LINDEN, AND R.M. BAX-
TER. 2016. Past, present, and future distribution of
Afromontane rodents (Muridae: Otomys) reflect cli-
mate-change predicted biome changes. Mammalia
80:359–375.
THORNE, R.F. 1967. A flora of Santa Catalina Island, Cali-
fornia. Aliso 6:1–77.
TICHÝ, L. 1999. Predictive modeling of the potential natural
vegetation pattern in the Podyjí National Park, Czech
Republic. Folia Geobotanica 34:243–252.
TINGLEY, R., M. VALLINOTO, F. SEQUEIRA, AND M.R. KEAR-
NEY. 2014. Realized niche shift during a global bio-
logical invasion. Proceedings of the National Acad-
emy of Sciences of the United States of America
111:10233–10238.
TULOWIECKI, S.J., AND C.P. LARSEN. 2015. Native Ameri-
can impact on past forest composition inferred from
species distribution models, Chautauqua County,
New York. Ecological Monographs 85:557–581.
TÜXEN, R. 1956. Die heutige potentielle natürliche Vege-
tation als Gegenstand der Vegetationskartierung.
Angewandte Pflanzensoziologie 13:4–42.
VAN DYKE, E., AND K. WASSON. 2005. Historical ecology
of a central California estuary: 150 years of habitat
change. Estuaries and Coasts 28:173–189.
VAN VUREN, D., AND B.E. COBLENTZ. 1987. Some ecologi-
cal effects of feral sheep on Santa Cruz Island, Cali-
fornia, USA. Biological Conservation 41:253–268.
VELÁSQUEZ-TIBATÁ, J., P. SALAMAN, AND C.H. GRAHAM.
2013. Effects of climate change on species distribu-
tion, community structure, and conservation of birds
in protected areas in Colombia. Regional Environ-
mental Change 13:235–248.
WESTMAN, W.E. 1983. Island biogeography: studies on
the xeric shrublands of the inner Channel Islands,
California. Journal of Biogeography 10:97–118.
WESTOFF, V., AND E. VAN DER MAAREL. 1978. The Braun-
Blanquet approach. Pages 287–399 in R.H. Whittaker,
editor, Classification of plant communities. Junk,
The Hague.
WOLLAN, A.K., V. BAKKESTUEN, H. KAUSERUD, G. GULDEN,
AND R. HALVORSEN. 2008. Modelling and predicting
fungal distribution patterns using herbarium data.
Journal of Biogeography 35:2298–2310.
ZERBE, S. 1998. Potential natural vegetation: validity and
applicability in landscape planning and nature con-
servation. Applied Vegetation Science 1:165–172.
ZHANG, Z.D., R.G. ZANG, AND M. CONVERTINO. 2013.
Predicting the distribution of potential natural vege-
tation based on species functional groups in frag-
mented and species-rich forests. Plant Ecology and
Evolution 146:261–271.
Received 28 February 2017
Revised 27 February 2018
Accepted 14 March 2018
Published online 17 December 2018
