The potential distribution of Zorotypus hubbardi caudell (Zoraptera: Zorotypidae) in North America, as predicted by ecological niche modeling

2006·5 MB·

by  Ismael A Hinojosa-Díaz

PROC. ENTOMOL. SOC. WASH.
108(4), 2006, pp. 860-867
THE POTENTIAL DISTRIBUTION OF ZOROTYPUS HUBBARDI CAUDELL
(ZORAPTERA: ZOROTYPIDAE) IN NORTH AMERICA, AS PREDICTED BY
ECOLOGICAL NICHE MODELING
ISMAEL A. HINoJOSA-DiAz, ELISA BONACCORSO, AND MICHAEL S. ENGEL
(IAH, MSE) Division of Entomology, Natural History Museum, 1460 Jayhawk
Boulevard, Snow Hall, and Department of Ecology and Evolutionary Biology,
University of Kansas, Lawrence, KS 66045-7523, U.S.A. (e-mail: [email protected]);
(EB) Division of Ornithology, Natural History Museum, 1345 Jayhawk Boulevard,
Dyche Hall, and Department of Ecology and Evolutionary Biology, University of
Kansas, Lawrence, KS 66045-7163, U.S.A.
Abstract.4Owing to their minute sizes, frail wings, and low abundances in
entomological collections, zorapterans have been assumed to be highly endemic and
poor dispersers. Wide distributions of some species are thought to be induced by
human activity. Herein, we use the Genetic Algorithm for Rule-set Prediction
(GARP) to model the potential distribution of Zorotypus hubbardi Caudell, one of 32
living species of the order Zoraptera. According to the model, the potential
distribution of Z. hubbardi covers the southeastern U.S., with western limits in
eastern Texas, eastern Oklahoma, and southeastern Kansas, and northern limits in
southeastern Iowa in the West and southernmost New York in the East; the model9s
predictions exclude Z. hubbardi from the Appalachian Mountains. The northern and
western boundaries of the potential distribution show that the ecological conditions
for the natural occurrence of the species are present, and that the presence of colonies
would not be solely dependent on coexistence with humans. New feral records from
the areas shown in the potential distribution would establish the certainty of the
model.
Key Words: ecological niche, Zorotypus, Polyneoptera, predictive modeling,
zorapteran
Zorotypus hubbardi Caudell is one of insects as a whole; however, there are
two species of Zoraptera occurring in phylogenetic and biological aspects that
the United States. This order of minute make this group of insects worthy of
(2-3.5 mm in total length), soft-bodied study. The monophyly of Zoraptera is
insects, contains a single family with 32 supported by several characters, among
living species restricted principally to them a unique pattern of wing venation,
tropical latitudes; only four species occur dimerous tarsi, and mating via <mating
north of the Tropic of Cancer (two in the hook= (Engel 2003a, b, 2004). However,
United States, and two in Tibet). The the phylogenetic placement of the order
low diversity of zorapteran species con- among hemimetabolous insects is con-
trasts with the overwhelming diversity of troversial. In the best-supported hypoth-
VOLUME 108, NUMBER 4 861
esis, Zoraptera is a sister group to METHODS
Embiodea (a.k.a., webspinners) (Engel
Occurrence points and environmental
and Grimaldi 2000, Grimaldi and Engel
data.4We based our ecological niche
2005).
models on 79 occurrence points of Z.
These enigmatic insects live gregari-
hubbardi in its native range. Occurrence
ously under the bark of decaying logs;
points were obtained from _ literature
they feed principally on fungal hyphae
records, and georeferenced to the nearest
and spores, but also prey on nematodes
minute using GEOLocate (Rios and Bart
or minute arthropods. Two distinct
2005). Environmental coverages used in
morphs (sometimes referred to as
the process included 24 electronic maps
<castes,= which should not be confused
summarizing aspects of topography
with the true castes of social insects) are
(U.S. Geological Survey9s Hydro-lK
present within each species and in the
data set http://edcdaac.usgs.gov/gtopo30/
same colony. Blind, wingless individuals,
hydro/) and climate (http://biogeo. berkeley.
which are more common during the life
edu/worldclim/down.htm). Because most
of the colony, and winged-eyed, disper-
occurrence points dated from the 1950's,
sive individuals, which are generally rare
we used a dataset of climatic coverages
in the colony; the latter individuals are
for the time period 1930-1960. All
produced when resources become de-
coverages were resampled at 0.1° pixel
pleted and the colony becomes crowded.
resolution and clipped to eastern North
Dispersive individuals shed their wings
America for model development. Com-
after arriving at a new log and found
pilation of environmental datasets and
new colonies. The northernmost limit of
visualization of the models was achieved
the distribution of Z. hubbardi (in south-
using Arc View (ESRI 2002).
central and southeastern U.S.), is likely
Ecological niche modeling.4To de-
the result of human activities, as most
velop ecological niche models, we used
colonies are found in sawdust piles
the Genetic Algorithm Rule-set Predic-
(rather than natural! logs), relying on
tion (GARP) (Stockwell and Noble
the warmth of the decaying material that
1992, Stockwell 1999, Stockwell and
is available for just a few seasons. As the
Peters 1999). This approach focuses on
sawdust pile becomes unsuitable for the
modeling ecological niches; the conjunc-
colony, dispersal is required; as a result,
the northernmost colonies become ex- tion of ecological conditions within
tinct and subsequent reintroductions are a species is able to maintain populations
required to maintain the presence of the without immigration (Grinnell 1917,
species (Engel 2003a). Formerly, zorap- MacArthur 1972). GARP relates ecolog-
terans were thought to have restricted ical characteristics of known occurrence
dispersal capabilities making them high- points to those of points sampled from
ly endemic, but wide distributions (as for the rest of the study region, developing
Z. hubbardi) and occurrence on oceanic a <series; jof, qdecision, rules; that, best
islands, suggest a different scenario. summarize those factors associated with
Herein, we asses the potential distribu- the species presence (Feria and Peterson
tion of Z. hubbardi, which in turn will 2002).
serve to formulate hypotheses about Occurrence points are first divided
the natural or introduced condition of into training and test data sets (Stock-
the northernmost colonies, as well as well and Peters 1999). GARP, then
ecological factors necessary for the applies an iterative process of rule
persistence of the species in_ specific selection, evaluation, testing, and incor-
areas. poration or rejection (Stockwell and
862 PROCEEDINGS OF THE ENTOMOLOGICAL SOCIETY OF WASHINGTON
Peters 1999). It chooses a method from graphic distribution of suitable condi-
a set of possibilities (e.g., logistic re- tions (i.e., a potential geographic
gression, bioclimatic rules) applied to the distribution for the species). Those mod-
training data, and a rule is developed els can also be projected onto alternate
or evolved; rules may evolve by a num- landscapes in space (Peterson and Vie-
ber of means (e.g., truncation, point glais 2001) and time (Rice et al. 2003).
changes, crossing-over) to maximize Herein, we projected the model for the
predictivity. Predictive accuracy (for in- native distribution of the species back
trinsic use in the model refinement) is onto eastern North America to estimate
then evaluated based on 1250 points the potential distribution of the species in
sampled randomly from the study region the whole region.
as a whole (Stockwell and Peters 1999). Since GARP uses a 8<8random-walk=9
The change in predictive accuracy from approach and predictions vary among
One iteration, to the <mext is\ used 8to runs, we generated 100 models for the
evaluate whether a particular rule should species, | Jand); selected) them L0G bese?
be incorporated into the model, and the models (those falling in an optimal
algorithm runs either 1000 iterations or combination of error measures; Ander-
until convergence occurs (Stockwell and son et al. 2003). These 10 models were
Peters 1999). summed to obtain maps for the potential
Extensive testing has demonstrated distribution of Z. hubbardi across eastern
GARP9s ability to predict ecological North America.
and geographic distributions of species Testing ecological niche models.4The
in diverse ecological, geographic, and ecological niche models generated were
taxonomic contexts (Anderson et al. tested as follows. Available distribution-
2003; Peterson and Cohoon 1999; Peter- al points were divided onto either side of
son 2001; Stockwell and Peterson 2002, the median latitude and median longi-
2003). Herein, we used a desktop version tude. Two of the quadrants (northwest
of GARP (http://www.lifemapper.org/ and southeast) were used to develop
desktopgarp). models in order to predict the distribu-
To reduce the environmental data set tions of points in the other two quad-
to coverages providing the highest pre- rants (southwest and northeast) and vice
dictivity accuracy, we performed a variety versa. Then, the proportional area pre-
of jackknife manipulations (Peterson et dicted present < number of extrinsic
al. 2003). We ran multiple iterations of data points was used as a random
models omitting coverages, or sets of expectation of successful prediction of
coverages, systematically. Then, we ex- points if no non-random association
amined the correlation between the in- existed between prediction and _ test
clusion and exclusion of each coverage points; a chi-square approach (1 df)
and omission error, and removed those was used to test the significance of the
coverages that had a strong detrimental departure from random _ expectations
contribution to model quality (when r > (Peterson and Shaw 2003). This test
O05): represents a challenge for model predic-
Ecological niche models developed tivity because it asseses the ability of
with GARP can be projected onto land- models to predict into broad, unsampled
scapes via spatial queries to detect regions.
conditions fitting those modeled as the
species9 niche. Projection of models onto RESULTS
the landscape from which input occur- According to the jackknife analyses,
rence points were taken, indicates geo- all topographic layers were highly corre-
VOLUME 108, NUMBER 4 863
Eig. I Prediction of the potential distribution of Zorotypus hubbardi in eastern North America. Dots
show occurrence data on which ecological niche models were based. Darker shading indicates greater
model agreement in predicting potential presence.
lated with high omission error (all The models developed to predict the
correlations r > 0.05); therefore, they potential distribution of Z. hubbardi
were excluded from further manipula- identified broad areas throughout the
tions. After all analyses, seven environ- southeastern United States, extending
mental layers (frequency of ground frost, centrally to the eastern extreme of Texas
annual precipitation, frequency of wet (particularly bordering the Gulf Coast),
days, and annual minimum, mean, and eastern half of Oklahoma, and south-
maximum temperature) were chosen to eastern Kansas. The northern limits
develop the final ecological niche models. reach southeastern Iowa in the West
Table 1. Summary of statistical tests of model accuracy in the quadrant test.
Proportional Area Expected Expected Observed Observed
Test! Predicted Present? N Correct Incorrect Correct Incorrect X2 12
NS predicts SN 30% 42 13 29 19 23 4.13 0.04
SN predicts NS 17% 37 6 3) 1] 26 4.48 0.03
! Two of the quadrants (northwest and southeast, NS) predict the distributions of points in the other
two quadrants (southwest and northeast, SN) and vice versa.
2 Area predicted present by all 10 models generated/total area of analysis.
864 PROCEEDINGS OF THE ENTOMOLOGICAL SOCIETY OF WASHINGTON
and the southernmost areas of New
York in the East. The model excludes
the Appalachian Mountains (Fig. 1).
The ecological niche models were
highly predictive of the potential distri-
bution of Z. hubbardi, based on the
quadrant test of model predictivity (Ta-
ble 1). In Fig. 2, the quadrant test is
illustrated: 42 and 37 points were avail-
able in the two pairs of quadrants. In the
first test, based on area predicted pres-
ent, 12 of 42 points were expected to be
predicted present correctly; however, 20
were predicted correctly (chi-square test,
P < 0.05). The reciprocal test had an
expected » correct prediction <of Gor
37 points, but 11 of 37 points were
correctly predicted (chi-square test, P <
0:05):
Visualization of patterns of distribu-
tion of species in ecological space pro-
vides a fertile ground for thinking about
why species are distributed in some areas
and not in others (Peterson and Shaw
2003). As an example, Fig. 3 shows the
ecological space available across the
species range in terms of Mean Annual
Temperature and Mean Annual Precip-
itation, and Mean Annual Temperature
and Frequency of Wet Days (light gray
diamonds), as well as the ecological
space occupied by the species according
to its potential distribution (dark gray
squares). In both cases, Z. hubbardi is
present in zones with intermediate values
for these three variables. It would be
interesting to compare this pattern with
other Zorotypus species to study whether
they overlap or replace each other in
Fig. 2. Illustration of the quadrant test used to
assess model predictivity for Zorotypus hubbardi. these ecological dimensions.
In panel A, we show the available distributional
points for the species, divided in two groups DISCUSSION
(northwest and southeast, southwest and north-
Zorapterans have been seen rarely by
east). In panel B, points in quadrants northwest
most entomologists and the few collec-
and southeast predict the distribution of points in
quadrants southwest and northeast. Panel C shows tion records have led to the assumption
the reciprocal test. of high endemism and poor powers of
dispersal. To account for these <8limited
dispersal capabilities,9 past authors have
suggested that the flimsy, narrow, pad-
VOLUME 108, NUMBER 4 865
(x1m0m)
ipitation
0 50 100 150 200 250 300
MAPrneenacun a l
160
120
80
d(Wxae1yt0s )
40
0 50 100 150 200 250 300
Mean Annual Temperature (Cx10)
Fig. 3. Visualization of modeled distribution of Zorotypus hubbardi in environmental space, viewed in
two bivariate plots each of two environmental dimensions. Dark gray diamonds indicate overall
availability in the area of analysis; light gray squares indicate the distribution of the species in ecological
space relative to availability.
dle-shaped wings of alates preclude tions for most of the species in the order
effective flight. Indeed, such assumptions (e.g., Huang 1980). Conversely, the low
have been considered the causal factors species diversity within the order and
for the <narrow= geographical distribu- poor sampling of populations has had
866 PROCEEDINGS OF THE ENTOMOLOGICAL SOCIETY OF WASHINGTON
more of an impact on the scattered over large areas. With global climate
geographical records for zorapterans. change and suitable human intervention
This is to say that the <narrow= dis- (e.g., sawdust piles near homes or lumber
tributions of most species is likely an yards for artificial warmth during win-
artifact of poor sampling (vide Engel ters), populations gradually push the
2001, 2004). When collectors develop extremes of their environmental limits.
a search image for zorapterans, they New records of feral colonies (1.e., those
tend to find more colonies and broader apart from human habitation and artifi-
geographic distributions of species than cial heat sources) from the areas shown
previously thought (e.g., Engel 2001, in the potential distribution, particularly
2004). Such evidence indicates that zor- the extremes of the range, would provide
apterans may be more mobile than greater support for our model.
entomologists have appreciated. The
wings of zorapterans are no less adept
ACKNOWLEDGMENTS
than those of many minute insects (e.g.,
Partial support for this study was
thrips, microhymenoptera) already known
provided by NSF EF-0341724 (to
to be capable dispersers. Furthermore,
MSE). This is contribution 3440 of the
owing to their small sizes, zorapterans
Division of Entomology, Natural Histo-
can be captured in wind currents and
ry Museum, University of Kansas. We
carried great distances. Thus, there are no
are grateful to A. T. Peterson anda!
a priori reasons to assume that zorapterans
Trueb for their useful comments on this
are incapable of regular dispersal.
manuscript.
Zorotypus hubbardi is the best known
species of zorapterans in terms of bi-
ology and also the one with the greatest LITERATURE CITED
number of locality records. Its potential Anderson, R. P., D. Lew, and A. T. Peterson. 2003.
distribution shows that the ecological Evaluating predictive models of species9 dis-
conditions suitable for the natural oc- tributions: Criteria for selecting optimal mod-
els. Ecological Modeling 162: 2114232.
currence of the species are present in
Engel, M. S. 2001. New neotropical records for
a widespread area, and suggest that
three Zorotypus species (Zoraptera: Zorotypi-
zorapteran dispersal capabilities have
dae). Entomological News 112: 278-280.
been underestimated. For Z. hubbardi, . 2003a. Zoraptera, pp. 1201-1203. Jn Resh,
in particular, the northern and western V. H. and R. T. Cardé, eds. Encyclopedia of
boundaries of the potential distribution Insects. Academic Press, San Diego.
produced by our model, show that 2003b. Phylogeny of the Zoraptera.
Entomologische Abhandlungen 61: 1474
occurrence of the species in those areas
148.
is feasible given the ecological variables
. 2004. Zoraptera, pp. 637-640. /n Llorente-
analyzed in this study, and indicate that Bousquets, J., J. J. Morrone, O. Yamez-
human activity would not be the only Ordonez, and I. Fernandez-Vargas, eds. Bio-
source for the presence of colonies. As diversidad, Taxonomia y Biogeografia de
Artropodos de México: Hacia una Sintesis de
suggested by Engel (2003a), zorapterans
su Conocimiento, Volumen IV. UNAM, Meéx-
are likely quite mobile and, at the
ico, D.F.
extremes of zorapteran distribution, col-
Engel, M. S. and D. A. Grimaldi. 2000. A winged
onies are being established regularly in Zorotypus in Miocene amber from the Domin-
favorable seasons and extirpated during ican Republic (Zoraptera: Zorotypidae), with
harsh winters. Indeed, populations may discussion on relationships of and within the
order. Acta Geologica Hispanica 35: 149-164.
be highly mobile and spreading regularly
Environmental Systems Research Institute (ESRD),
throughout areas of suitable habitability,
Inc. 2002. ArcView 3.3. 380 New York St.,
thereby maintaining specific integrity Redlands.
VOLUME 108, NUMBER 4 867
Feria, T. P. and A. T. Peterson. 2002. Using point Peterson, A. T., M. Papes, and D. A. Kluza. 2003.
occurrence data and inferential algorithms to Predicting the potential invasive distributions
predict local communities of birds. Diversity of four alien plant species in North America.
and Distributions 8: 494S6. Weed Science 51: 863-868.
Grimaldi, D. and M. S. Engel. 2005. Evolution of Rice, N. H., E. Martinez-Meyer, and A. T.
the Insects. Cambridge University Press, Cam- Peterson. 2003. Ecological niche differentiation
bridge. in the Aphelocoma jays: A phylogenetic per-
Grinnell, J. 1917. Field tests of theories concerning spective. Biological Journal of the Linnaean
distributional control. American Naturalist 51: Society 80: 369-383.
115-128. Rios, N. and H. Bart. 2005. GEOLocate Georefer-
Huang, F.-S. 1980. The order Zoraptera and its encing Software for Natural History Collec-
geographical distribution, pp. 1087-1090. In tions. URL _http://www.museum.tulane.edu/
Liu, D.-S., ed. Geological and Ecological geolocate
Studies of Qinghai-Xizang Plateau: Proceed- Stockwell, D. R. B. 1999. Genetic algorithms II,
ings of Symposium on Quinghai-Xizang (Ti- pp. 123-144. Jn Fielding, A. H., ed. Machine
bet) Plateau. Science Press, Beijing. Learning Methods for Ecological Applica-
MacArthur, R. H. 1972. Geographical Ecology: tions. Kluwer Academic Publishers, Boston.
Patterns in the Distribution of Species. Harper Stockwell D> IR= BS and Mie Re Nobles 1992.
& Row, New York. Induction of sets of rules from animal distri-
Peterson, A. T. 2001. Predicting species9 geograph- bution data: A robust and informative method
ic distributions based on ecological niche of analysis. Mathematics and Computers in
modeling. Condor 103: 599-605. Simulation 33: 385-390.
Peterson, A. T. and K. C. Cohoon. 1999. Sensitiv- Stockwell, D. R. B. and D. P. Peters. 1999. The
ity of distributional prediction algorithms to GARP modeling system: Problems and solu-
geographic data completeness. Ecological tions to automated spatial prediction. Interna-
Modeling 117: 159-164. tional Journal of Geographic Information
Peterson, A. T. and J. Shaw. 2003. Lutzomyia Systems 13: 143-158.
vectors for cutaneous leishmaniasis in southern Stockwell, D. R. B. and A. T. Peterson. 2002.
Brazil: Ecological niche models, predicted Effects of sample size on accuracy of species
geographic distributions, and climate change distribution models. Ecological Modeling 148:
effects. International Journal for Parasitology 1-13.
33: 919-931. 44. 2003. Comparison of resolution of
Peterson, A. T. and D. A. Vieglais. 2001. Predicting methods used in mapping biodiversity patterns
species invasions using ecological niche mod- from point occurrence data. Ecological Indi-
eling. BioScience 51: 363-371. cators 3: 213-221.