Introduction

Modeling species distributions is a powerful tool that has many applications to conservation, including forecasting potential distributions of invasive species and predicting future distributions with climate change (Elith et al. 2011). By utilizing presence-only samples with their associated geographic coordinate data and environmental data, Maximum Entropy Modeling software (or MaxEnt) can predict species occurrences, fitting a model to these data by minimizing the relative entropies of the probability densities of the sites from the samples and of the landscape or potential sites of occurrence.

In this study, we analyzed samples of the genus Dissochaetus, which belongs to the Leiodidae family of beetles, colloquially known as the round fungus beetles (Evans 2014). Since these beetles are fully winged, capable of flight, and feed on decaying organic matter, they can be caught in either dung- or carrion-baited pitfall traps or flight interception traps (Evans 2014). A review of the literature reveals that these beetles are relatively understudied, especially with regards to habitat preference, as compared to other genera of the same family, such as the small car­rion beetle. (Kočárek 2002, Costas 2014). Likewise, more detailed fossil evidence is noted for Catopocerinae beetles of the Leiodidae family, such as from the Pleistocene; their flightless and eyeless morphology makes it more likely that one can assume they were widely distributed during west­ern Pangaea. Therefore, their wide distribution during western Pangea could be responsible for their now disjunct distribution across North and South America (Peck & Cook 2011). It could also be true that the disjunct distribution of Dissochaetus across North and South America is also due to a broad historic distribution when the continents were connected. How­ever, because they are capable of flight and therefore better dispersalists, it cannot be concluded that this is the case. With modeling analysis, one can predict from species’ modern locations whether they could survive in other regions, but only a study incorporating phylogenetics and fossil evidence could truly address this historical question.

Hypothesis 1: Neotropical vs. Nearctic Realms

In Janzen’s (1967) article on mountain passes and temperature variability in the tropics as compared to the temperate region, he showed that, since temperature varies more in the temperate than in the tropics, species in the tropics would have a narrower niche breadth relative to their counterparts in the temperate region. Therefore, the changes in tem­perature with elevation would pose a significantly greater barrier to tropi­cal species than temperate because they would be less likely to tolerate this thermal shift with elevation. This idea had been supported by further findings. For example, North American mammalian immigrants survived and outcompeted more South American mammals in South America than South American mammalian immigrants did in North America after the North American Interchange (Marshall et al. 1982). Beetles that are more geographically and elevationally widespread are known to have a higher thermal tolerance, so from this fact and Janzen’s study, it can be inferred that Dissochaetus of the Nearctic would have a higher thermal tolerance and therefore, broader distribution than Dissochaetus of the Neotropical realm (Garcia-Robledo et al. 2016). Therefore, I would also predict Dis­sochaetus of the Nearctic realm to be more prevalent in the Neotropical realm, and Dissochaetus of the Neotropical realm to be less prevalent in the Nearctic realm. I also would predict that elevation will be a more significant factor in the model of the distribution of Dissochaetus from the Neotropical data as compared to Nearctic data.

Hypothesis 2: Bolivia vs. Neotropical Realm

Most of the samples collected in Bolivia were extremely close in geographic location (Fig. 1). This area of Bolivia (in Cochabamba) is known for its “eternal spring” as the temperature is warm year-round (Ca­valleri 2013). As Janzen and others have also demonstrated, this stability in temperature is associated with narrower niches of species (Janzen 1967). Therefore, I expect the distribution predicted from the Bolivia data to be restricted to only thermally stable regions and less widespread than the Neotropical model’s realm as whole. Thus, temperature seasonality would be a larger contributing factor for the Bolivia model than in the Neotropical model. I would also predict landcover to be a more significant contributing factor because of the restriction in sampling site as compared to the Neotropical realm. These predictions could then suggest that there are species endemic to Bolivia or that there is a sampling bias in the model.

Hypothesis 3: Ground Traps vs. Flight Interception Traps

Two types of traps, flight interception traps (FIT) and ground traps, capture beetles in different types of activity – flight and ground movement. Flight interception traps consist of a pane of glass covered in a fine mesh that is not visible to beetles, causing them to hit the glass and be collected in a container with a preserving solution (Yi et al. 2012). Malaise traps were included in this study because they can also catch beetles in flight. Malaise traps are large tents with openings at the front and back and a fabric barrier in between, also with a container with preservative to collect the beetles. Meanwhile, the ground traps, which are most likely where the baited “carrion traps” and “dung traps” were placed, can be used for collecting surface-dwelling beetles, including flightless ones (Yi et al. 2012). Trapping method is known to affect the assemblages of species trapped (Holland 2006). For one thing, since some species of beetles live inside decaying tree trunks for generations, both passive methods of ground traps and flight interception traps (FIT) fail to capture these species (Majka & Klimaszewski 2009). Thus, both these methods are considered to be activity-biased (Yi et al. 2012). One review of studies conducted in Oulanka National Park found that FITs caught 44-48% of species, while a combination of FITs and pitfall traps caught about 60%. Only manual sweeping was able to recover enough species to reach 91.4% of those known in the area (Majka & Klimaszews­ki 2009). These authors were also further concerned with this discrepancy by the possibility that FITs may sample more generalist species and fewer specialists that fly less (Majka & Klimaszewski 2009). Therefore, I would predict that modeling Dissochaetus distributions using FIT traps will produce a broader distribution than using ground traps. Since the higher diversity of species in the tropics is often explained by increasing spe­cialization, I would also expect the ground traps map to be more similar to the Neotropical realm map than the Nearctic map because it should appear more restricted than the FIT model (Brown 2014). This hypothesis is confounded by the choice of trap of the experimenter, so it assumes that the experimenter chose the trap that would collect the most beetle species in the area.

Methods

Prof. Menke’s Biogeography class first digitized the labels of 103 Dissochaetus beetle samples from the Field Museum’s collection. Then, using the location information provided, these samples were assigned latitude and longitude coordinates if they were not already pro­vided on the label using Google Earth. These latitude and longitude coor­dinates were then verified in ArcMap using ArcGIS by importing the data from a comma delimited Excel file, mapping the coordinates and checking that they were in the right country. In this way, we were able to prevent the error inherent in identifying coordinates in Google Earth by correcting any samples that were incorrectly assigned coordinates in the ocean, which would have subsequently influenced our distribution models.

Then, I used Maximum Entropy Modeling (or MaxEnt) to model the possible distribution of Dissochaetus according to the location data for the samples subsetted by realm (Nearctic and Neotropical), country (Bolivia), and trap type (flight interception traps [FIT] including malaise traps, and ground traps [most likely most were pitfall traps, although one was specified as a “cup”] baited with carrion or dung). The models were set up according to the tutorial provided by Young et al. (2011) and Prof. Menke’s instructions. 10 replicates were performed for each model. Prof. Menke also provided the environmental layers used in the model for temperature and rainfall patterns, as well as elevation and landcover val­ues (see Table 1 for the complete list of variables used excluding Bio_5, Bio_6, and Bio_7, which could not be run but are included for reference to understand other variables), and the results were also interpreted ac­ cording to his guide and the explanation of modeling presence-only data provided by Elith et al. 2011.

Results

The coordinates that we used for the MaxEnt modeling were all verified to country in ArcGIS, so none of the climate, elevation, or land cover variables could have been so erroneous as to affect the modeling significantly (Fig. 1). The models were also confirmed to fit better than random chance before the results were evaluated (Fig. S1).

Hypothesis 1: Neotropical vs. Nearctic Realms

The probability map for Dissochaetus presence produced by the model from the Nearctic data (Fig. 2) does appear much more widely distributed than the probability map for the Neotropical data (Fig. 3), as predicted. The Nearctic model particularly predicts a much higher prob­ability and wider distribution for Dissochaetus in the Palearctic than the Neotropical model, as well as for the Australian and Afrotropical realms (Figs. 2, 3). The Nearctic data also predicted the presence of Dissochae­tus in the Neotropics (Fig. 2), as predicted and known for some locations from our data. However, the Neotropical data did not predict the presence of Dissochaetus in the northern and southwestern United States, contrary to the fact that we did have samples from those regions, including ones from Colorado, Indiana, Texas, and Arizona (Fig. 3).

For the Nearctic model, mean diurnal range had the high­est permutation importance in MaxEnt, with 70.1% (Table 2). The next highest permutation importance was the coefficient of variation of precipitation, or precipitation seasonality, with 22.9% (Table 2). While a few factors had smaller contributions (precipitation of warmest quarter, precipitation of driest quarter, and isothermality), the rest of the factors had a zero value for permutation importance in the final model (Table 2). Conversely, the Neotropical model did not have one variable with a huge importance but rather several that were above 20%: temperature season­ality (20.2%), isothermality (23.3%), and precipitation of the driest month (25.9%). Most of the other factors had smaller permutation importances (Table 3). Elevation had a 0.3% permutation importance for the Neotropi­cal model, while it had zero importance to the final model for the Nearctic, a much smaller difference in factor contributions than between other factors of the models (Tables 2, 3).

Hypothesis 2: Bolivia vs. Neotropical Realm

The probability map of Dissochaetus distribution using only the place data from Bolivia (Fig. 4) is relatively similar in overall spread to the map for the entire Neotropical realm (Fig. 3). The most noticeable difference is the difference in African distribution: based on the complete Neotropical data, Dissochaetus is more likely to be found on the east coast of Africa than based on the model with only the Bolivia data, which predicts a higher likelihood in central Africa (Figs. 3, 4). While the map for Bolivia appears slightly less widespread (especially with respect to Africa, southern Asia, and parts of central South America), it has similar likeli­hoods for Dissochaetus presence and there are no other major regions of Neotropical distribution that are absent from the map produced from the Bolivia data alone (Figs. 3, 4).

The Bolivia and Neotropical models also share temperature seasonality as a major contributing factor to each model (20.2% for Neotropical, 26.6% for Bolivia, Tables 3, 4) but diverge from there. As previously stated, isothermality and precipitation of the driest month were important contributing factors in the Neotropical model (Table 3) but they both had zero permutation importance values for the Bolivia model (Table 4). Instead, mean diurnal range and precipitation of the warmest quarter were important contributing factors to the Bolivia model (Table 4).

Hypothesis 3: Ground Traps vs. Flight Interception Traps

The probability maps for the distribution of Dissochaetus using the location data with FITs (Fig. 5) and ground traps (Fig. 6) appear broadly and remarkably similar. There are some small differences; the ground trap model predicts a broader distribution in Africa, and a more likely presence in Madagascar. Nonetheless, the ground trap model is not less broadly distributed than the FIT model, as was predicted; in fact, it is more broadly distributed in some areas.

While the probability maps appear relatively similar, there were some distinct differences in the permutation importance variables for the two models. Both had the precipitation of the driest month, isothermality, and precipitation of the warmest quarter as similarly important contrib­uting variables, but the permutation importance for landcover was zero for the ground trap model, while it was 28.1% in the FIT model (Tables 5, 6). Further, mean diurnal range was also a more important contributor in the FIT model than the ground trap model, although less drastically so (Tables 5, 6).

Discussion

Hypothesis 1: Neotropical vs. Nearctic Realms

The comparison between the Neotropical and Nearctic models’ probability distribution maps do support the prediction that the Nearctic species would be more prevalent throughout the globe and more tolerant of the Neotropics than the Neotropical species would be of the Nearc­tic, as the Neotropical model did not predict Dissochaetus occurrence where they are actually known to occur in the United States (Fig. 3). This suggests that Nearctic species of Dissochaetus could likely colonize the Neotropical region, as is shown in the model, but Neotropical species are less likely to be able to colonize the Nearctic region. Further, it suggests that if Dissochaetus is found in regions predicted by the Nearctic but not the Neotropical model, it may be more likely that it dispersed from the Nearctic and shares a more recent common ancestor with a Nearctic species.

These results also further illustrate another biogeographic principle: Rapoport’s rule, which states that species range increases with increasing latitude, which he derived from a study of the distributions of species and subspecies of mammals of North and Central America (Stevens 1989). This idea builds upon the idea that species with smaller niches encounter “higher” barriers and are also less successful at coloniz­ing new environments that differ in temperature and precipitation (Janzen 1967, Marshall et al. 1982). By sampling just six locations in the Nearctic, a much wider distribution was predicted than from many more locations sampled throughout the Neotropical realm because of the broader vari­ability in thermal and precipitation conditions experienced within these areas.

However, elevation was not a much more significantly important factor in permutation importance as predicted from Janzen’s findings (0.3% for Neotropical as compared to 0% for the Nearctic model), per­haps because the samples in the Neotropical realm were collected from a variety of elevations, such that the difference in elevation was not signifi­cant between the two groups (Janzen 1967). However, the fact that mean diurnal range, the particular difference that Janzen was also interested in when looking at thermal overlap, had the largest permutation importance in the Nearctic model but not the Neotropical model supports the idea that the distribution of Nearctic-adapted species could be much broader than Neotropical species due to adaptation to this temperature range (Janzen 1967).

Hypothesis 2: Bolivia vs. Neotropical Realm

The hypothesis that the restricted sampling strategy in Bolivia would result in a much less widely spread Dissochaetus distribution as compared to that predicted from the Neotropical realm as a whole was not supported by the resulting maps, which were remarkably similar in spread (Figs. 3, 4). Further, landcover was not a much more significant contributing factor to the model, as was also predicted, and the major contributing factors were not significantly different, as isothermality is based on the mean diurnal range, and both have a precipitation-related contributing factor during different temporal periods (driest month for the Neotropical realm, and warmest quarter for Bolivia). These results may be explained by a re-evaluation of the original locality data, which reveal that beetles were collected along an elevational gradient spanning 300 to 3250 m ASL. With this elevation difference, there was a noted difference in habitat type: from rainforest to elfin forest to cloud forest to paramo (tundra-like) shrub. In this way, a variety of habitats, rainfall, and tempera­tures were represented in the Bolivia sample such that it could mostly approximate the predictions from the whole Neotropical realm. While at first glance it appeared to be a locally restricted sample, it was in fact quite diverse in environment-type via the elevational gradient. Therefore, this finding further illustrates the use of elevational gradients as a proxy for latitudinal gradients and other geographical distances because of the remarkable variety of habitats that can be sampled on mountains (Halbrit­ter et al. 2013).

As Loiselle et al. (2008) emphasize that referring to expert opinion of geographic distributions is important for verifying MaxEnt mod­els, reference to one recently published species key that reviewed three species of Dissochaetus all found in Bolivia (similar in both geographic coordinates and reported habitat type of nine of the samples in this study) also reported that two of the three were known to be widely distribut­ed, while the third distribution is not well known, further supporting the similarity between these models (Costas 2014). Therefore, it is then not surprising that Bolivia could actually be somewhat representative of the Neotropical realm as a whole.

Hypothesis 3: Ground Traps vs. Flight Interception Traps

The distributions of Dissochaetus predicted by trap type did not appear to be very different, ultimately failing to support the prediction that the ground traps would be used more in the Neotropical region because of the greater number of specialists in the region. Thus, the ground trap model did not produce a more restricted distribution as predicted (Figs. 5, 6). The difference in FIT and ground traps previously observed in a review of studies could be due to differences in flight activity between different Dissochaetus species. However, flight behavior does not vary significantly between the threshold physiological thermal extremes for some insects (Majka & Klimaszewski 2009, Taylor 1963). Further, all Dissochaetus species are winged and capable of flight (Evans 2014), so while the previously discussed study found a large discrepancy between the two trapping methods in Oulanka National Park, it is likely this was due to the presence of flightless beetle species (Majka & Klimaszewski 2009). It is possible that a more specific study of Dissochaetus would not find a significant difference between the two trapping types. It can also be seen in the data that both types of traps were used in many of the same countries, thus resolving the previously-referenced issue of sampling bias in species, such that the model did not produce a difference in predicted distribution.

Further, there are other possible reasons why an experi­menter may choose to use a flight interception trap instead of a baited ground trap besides maximizing captured species richness, such as the possibility that carrion-feeding vertebrates may try to consume the bait if not secured. As a result, these vertebrates’ presence could confound the results, although not significantly enough to really differentiate the two models (Shubeck 1976). One of these possible reasons is actually sug­gested in the model. The model suggests that the landcover was a much more significant contributing variable to the final model for the FIT data than for the ground trap data (Tables 5, 6), such that the habitat type may actually influence experimenters’ choice to use FITs and could ultimately represent different species of Dissochaetus that may fly more frequently in different habitat types. This could explain the differential success in the types of traps in different landcover types. This difference and the difference in mean diurnal range in permutation importance between the two models, however, could also be an artifact of the small sample size or nonrandom sampling as described by Loiselle et al. (2008), as they have the potential to bias results.

In conclusion, MaxEnt is a powerful tool for distribution mod­eling that can reveal differences and unexpected similarities of predicted distributions between models created by subsampling Dissochaetus data by biogeographical realm, country, and trapping method. While only the genus of these samples was known and different species could only be inferred from the predicted distribution probabilities, these models were able to support the difference between Nearctic and Neotropical species’ colonization ability and the representative power of sampling an eleva­tional gradient of a mountain to approximate the habitats and climate conditions experienced across a larger region. Further, Dissochaetus species sampling may not be significantly affected by trapping method because both ground traps and flight interception traps produced similar models of distribution, but future studies that actually identify and quantify the species richness collected by each trap type could better address this question.

Acknowledgements

The author would like to thank Hailey Broeker for moral support and technical assistance with the ArcGIS and MaxEnt software.

References

Brown, J. H. 2014. Why are there so many species in the tropics?. Journal of Biogeography, 41(1), 8-22.

Cavalleri, G. 2013. Cochabamba: Bolivia’s city of eternal spring. Retrieved from: https://southamericabasics.com/2013/02/25/cochabamba-bo­livias-city-of-eternal-spring/

Costas, J. M. S. 2014. Dissochaetus truncatus n. sp. Keys to species in the group of D.

Brunneicollis Salgado, 2010. Information on other species in the genus Disso­chaetus (Coleoptera, Leiodidae, Cholevinae, Anemadini). Boletín de la SEA, (54), 121-125.

Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., & Yates, C. J. 2011. A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17(1), 43-57.

Evans, A. V. 2014. Beetles of Eastern North America. Princeton University Press.

García-Robledo, C., Kuprewicz, E. K., Staines, C. L., Erwin, T. L., & Kress, W. J. (2016). Limited tolerance by insects to high temperatures across tropical elevational gradients and the implications of global warming for extinction. Proceedings of the National Academy of Sciences, 113(3), 680-685.

Halbritter, A. H., Alexander, J. M., Edwards, P. J., & Billeter, R. 2013. How comparable are species distributions along elevational and latitudi­nal climate gradients?. Global Ecology and Biogeography, 22(11), 1228-1237.

Holland, J. D. 2006. Cerambycidae larval host condition predicts trap efficien­cy. Environmental Entomology, 35(6), 1647-1653.

Janzen, D. H. 1967. Why mountain passes are higher in the tropics. The Amer­ican Naturalist, 101(919), 233-249.

Kočárek, P. 2002. Small carrion beetles (Coleoptera: Leiodidae: Cholevinae) in Central European lowland ecosystem: seasonality and habitat preference. Acta Soc. Zool. Bohem, 66, 37-45.

Loiselle, B. A., Jørgensen, P. M., Consiglio, T., Jiménez, I., Blake, J. G., Lohmann, L. G., & Montiel, O. M. 2008. Predicting species distri­butions from herbarium collections: does climate bias in collection sampling influence model outcomes?. Journal of Biogeography, 35(1), 105-116.

Marshall, L. G., Webb, S. D., Sepkoski, J. J., & Raup, D. M. 1982. Mammalian evolution and the great American interchange. Science, 215(4538), 1351-1357.

Majka, C. G, & Klimaszewski, J. (Eds.). 2009. Biodiversity, Biosystematics, and Ecology of Canadian Coleoptera. Pensoft: Sofia-Moscow.

Peck, S. B., & Cook, J. 2011. Systematics, distributions and bionomics of the Catopocerini (eyeless soil fungivore beetles) of North America (Co­leoptera: Leiodidae: Catopocerinae). Zootaxa, 3077, 1-118.

Shubeck, P. P. 1976. An alternative to pitfall traps in carrion beetle studies (Coleoptera). Entomol. News, 87(5-6), 176-178.

Stevens, G. C. 1989. The latitudinal gradient in geographical range: how so many species coexist in the tropics. The American Naturalist, 133(2), 240-256.

Taylor, L. R. 1963. Analysis of the effect of temperature on insects in flight. The Journal of Animal Ecology, 99-117.

Yi, Z., Jinchao, F., Dayuan, X., Weiguo, S., & Axmacher, J. C. (2012). A comparison of terrestrial arthropod sampling methods. Journal of Resources and Ecology, 3(2), 174-182.

Young, N., Carter, L., & Evangelista, P. 2011. A MaxEnt model v3.3.3e tutorial (ArcGIS v10), tutorial, Colorado State University.

 

Figures

 

Figure 1: Map created in ArcGIS using the WGS84 coordinate system. The legend labels the samples by the country that was associated with the sample, not the country that its coordinates locate it in on the map. All samples appear to have been assigned coordinates consistent with their associated countries.

 

Figure 2: Picture of the MaxEnt model of Dissochaetus distribution from Ne­arctic data, as colored by probability of occurrence (see color bar: violet=0% to red=100%).

 

Figure 3: Picture of the MaxEnt model of Dissochaetus distribution from Neo­tropical data, as colored by probability of occurrence (see color bar: violet=0% to red=100%).

 

Figure 4: Picture of MaxEnt model of Dissochaetus distribution from Bolivia data, as colored by probability of occurrence (see color bar: violet=0% to red=100%).

 

Figure 5: Picture of MaxEnt model of Dissochaetus distribution from flight intercep­tion trap data, as colored by probability of occurrence (see color bar: violet=0% to red=100%).

 

Figure 6: Picture of MaxEnt model of Dissochaetus distribution from ground trap data, as colored by probability of occurrence (see color bar: violet=0% to red=100%).

 

Supplementary figure description

 

Figure S1: Receiver operating curves for test data (blue), training data (red) and the random model (gray) for the Nearctic model (a), Neotropical model (b), Bolivia mod­el (c), FIT model (d), and ground trap model (e). All models have test lines above the gray line, indicating that the predicted models fit better than random models.

 

Tables

Bioclim climate data variable key
Bio_1	Annual Mean Temperature
Bio_2	Mean Diurnal Range (Mean of monthly (max temp – min temp))
Bio_3	Isothermality (Bio_2/Bio_7)(*100)
Bio_4	Temperature Seasonality (stan­dard deviation*100)
Bio_9	Mean Temperature of Driest Quarter
Bio_10	Mean Temperature of Warmest Quarter
Bio_11	Mean Temperature of Coldest Quarter
Bio_12	Annual Precipitation
Bio_13	Precipitation of Wettest Month
Bio_14	Precipitation of Driest Month
Bio_15	Precipitation Seasonality (Coeffi­cient of Variation)
Bio_16	Precipitation of Wettest Quarter
Bio_17	Precipitation of Driest Quarter
Bio_18	Precipitation of Warmest Quarter
Bio_19	Precipitation of Coldest Quarter
Elevation	Elevation
Global Land Cover (Vegetation – Categorical) = Landcover
0	Water
1	Evergreen Needleleaf Forest
2	Evergreen Broadleaf Forest
3	Deciduous Needleleaf Forest
4	Deciduous Broadleaf Forest
5	Mixed Forest
6	Woodland
7	Wooded Grassland
8	Closed Shrubland
9	Open Shrubland
10	Grassland
11	Cropland
12	Bare Ground
13	Urban and Built

Table 1: Climate (temperature and rainfall), elevation, and landcover variable names, and landcover numerical values for reference, that were used as environ­mental layers (excluding Bio_5, Bio_6, and Bio_7, which are included for reference) in the MaxEnt analysis.

 

Variable	Percent contribution	Permutation importance
bio_2	51	70.1
bio_15	40.7	22.9
bio_17	4	0.6
bio_18	3.2	4
bio_3	0.7	2.4
bio_14	0.4	0
lancover	0	0
bio_19	0	0
bio_4	0	0
bio_16	0	0
bio_9	0	0
elevation	0	0
bio_13	0	0
bio_12	0	0
bio_11	0	0
bio_10	0	0
bio_1	0	0

Table 2: Variable contribution to the path to the model (percent contribution) and the final model (permutation importance) for the Nearctic model.

 

Variable	Percent contribution	Permutation importance
bio_1	24.2	6.5
bio_4	24	20.2
bio_3	19.6	23.2
bio_14	9.9	25.9
landcover	5.9	1.6
bio_16	4.5	2.1
bio_19	3.3	1
bio_15	3	6.7
bio_11	2.4	3.6
elevation	1.9	0.3
bio_2	0.7	0.6
bio_12	0.2	0.3
bio_17	0.2	6.4
bio_9	0.1	1.3
bio_1	0	0
bio_10	0	0
bio_13	0	0

Table 3: Variable contribution to the path to the model (percent contribution) and the final model (permutation importance) for the Neotropical model.

 

Variable	Percent contribution	Permutation importance
bio_4	27.3	26.6
landcover	26.1	1.2
bio_18	22.2	20
bio_2	13.4	37.9
elevation	3.3	0.4
bio_15	2.7	8.4
bio_13	2.1	5.4
bio_16	1.7	0
bio_19	0.6	0
bio_3	0.5	0
bio_12	0.2	0
bio_9	0	0
bio_17	0	0
bio_14	0	0
bio_11	0	0
bio_10	0	0
bio_1	0	0

Table 4: Variable contribution to the path to the model (percent contribution) and the final model (permutation importance) for the Bolivia model.

 

Variable	Percent contribution	Permutation importance
bio_18	31.4	11
bio_3	22.3	9
landcover	16.3	28.1
bio_14	10.2	22.5
bio_4	7	1.7
bio_2	5.9	16.5
bio_19	3.3	3.6
bio_15	1.8	0.9
bio_12	0.7	3.1
elevation	0.4	0.2
bio_1	0.2	0.1
bio_10	0.1	0.2
bio_9	0.1	1.7
bio_17	0.1	1.3
bio_11	0	0
bio_13	0	0
bio_16	0	0

Table 5: Variable contribution to the path to the model (percent contribution) and the final model (permutation importance) for the FIT model.

 

Variable	Percent contribution	Permutation importance
bio_18	24.5	3.1
bio_3	21.2	19.5
bio_14	11.4	39.8
bio_4	10	18.8
bio_13	6.3	0.6
bio_11	6.3	2.4
bio_12	5.3	2.6
bio_19	4.5	5.4
bio_16	4.1	0.3
bio_15	3.2	5.4
elevation	1.7	0.6
bio_2	0.7	1.4
landcover	0.5	0
bio_17	0.3	0
bio_1	0.1	0.1
bio_9	0	0.1
bio_10	0	0

Table 6: Variable contribution to the path to the model (percent contribution) and the final model (permutation importance) for the ground trap model.

Related Links:

• 142-147_McDonald_PrimaryArticle (PDF)