To predict areas where Xylocopa appendiculata could establish in the USA, we used location data for specimens at the American Museum of Natural History (AMNH, Jerome Rozen and Lance Jones), The Snow Entomological Collection at the University of Kansas Biodiversity and Natural History Museum (SEMC, Jennifer Thomas), and verified distribution data from iNaturalist (2019), Discover Life (2019), and the Global Biodiversity Information Facility (GBIF 2019) for specimens ID validated by taxonomists. For the iNaturalist data, we used research grade detections, which are identifications that over 2/3 of the identifiers agree on (iNaturalist 2019). The combined dataset contained 325 X. appendiculata occurrences (duplicate records were removed). We also included in the analysis the type localities for X. appendiculata Smith, 1874 which is Ning-po-foo, China and X. appendiculata circumvolans (Smith, 1873) which is Hiogo, Japan (Appendix 1).

We used this dataset to predict where X. appendiculata could establish in the USA based on the associated Plant Hardiness Zones and Maxent niche modeling. We used two modeling approaches to increase the rigor of the analysis and better account for uncertainty regarding the bee’s potential distribution in the USA.

The Plant Hardiness Zones are calculated based on the average annual extreme minimum temperature for an area in 10 °F (5.6 °C) increments (ARS 2012). They can be used to predict where plant pests (e.g., insects) could establish based on the Plant Hardiness Zones that match with the pest’s native range (PERAL 2013). We overlaid the 325 X. appendiculata occurrence points with a Plant Hardiness Zone layer that was based on average climate data from 1988 to 2017 (Takeuchi et al. 2018). We then mapped the corresponding plant hardiness zones in the United States to predict its distribution.

We used maximum entropy niche modeling (Maxent 3.3.3k; Phillips et al. 2006, 2017) to estimate the potential distribution of X. appendiculata in the United States. We removed duplicate records (>1 presence point within a ~4×4 km grid cell) and reduced spatial autocorrelation using ‘spatial filtering’ (i.e. reducing density of occurrences) using SDMToolbox (Brown 2014; Kumar et al. 2016). The total occurrences from the native range (Japan, Korean peninsula, and China) and invaded range (California, USA) were reduced from 325 to 158 (Appendix 1). We then calculated a Gaussian Kernel Density layer of occurrence data in ArcMap using SDMToolbox, which we used to account for potential sampling bias in occurrence data. We obtained climatic data from CHELSA website (Climatologies at High resolution for the Earth’s Land Surface Areas; http://chelsa-climate.org/; Karger et al. 2017). We downloaded 19 bioclimatic variables data layers (~4×4 km spatial resolution) which represent average monthly temperature, precipitation, seasonal variables, and climatic extreme indices data from 1979–2013 (Hijmans et al. 2005; Appendix 2). We examined all 19 variables for cross-correlation (Pearson correlation coefficient, r) and highly correlated variables (|r| > 0.80) were excluded to reduce multicollinearity. The decision to exclude or include a variable was based on its potential biological relevance to X. appendiculata and its relative predictive power in the model (Appendix 2, 3).

Multiple models were fitted with different feature types and regularization multiplier values, and the best model with the optimal level of complexity was selected. Performance of the model was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC; Peterson et al. 2011). We used the 10-fold cross-validation procedure for evaluating model performance, and reported averaged test AUC values across the ten replicates. Unsuitable areas for X. appendiculata were defined using the 100 percent sensitivity threshold (Liu et al. 2013). Therefore, areas with > 0.107 probability of presence in Maxent modeling results were considered as suitable for X. appendiculata.

The analysis of potential U.S. areas for Xylocopa appendiculata establishment based on plant hardiness zones (Figure 1) predicted that X. appendiculata could establish in Plant Hardiness Zones 5 to 10. This area includes most of the contiguous United States, and parts of southern and coastal Alaska (Figure 1). Colder regions like North Dakota, most of Alaska and the Rocky Mountains, and warmer regions like southern Florida and Hawaii were predicted to be unsuitable. Xylocopa appendiculata’s predicted cold hardiness is due to it’s occurrence in colder parts of Japan.

The plant hardiness zone model likely represents a worst case scenario for X. appendiculata’s potential U.S. distribution due to the coarseness of the approach. This conclusion is supported by the fact that no western species of Xylocopa reach Canada, except for X. virginica which is present in the most southern parts of Ontario and Quebec.

Figure 1. 

Suitable U.S. areas for Xylocopa appendiculata establishment based on plant hardiness zones – This map shows suitable areas for X. appendiculata establishment based on its global distribution and the associated plant hardiness zones 5 to 10.

The second analysis, or Maxent model, predicted climatically suitable areas for X. appendiculata in the eastern United States, parts of California, southwestern Alaska, and parts of Hawaii (Figure 2). Our model indicated medium to low climatic suitability (probability between 0.40–0.10) in parts of California, and very low suitability in Hawaii (probability 0.10–2.0) (Figure 2). The Maxent model for X. appendiculata performed well with a test AUC value of 0.98 (an AUC value of 1.0 indicates superior performance, and 0.5 indicates performance no better than random; Peterson et al. 2011). Precipitation of warmest quarter (Bio18) and temperature seasonality (Bio4) were the two top predictors associated X. appendiculata’s distribution, with 45% and 28% contributions, respectively (Appendix 2).

The climatic response curves fitted by the Maxent model suggest that areas with average annual temperatures (Bio1) between 4 °C and 22 °C, and average annual precipitation (Bio12) between 200 mm and 3,800 mm are likely suitable for X. appendiculata (Appendix 3). The distinct contrast in the patterns in predicted climatic suitability for X. appendiculata in northeastern Texas, Arkansas, northern Louisiana, western Mississippi and Tennessee, is probably due to variation in average summer precipitation (June, July and August) in that part of the country (Appendix 4). Based on where the plant hardiness zone and Maxent models agree, the areas at greatest risk for X. appendiculata establishment include parts of southern Alaska and portions of eastern Virginia and the Carolinas (Figures 1, 2). Other at-risk areas include portions of western California and the eastern, north central, and central United States.

Figure 2. 

Predicted climatic suitability for Xylocopa appendiculata in the United States.

USA records

USA. 1♀; California: San Jose; Sept. 2012; on flowers of Salvia azurea Michx. ex Lam. (Lamiaceae).1 ♂; same locality data; May 2013; on flowers of Erysimum linifolium (Pers.) J. Gay (Brassicaceae). 2♂; same locality data; Apr. 2015; on flowers Erysimum linifolium (Brassicaceae). 1♀; same locality data; Mar.2017; on flowers of Prunus cerasifera Ehrh (Rosaceae). 1♀; same locality data; May 2019; on flowers Erysimum linifolium (Brassicaceae).

USA. 1♀; California: Castro Valley; 37°42'13"N, 122°03'35"W; Sept. 2017; on flowers of Grewia occidentalis L. (Malvaceae). 1♀; same locality data; Aug. 2018; on flowers of Passiflora sp. L. (Passifloraceae). These two records show the movement of the species further north (~32–45 kilometers) from where the species was first reported in northern California.

Bees of the subgenusAlloxylocopa Hurd and Moure (before the introduction of X. appendiculata, the subgenus was not known to be present in the Western hemisphere) can be distinguished from all other Xylocopa subgenera that are present in the USA and Canada by the following combination of characters: (female) pygidial plate without sub-apical spines (Fig. 3d, pp: pygidial plate), (both sexes) mesoscutellum with sub-horizontal dorsal surface abruptly and angulately separated from sub-vertical surface (Fig. 3e), postero-dorsal margin of mesoscutellum not surpassing posterior margin of metanotum and not projecting posteriorly beyond its posterior surface as a thin-edged flange., vertical fold of metasomal tergum 1 with foveate, hollow-like depression (Fig. 3f, arrow),

Specimens of X. appendiculata can be distinguished from native U.S. species of carpenter bees by the following combination of features: mandible with two teeth (Fig. 3g, arrow); clypeus and paraocular areas strongly punctate (Fig. 3b, square); occipital area, mesosoma and external side of fore tibiae with pubescence bright yellow, black on most of metasoma (Fig. 3a), except for few light brown setae close to pygidial plate on female (Fig. 3d, arrows); (Fig. 3c, arrow), Fig. 3a, e) and (Fig. 3a, arrow). The mesosoma of the native, North American species can be covered with black, grayish, brownish, or pale yellow setae and often have pale hairs elsewhere on the body and/or metallic highlights on the integument (Hurd and Moure 1963). The most similar of the native carpenter bees in the U.S. is Xylocopa virginica L. (the Eastern carpenter bee) which also has two mandibular teeth and a strongly punctuate clypeus, but differs from X. appendiculata because it lacks the extensive bright yellow hairs in the mesosoma and vertex. The males of X. appendiculata also lack the extensive bluish or greenish reflections of the body present in males of X. virginica.

Figure 3. 

Some of the diagnostic features of Xylocopa appendiculata Smith: a lateral habitus showing coloration of pubescence on mesosoma and metasoma (the yellow coloration may not be as fluorescent in field specimens and is mostly due to reflections from the imaging system) b face showing integument sculpturing of clypeus and paraocular area c coloration of pubescence around occipital area (arrow) d pygidial plate (pp) and coloration of pubescence close to it (arrows) e shape of mesoscutellum from lateral view f Shape of metasomal tergum 1 (T1) from antero-lateral view and the presence of a foveate, hollowed like depression (arrow) g dentation of mandibles (arrow showing the two teeth on mandible).

Little is known about the preferred plants visited and pollinated by X. appendiculata in its native range, but based on the observations of visits in the USA this species seems, at least potentially, polylectic in its floral preferences. In the new habitat in the Bay Area this species was seen visiting flowers of introduced plants such as Grewia occidentalis (originally from Africa) and Passiflora sp. (originally from tropical America). Both plants were in the senior author’s bee garden in Castro Valley, CA.