Available via license: CC BY 4.0
Content may be subject to copyright.
1
Assessing the ecological niche and invasion potential of the Asian giant hornet
1
2
Gengping Zhua,*, Javier Gutierrez Illana, Chris Looneyb, David W. Crowdera
3
a Department of Entomology, Washington State University, Pullman, WA 99164, USA
4
b Washington State Department of Agriculture, Olympia, WA, 98501, USA
5
* gengping.zhu@wsu.edu
6
7
Author Contributions
8
All authors designed the study and wrote the manuscript, GZ conducted the analyses
9
10
Competing Interest Statement: None
11
12
This file includes:
13
Main Text
14
Figures 1 to 4
15
16
Word Count: 3484
17
18
Reference Count: 33
19
20
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
2
Abstract
21
The Asian giant hornet (Vespa mandarinia) is the world’s largest hornet. It is native to East Asia,
22
but was recently detected in British Columbia, Canada, and Washington State, USA. Ves pa
23
mandarinia are an invasion concern due to their potential to negatively affect honey bees and act
24
as a human nuisance pest. Here, we assessed effects of bioclimatic variables on V. mandarinia
25
and used ensemble forecasts to predict habitat suitability for this pest globally. We also simulated
26
potential dispersal of V. mandarinia in western North America. We show that V. mandarinia are
27
most likely to invade areas with warm to cool annual mean temperature but high precipitation,
28
and could be particularly problematic in regions with these conditions and high levels of human
29
activity. We identified regions with suitable habitat on all six continents except Antarctica. The
30
realized niche of introduced populations in the USA and Canada was small compared to native
31
populations, implying high potential for invasive spread into new regions. Dispersal simulations
32
showed that without containment, V. mandarinia could rapidly spread into southern Washington
33
and Oregon, USA and northward through British Columbia, Canada. Given its potential negative
34
impacts, and the capacity for spread within northwestern North America and worldwide, strong
35
mitigation efforts are needed to prevent further spread of V. mandarinia.
36
37
Key words: ecological niche model, climate suitability, human disturbance, ensemble forecast,
38
pest management, dispersal
39
40
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
3
Introduction
41
The Asian giant hornet (Vespa mandarinia Smith) is the world’s largest hornet and is native to
42
temperate and sub-tropical Eastern Asia (Fig. 1A), where it is a predator of honey bees and other
43
insects (Matsuura & Sakagami 1973; Archer 1995; McGlenaghan et al. 2019). Coordinated
44
attacks by V. mandarinia on beehives involve pheromone marking to recruit other hornets,
45
followed by rapid killing of worker bees until the hive is destroyed (McClenaghan et al. 2019).
46
Japanese honey bees (Apis cerana) co-evolved with V. mandarinia and have defensive behaviors
47
to counter these attacks, including recognizing and responding to marking pheromones and “bee
48
ball” attacks on hornet workers (Sugahara et al. 2012; McGlenaghan et al. 2019). Apis mellifera,
49
the European honey bee, however, has no co-evolutionary history with V. mandarinia and lacks
50
effective defensive behaviors, making it highly susceptible to attack (McGlenaghan et al. 2019).
51
In September 2019, a nest of V. mandarinia was found on Vancouver Island in Canada, and
52
two workers were found 90 km away in Washington State, USA, later that year (USDA 2019)
53
(Fig. 1B). The introduction of Asian giant hornet into western North America is concerning
54
because of the vulnerability of A. mellifera, which is widely used for crop pollination, to hornet
55
attacks. Predation by V. mandarinia on A. mellifera in Asia causes major losses (McGlenaghan et
56
al. 2019). Vespa mandarinia is also medically significant, and can deliver painful stings and large
57
doses of cytolytic venom. Multiple stings can be fatal even in non-allergic individuals, although
58
recent mortality rates are much lower than the historic reports of more than 30 yearly deaths in
59
Japan (Yanagawa et al. 2007). Currently, V. mandarinia is listed as a quarantine pest of the
60
United States and efforts are underway to prevent establishment and spread (USDA 2019).
61
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
4
Invasions from species such as V. mandarinia are governed by arrival, establishment, and
62
spread (Liebhold & Tobin 2008). Ecological niche models, which involve model calibration
63
using climate variables in native ranges, followed by extrapolation to introduced areas, are often
64
used to assess habitat suitability for invasive species (Leibhold & Tobin 2008; Peterson et al.
65
2013). Invasions can be particularly problematic in regions with high human activity, which can
66
also facilitate invasions through transport of introduced species. To assess spread of invasive
67
species, models can also simulate processes (Liebhold & Tobin 2008; Engler et al. 2012). Models
68
thus guide efforts to prevent establishment and spread, which are cost-effective early in invasions
69
(Liebhold & Tobin 2008).
70
It is not yet clear if V. mandarinia is established in North America, and federal and local
71
agencies are implementing trapping and monitoring programs to identify areas of introduction
72
and prevent establishment and spread (USDA 2019). However, several factors that could guide
73
mitigation efforts remain unknown, including the potential habitat suitability for V. mandarinia.
74
Moreover, the potential rate of population dispersal into new areas is poorly understood. In
75
Europe, an invasion by the congener V. velutina has expanded via natural and human-assisted
76
dispersal from 19 to nearly 80 km per year (Bertolino et al. 2016; Robinet et al. 2017). Here, we
77
assess these questions by modeling responses of V. mandarinia to bioclimatic variables in the
78
native range and extrapolating to introduced ranges. We also used dispersal simulations to
79
estimate potential rates of spread throughout western North America. These complementary
80
approaches can guide efforts to prevent the establishment and spread of this invasive species.
81
82
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
5
Material and Methods
83
Environmental factors affecting occurrence of V. mandarinia
84
We first assessed relationships between occurrence of V. mandarinia and environmental
85
factors. Occurrence data were attained with the “spocc” package in R (R Core Team 2020) from
86
the Global Biodiversity Information Facility, Biodiversity Information Serving Our Nation,
87
Integrated Digitized Biocollections, and iNaturalist (Scott et al. 2017) (Fig. 1). Additional data
88
were collected from published studies (Archer 1995; Lee 2010). Occurrence records located
89
within oceans or without geographic coordinates were removed. 422 unique records from V.
90
mandarinia’s native range in Asia (Japan, South Korea, Taiwan) were obtained (Fig. 1A). Of
91
these, 200 were filtered out by enforcing a distance of 10 km between records (Fig. 1A); we used
92
this filtering process because ecological niche models are sensitive to sample bias (Warren &
93
Seifert 2011). Our assembled 222 records from east Asia are consistent with published records
94
(Archer 1995; Lee 2010), suggesting we effectively captured the distribution of V. mandarina.
95
Vespine wasps have high endothermic capacity and thermoregulatory efficiency, and can
96
survive broad temperature ranges (Käfer et al. 2012). To determine climate factors that constrain
97
V. mandarinia, we obtained 7 Worldclim variables (Fick & Hijmans 2017): (i) annual mean
98
temperature, (ii) mean diurnal range, (iii) max temperature of warmest month, (iv) minimum
99
temperature of coldest month, (v) annual precipitation, (vi) precipitation of wettest and (viii)
100
driest months (Bio14); we also considered annual mean radiation (Fig. S1A). Although some of
101
these variables were correlated (Fig. S1B), highly correlated variables have little impact on
102
ecological niche models that account for redundant variables (Feng et al. 2019).
103
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
6
After selecting variables, we used generalized linear models (GLM) with Bernoulli errors
104
to model the probability of occurrence of V. mandarinia as a function of each bioclimatic factor.
105
This approach was used to minimize the chances of overfitting models, and Hosmer Lemeshow
106
goodness of fit test were used to evaluate GLM model performance (Hosmer et al. 1989). Rather
107
than plotting a single partial response curve (i.e., fitting response curves for specific predictors
108
while keeping the other predictors at their mean value), we adopted inflated response curves to
109
explore species-environment relationships along the entire gradient while keeping the other
110
predictors at their mean, minimum, median, maximum, and quartile values (Zurell et al. 2012).
111
112
Realized niche modeling of native and introduced populations
113
After assessing environmental factors affecting V. mandarinia occurrence, we next assessed
114
realized niches occupied by native and introduced populations. Given that only 4 occurrence
115
points exist in the introduced range of western North America, two of which are within 10 km,
116
we could not use a strict test of whether realized niches shifted during the introduction of V.
117
mandarinia (i.e., niche equivalency and similarity test; Warren et al. 2010). Rather, we used
118
minimum ellipsoid volumes to display and compare the two realized niches; this technique
119
provides a clear vision of niche breadth for two populations and their relative positions in
120
reduced dimensions (Qiao et al. 2016). We generated three environmental dimensions that
121
summarized 90% of overall variations in the 8 global bioclimatic dimensions using principle
122
component analysis in NicheA version 3.0 (Qiao et al. 2016).
123
124
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
7
Ecological niche modeling
125
We used classical niche models to assess worldwide habitat suitability for potential spread
126
of V. mandarinia (Peterson et al. 2013). Given that uncertainty exists with any individual model,
127
we used an ensemble approach that averaged predictions of five algorithms: (i) generalized
128
additive models, (ii) general boosted models, (iii) generalized linear models; (iv) random forests,
129
and (v) maximum entropy models. Such ensemble models have been proposed as a consensus
130
approach to more effectively estimate climate suitability, achieve a higher predictive capacity,
131
and reduce uncertainty (Araújo & New 2007; Thuiller et al 2009; Zhu & Peterson 2017). To
132
build models, 50% of observed records were used for model training and 50% for validation
133
(Tsoar et al. 2007). We used a “random” method in biomod2 to select 10,000 pseudo-absence
134
records from “accessible” areas of V. mandarinia in Asia, which were delimited by buffering
135
minimum convex polygons of observed points at 400 km (Owen et al. 2013). This selection of
136
pseudo-absence records improves ecological niche model performance (Phillips & Dudík 2008).
137
For evaluation of models, we used Area Under the Curve (AUC) of Receiver Operating
138
Characteristic (ROC) plots as a measure of model fit (Jiménez-Valverde et al. 2012). AUC has
139
been criticized in niche model literature, and inference upon its values should be taken cautiously
140
as we didn’t have reliable absence data (Leroy et al. 2012). However, here we simply tested
141
niche model discriminability in native areas and not introduced areas. Final niche models were
142
fitted using overall trimmed occurrence points for combination with footprint and displaying.
143
Habitat modification and disruption has been linked to invasiveness in some Vespidae, and
144
invasions could be particularly problematic in regions with high human activity (Beggs et al.
145
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
8
2011, Robinet et al. 2017). In its native range, V. mandarinia is able to colonize green areas in
146
cities, although at lower abundance than other vespine wasps (Choi et al. 2012). Human-assisted
147
movement has affected expansion of V. velutina in Europe, and may also affect V. mandarinia
148
(Robinet et al 2007). Models that combine climate suitability with measures of human activity
149
may provide more accurate estimates of site vulnerability to colonization, particularly arrival and
150
establishment processes (Liebhold & Tobin 2008). We measured human footprint as an indicator
151
of human-mediated disturbances, a metric that combines population pressure and human
152
infrastructure. We combined human footprint with climate suitability using a bivariate mapping
153
approach (Fig. 2). All variables selected for analyses were used at a resolution of 5 arcmin.
154
155
Dispersal simulation
156
Vespa mandarinia is a social insect that forms colonies with one queen and many workers,
157
and population dispersal is mediated by the spread of queens. Workers typically fly 1 to 2 km
158
from their nest when foraging, although they can forage up to 8 km (Matsuura & Sakagami
159
1973). Data on queen dispersal appears to be unknown, but V. mandarinia queens are likely to
160
have flight capacity greater than workers. Flight mill simulations with the congener V. velutina
161
suggest that queens can fly 18 km in a single day (Robinet et al 2017), although flight distance
162
under field conditions is likely to be smaller.
163
To simulate potential spread of V. mandarinia based on these dispersal capacities and
164
occurrence points in western North America, we used the “MigClim” package (Engler et al.
165
2012). This approach simulates species expansion using a species’ occurrence as well as habitat
166
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
9
suitability and different dispersal scenarios. Short-distance dispersal considers innate dispersal of
167
a species to move through diffusion-based processes, whereas long-distance dispersal considers
168
passive dispersal over long distance, such as dispersal by hitchhiking on human activity (Engler
169
et al. 2012). MigClim uses a dispersal step as a basic time unit to simulate the dispersal, with
170
dispersal steps often equal to one year since most organism dispersal occurs yearly or can be
171
modeled as such, particularly for social insects where queens form colonies only once a year
172
(Engler et al. 2012). We ran a simulation with a total of 20 dispersal steps for V. mandarinia.
173
In our simulations, we created combined suitability using climate suitability from ensemble
174
models and human footprint. We then chose 3 different dispersal scenarios for simulations: (i)
175
short-distance dispersal only, (ii) long-distance dispersal only, and (iii) both short- and
176
long-distance dispersal. These three scenarios seek to capture both biological and
177
human-mediated dispersal potential of V. mandarinia, as MigClim does not account for
178
population demography (Engler et al. 2012). Simulations of short-distance dispersal were based
179
on physical barriers and the dispersal kernel, which is the dispersal probability as a function of
180
distance, whereas long-distance dispersal simulations depend on frequency of movement and
181
distance range. MigClim uses a dispersal step as a basic time unit to simulate the dispersal, with
182
dispersal steps often be equal to one year since most organism dispersal occurs yearly or can be
183
modeled as such, particularly for social insects where queens form colonies only once a year
184
(Engler et al. 2012).
185
We ran simulations with 20 dispersal steps. In MigClim, the dispersal kernel is the dispersal
186
probability as a function of distance (Pdisp) and the propagule production potential (Pprop). Our
187
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
10
raster data had a resolution of 5 arcmin (
≈
5.5 km); we defined short-distance dispersal as less
188
than 6 pixels (~ 33km). Dispersal more than 6 pixels was considered long-distance dispersal,
189
which had a maximum 20 pixels (~110km). We used a dispersal kernel of 1.0, 0.4, 0.16, 0.06,
190
and 0.03 pixel for short-distance dispersal, which is an average of 10 km/dispersal step, with a
191
maximum of 33 km. We set Pprop as 1 since V. mandarinia is a social insect and we assumed that
192
the probability of a source cell to produce propagules is 100%. We assumed there were no
193
barriers to either short- or long-distance dispersal.
194
195
Results and Discussion
196
Generalized linear models showed no significant differences between models fit to the 8
197
environmental variables and observed data (
χ
2 = 8.2, df = 8, P = 0.41). We show V. mandarinia is
198
most likely to occur in regions with low to warm annual mean temperatures and high annual
199
precipitation (Fig. 2). However, our models show that they can tolerate broad temperature ranges
200
(Fig. 2, S3), and that they are not particularly sensitive to radiation and extremes of precipitation
201
(Fig. S3). The most suitable habitats are predicted to be in regions with maximum temperature of
202
39 °C in the warmest month (Fig. S3). Our results thus support the existence of a thermal
203
threshold beyond which V. mandarina would be unable to establish, and this could be crucial for
204
management and policy making in case of a prolonged invasion of the hornet in North America.
205
The minimum ellipsoid volumes show that the realized niche of introduced individuals in
206
western North America were nested within the realized niche of native populations (Fig. 3). As
207
the introduced locations represent a small fraction of the realized niche occupied by native
208
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
11
populations, there is widespread potential for the introduced range to expand without mitigation
209
(Fig. 3). However, the contrasting volume sizes occupied by native and introduced populations
210
may simply be due to the limited number of occurrences outside of the native range (4 points)
211
rather than any reduction in the niche space available to introduced populations.
212
Our ecological niche models showed excellent performance in discriminability evaluations
213
(generalized additive model [GAM]: AUC = 0.89; general boosted model [GBM]: AUC = 0.93;
214
generalized linear model [GLM]: AUC = 0.91; Maxent: AUC = 0.93; Random Forest [RF]: AUC
215
= 0.91). However, the five niche models had variability in habitat suitability across the globe
216
(Fig. S4), and the ensemble model (Fig. 2) had better discriminability and outperformed these
217
individual models (AUC = 0.94). Outside of the native area, our ensemble models captured
218
detection points in North America as occurring in regions with highly suitable habitat (Fig. 2).
219
The ensemble models suggested that suitable habitat for V. mandarinia exists along much of
220
the coastline of western North America as most of the eastern USA and adjacent parts of Canada,
221
much of Europe, northwestern and southeastern South America, central Africa, eastern Australia,
222
and most parts of New Zealand. Each of these regions is also associated with high human activity,
223
although we did identify suitable climatic areas with low human activity (e.g., central South
224
America, Fig. 2). Yet, given that many suitable regions were identified by the ensemble model
225
that had both high climatic suitability and high human activity, it is likely that human activity
226
could facilitate future invasions of V. mandarinia. The model predicts that much of the interior of
227
North America is unsuitable habitat, likely due to inhospitable temperatures and low
228
precipitation. This includes the eastern portions of British Columbia and the Pacific Northwest
229
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
12
states, and the Central Valley of California, all of which have extensive agricultural production
230
(e.g. tree fruit and tree nuts) that relies almost exclusively on A. mellifera pollination.
231
Our simulations of V. mandarinia dispersal in western North America showed high potential
232
for spread within western North America (Fig. 4). When considering short-distance dispersal,
233
mediated by hornets flying an average of 10 km/yr and a maximum of 33 km/yr, populations of V.
234
mandarinia could rapidly spread along the western coast of North America, reaching Oregon in
235
20 yr. Northward expansion into Canada would likely be limited to the southern coast of British
236
Columbia (Fig. 4). When we accounted for long-distance human-mediated dispersal, the
237
expansion of V. mandarinia extended dramatically toward the north along coastal areas of British
238
Columbia, and showed a faster rate of expansion into southern Washington State and into Oregon,
239
USA (Fig. 4). This suggests dispersal throughout the western USA could occur within 20 or less
240
yr even without human-mediated transport or new introduction events.
241
Ecological impacts are difficult to predict for vespids (Beggs et al. 2011). While many
242
transplanted Vespidae appear to have only minor impacts, others are known to displace
243
congeners through multiple, idiosyncratic mechanisms (Beggs et al 2011). There are no other
244
Vespa, native or introduced, in the region of North America where V. mandarinia has been
245
detected, and no native Vespa where suitable habitat is predicted by this model. However, Asian
246
giant hornets are known to prey on social Hymenoptera other than bees (Matsuura 1984), and
247
thus could affect populations of several vespid genera in the Pacific Northwest. Ve s p a
248
mandarinia also preys upon many other insects, with chafer beetles comprising a large part of its
249
diet in parts of Japan (Matsuura 1984). It is unknown how it might impact native insects if it
250
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
13
becomes established, but the habitat suitability predicted here indicates that negative effects
251
could be distributed over a fairly expansive area.
252
We also anticipate considerable impacts on beekeepers. Established populations of V.
253
mandarinia would likely prey on readily-available hives late in the season, weakening any that
254
aren’t killed outright. In Europe, the congener V. velutina has reportedly caused losses ranging
255
from 18 to 80% of domestic hives, depending on the region (Laurino et al. 2020). The results
256
presented here suggest that large expanses of the Pacific Coast in North America could become
257
challenging for beekeeping operations, especially during the late summer and fall when attacks
258
are greatest. Unchecked, this species of hornet could cause major disruption in the western US
259
and Canada, possibly forcing beekeepers to invest in extensive hornet management or relocate
260
parts of their operations to areas of unsuitable V. mandarinia habitat.
261
262
Acknowledgement
263
We thank D. Zurell, R. Engler, and E. Ugene for help developing ecological models. The
264
work was funded by USDA Hatch Project 1014754.
265
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
14
References
266
Araújo MB, New M (2007) Ensemble forecasting of species distributions. Trends in Ecol. Evol.
267
22, 42–47.
268
Archer ME (1995) Taxonomy, distribution and nesting biology of the Vespa mandarinia group
269
(HYM. Vespinae). Entomol. Mon. Mag. 131, 47–53.
270
Beggs JR et al. (2011) Ecological effects and management of invasive alien Vespidae. Biocontrol
271
56: 505-526.
272
Bertolino S, Lioy S, Laurino D, Manino A, Porporato M (2016) Spread of the invasive yellow-
273
legged hornet Vespa velutina (Hymenoptera: Vespidae) in Italy. Appl. Entomol. Zool., 51,
274
589–597.
275
Choi MB, Kim JK, Loo JW (2012) Increase trend of social Hymenoptera (wasps and honeybees)
276
in urban areas, inferred from moving-out case by 119 rescue services in Seoul of South
277
Korea. Entomol. Res. 42, 308–319.
278
Engler R, Hordijk W, Guisan A (2012) The MIGCLIM R package – seamless integration of
279
dispersal constraints into projections of species distribution models. Ecography 35,
280
872–878.
281
Feng X, Park DS, Liang Y, Pandey R, Pape
ş
M (2019) Collinearity in ecological niche modeling:
282
Confusions and challenges. Ecol. Evol. 9, 10365–10376.
283
Fick SE, Hijmans RJ (2017) WorldClim 2: new 1km spatial resolution climate surfaces for global
284
land areas. Int. J. Clim. 37, 4302-4315.
285
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
15
Hosmer DW, Jovanovic B, Lemeshow S (1989) Best subsets logistic regression. Biometrics 45,
286
1265–1270.
287
Käfer H, Kovac H, Stabentheiner A (2012) Resting metabolism and critical thermal maxima of
288
vespine wasps (Ves pu la sp.) J. Insect Phys. 58, 679–689.
289
Laurino D, Lioy S, Carisio L, Manino A, Porporato M (2020) Vespa velutina: An alien driver of
290
honey bee colony loss. Diversity 12, 5.
291
Lee JX (2010) Notes on Vespa analis and Vespa mandarinia (Hymenoptera, Vespidae) in Hong
292
Kong, and a key to all Vesp a species known from the SAR. Hong Kong Entomol. Bull. 2,
293
31–36.
294
Leroy B et al. (2018) Without quality presence–absence data, discrimination metrics such as TSS
295
can be misleading measures of model performance. J. Biogeog. 45, 1994–2002.
296
Liebhold AM, Tobin PC (2008) Population ecology of insect invasions and their management.
297
Annu. Rev. Entomol. 53, 387–408.
298
Matsuura M (1984) Comparative biology of the five Japanese species of the genus Ve spa
299
(Hymenoptera, Vespidae). Bull Fac. Agric. Mie Univ. 69, 1–131.
300
Matsuura M, Sakagami SF (1973) A bionomic sketch of the giant hornet, Vespa mandarinia, a
301
serious pest for Japanese apiculture. J. Fac. Sci Hokkaido Univ. (Zoology) 19, 125–162.
302
McClenaghan B et al. (2019) Behavioral responses of honey bees, Apis cerana and Apis
303
mellifera, to Vespa mandarinia marking and alarm pheromones. J. Apic. Res. 58, 141–148.
304
Owens HL et al. (2013) Constraints on interpretation of ecological niche models by limited
305
environmental ranges on calibration areas. Ecol. Model. 263, 10–18.
306
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
16
Peterson AT et al. (2011) Ecological Niches and Geographic Distributions: A Modeling
307
Perspective. Princeton University Press, Princeton, USA.
308
Phillips SJ, Dudík M (2008) Modeling of species distributions with Maxent: new extensions and
309
a comprehensive evaluation. Ecography 31, 161–175.
310
Qiao H, Peterson AT, Campbell LP, Soberon J, Ji L, Escobar LE (2016) NicheA: creating virtual
311
species and ecological niches in multivariate environmental scenarios. Ecography 39,
312
805-813.
313
R Core Team (2020) R: a language and environment for statistical computing (v. 4.0.0). R
314
Foundation for Statistical Computing, Vienna, Austria.
315
Robinet C, Suppo C, Darrouzet E (2017) Rapid spread of the invasive yellow-legged hornet in
316
France: The role of human-mediated dispersal and the effects of control measures. J. Appl.
317
Ecol. 54, 205–215.
318
Scott A, Ram K, Hart T, Chamberlain MS (2017) spocc: interface to species occurrence data
319
sources, R package version 0.4.0. Available: http://CRAN.R-project.org/package=spocc.
320
Sugahara M, Nishimura Y, Sakamoto F (2012) Differences in heat sensitivity between Japanese
321
honeybees and hornets under high carbon dioxide and humidity conditions inside bee balls.
322
Zool. Sci. 29, 30–36.
323
Thuiller W, Lafourcade B, Engler R, Araújo MB (2009) BIOMOD–a platform for ensemble
324
forecasting of species distributions. Ecography 32, 369–373.
325
Tsoar, A., Allouche, O., Steinitz, O., Rotem, D. and Kadmon, R., 2007. A comparative evaluation
326
of presence
only methods for modelling species distribution. Div. Dist. 13, 397-405.
327
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
17
USDA (2019) New pest response guidelines for Asian giant hornet (Vespa mandarinia). United
328
States Department of Agriculture, Animal and Plant Health Inspection Service, Plant
329
Protection and Quarantine, Riverdale Park, MD, USA.
330
Warren DL, Glor RE, Turelli M (2010) ENMTools: a toolbox for comparative studies of
331
environmental niche models. Ecography 33: 607–611.
332
Warren DL, Seifert SN (2011) Ecological niche modeling in Maxent: the importance of model
333
complexity and the performance of model selection criteria. Ecol. Appl. 21, 335–342.
334
Yanagawa Y, Morita K, Sugiura T, Okada Y (2007) Cutaneous hemorrhage or necrosis findings
335
after Vespa mandarinia (wasp) stings may predict the occurrence of multiple organ injury:
336
A case report and review of literature. Clin. Toxicol. 45, 803–807.
337
Zhu GP, Peterson AT (2017) Do consensus models outperform individual models? Transferability
338
evaluations of diverse modeling approaches for an invasive moth. Biol. Invasions 19,
339
2519–2532.
340
Zurell D, Elith J, Schroeder B (2012) Predicting to new environments: tools for visualizing
341
model behavior and impacts on mapped distributions. Div. Dist. 18, 628–634.
342
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
18
Figure 1. Present distribution of Asian giant hornet in (A) native and (B) introduced regions. In
343
(A) points denote trimmed records used to fit models.
344
345
346
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
19
Figure 2. Ensemble forecast of potential invasion of Vespa mandarinia. Increasing intensities of
347
yellow represent increasing climate suitability, and increasing blue represent increasing
348
establishment potential due to human activity, where increasing red mean increasing potential
349
invasion due to high climate suitability and human activity. Scores of bivariate maps are divided
350
into 6 equal quantiles in the data ranges of climate suitability and human footprint respectively.
351
352
353
354
355
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
20
Figure 3. Realized niche occupied by native and introduced populations shown as minimum
356
ellipsoid volumes. The pink volume represents the native niche, the blue volume represents the
357
introduced niche, and points denote environmental conditions across the globe. The three PCA
358
axes were estimated in NicheA and captured 90% of the variation in the 8 bioclimatic variables.
359
360
361
362
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
21
Figure 4. Combined suitability (A) and estimated expansion (B-D) of Vespa mandarinia over 20
363
yr in western North America under three dispersal scenarios: (B) short dispersal distance only
364
(SSD), (C) long dispersal distance only (LDD) and (D) combined (LDD & SDD) scenarios. Each
365
color represents two dispersal steps (total 20 dispersal steps) in dispersal simulations.
366
367
.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under a The copyright holder for this preprint (whichthis version posted June 1, 2020. . https://doi.org/10.1101/2020.05.25.115311doi: bioRxiv preprint
