Among the five sites, 134 species in 23 genera were collected (data not corrected for sampling intensity). Table 1 provides a full list of species and their abundances at each site and Table 2 gives a list of observed richness, estimated richness, and Fisher's alpha for each site. The most speciose site was Vitobamba, followed by Atalaya, Los Amigos, and Culebrayoc; Wayquecha was the least speciose. A clear unimodal pattern is present, though it is skewed toward the lower elevations (Fig. 2). Species richness decreases faster at the lower elevations than higher elevations. This is evidenced by the fact that Los Amigos and Culebrayoc have nearly equal estimated species richness values, but Culebrayoc is 900 m up slope from Vitobamba while Los Amigos is only 500 m down slope. Both species richness estimations and Fisher's alpha positively correlated with temperature. The relationship was weak and not significant when considered for the whole elevational range for estimated species richness (r² = 0.552, p = 0.150) but the relationship had a higher correlation for Fisher's alpha (r² = 0.726, p = 0.067) (Figs 3–6). When the relationship was only considered for the three highest elevations it was relatively highly correlated, but still non-significant, for species richness (r² = 0.939, p = 0.159), and not significant for Fisher's alpha (r² = 0.736, p = 0.343) (Fig. 7). Species richness was positively, but not significantly, correlated with (log)elevation below the breakpoint (300–818 m a.s.l.; r² = 0.953, p = 0.139) and was negatively, significantly correlated with elevation above the breakpoint (818–2,886 m a.s.l.; r² = 0.999, p = 0.003); Fisher's alpha did not significantly correlate with (log)elevation above or below the breakpoint, and was only strongly correlated with (log)elevation above the breakpoint (300–818 m a.s.l.; r² = 0.357, p = 0.593; 818–2,886 m a.s.l.; r² = 0.901, p = 0.203).

Figure 2. 

The relationship between elevation and species richness. There is a clear low-elevation skewed unimodal pattern, which peaks around 800 m.

There were only 26 cosmopolitan species, which accounted for 21.0% of the total species richness and 34.2% of the total abundance (Table 3). Lowland species comprised 34.7% of total species richness and 23.2% of total abundance. Low land species had similar contributions to total species richness and total abundance at Los Amigos and Vitobamba. At Culebrayoc, montane species only accounted for 17.9% of the species richness but 71.9% of the total abundance, which is far greater than at either Vitobamba or Wayquecha (Table 3). This is primarily due to three species: Xylophanes docilis (Butler, 1875), X. resta Rothschild & Jordan, 1903, and Xylophanes sp. 3, whose combined abundances accounted for 964 of the 1,845 moths collected at Culebrayoc (Table 1). Endemic species contributed 21.0% to total species richness, but much less to total abundance (1.2%), and very little to the species richness or abundance at any one site (Table 3). Only four endemic species: Cocytius belzebuht (Boisduval, 1875), Lintneria aurigutta (Roghschild & Jordan, 1903), Nyceryx nictitans saturate Rothschild & Jordan, 1903, and Xylophanes sp. 2 were represented by more than three specimens (Table 1).

Figures 3–6. 

Regression analysis across the entire gradient for average temperature with (3) estimated species richness, (4) Fisher's alpha, and regression analysis at the three highest elevations for average temperature with (5) estimated species richness, and (6) Fisher's alpha. (3) r² = 0.5518, p = 0.150; (4) r² = 0.7264, p = 0.067; (5) r² = 0.9391, p = 0.159; (6) r² = 0.7364, p = 0.343.

Figure 7. 

Breakpoint regression analysis for (log)elevation with estimated species richness. Vitobamba (800 m), which was the break point, is included in both graphs. The characteristic mid-elevation unimodal richness pattern can be seen with the increase in species richness with elevation up to Vitobamba (r² = 0.953, p = 0.139), and the decrease in species richness with elevation above Vitobamba (r² = 0.999, p = 0.003).

On a species level, there appears to be a division between lowland species, those collected at Los Amigos, Atalaya, and Vitobamba, and montane species, those with populations centered at Culebrayoc and Wayquecha (Table 1). Even cosmopolitan species and species found at both lowland and montane sites can be divided into two groups: those with Los Amigos-Atalaya-Vitobamba centered populations or those with Vitobamba-Culebrayoc-Wayquecha centered populations. These species had individuals extending beyond their population centers, but only in small numbers. Protambulyx strigilis (Linnaeus, 1771), for example, had 73 individuals collected at lowland sites but only five individuals from Culebrayoc. Few species showed uniform distributions among high and low elevations.

Most genera were cosmopolitan, with species along the entire altitudinal gradient. Some genera, however, appeared to be more restricted in range, and most of these were lowland centered. The genera Callionima, Eumorpha, Hemeroplanes, Madoryx, Manduca, Nyceryx, Oriba, and Protambulyx all contain predominantly lowland species, with high elevation representatives being rare. Three genera, Hemeroplanes, Madoryx, and Oriba, appear to be lowland endemics, with all individuals coming from Los Amigos, Atalaya or Vitobamba. Only one genus, Euryglottis, was completely restricted to montane forests. Individuals of all three species, Euryglottis aper (Walker, 1856), E. dognini Rothschild, 1896, and E. guttiventris (Rothschild & Jordan, 1903), were only collected in Culebrayoc and Wayquecha, with the majority of individuals collected in Culebrayoc.

The contributions of the three sphingid subfamilies, Macroglossinae, Sphinginae, and Smerinthinae, to species richness and total abundance were similar at all five sites (Figs 8–9). Moreover, at each site the contribution of each subfamily to species richness was similar to its contribution to abundance. There were no sites in which a subfamily made up a large proportion of species richness but a small proportion of the abundance, or vice versa. For example, at Los Amigos Macroglossinae made up 69.2% of the species richness and 71.9% of the abundance.

Figures 8–9. 

Contribution of sphingid subfamilies to (8) species richness and (9) total abundance. Blue bars = Macroglossinae; orange bars = Sphinginae; grey bars = Smerinthinae. Contributions of each subfamily are similar among sites, and contributions of any particular subfamily to richness and abundance are similar within a site.

Hawkmoth diversity did not correlate well with temperature across the entire elevational range. However, Hawkins et al. (2003) and McCain (2007) noted that with water-energy dynamics the importance of energy depended on the latitude or altitude, and Beck and Chey (2008) stated that temperature could be expected a priori to only be correlated with richness at higher elevations. When the temperature-species richness relationship is only considered for the three highest elevations the predictive power of temperature increases dramatically, which is consistent with other studies. Temperature should, then, only be part of what drives species richness. Longino and Colwell (2011) found that although temperature was the best predictor of elevational species richness, no single environmental variable matched the unimodal, low-elevation skewed richness pattern they observed for litter ants in Costa Rica.

Species richness peaked at ~800 m. This is consistent with the findings of other studies on altitudinal insect diversity patterns (e.g., Samson et al. 1997 for ants, Beck and Chey 2008 for geometrids, Beck and Kitching 2009 for sphingids, Longino and Colwell 2011 for litter ants). Transitional plant communities at the sites of peak richness can potentially explain a unimodal pattern. The two sites with the highest species richness (Atalaya and Vitobabma) both contain multiple plant communities. The vegetation at Atalaya is characterized by a combination of moist lowland tropical forest and pluvial submontane tropical forest, while Vitobamba is characterized by a transition from lowland to montane forest. These transitional plant communities likely contain food plants for montane and lowland forest hawkmoths. Beck et al. (2002) and Beck and Chey (2008) reported positive correlations between moth and plant species richness, though Beck and Chey only found this at higher elevations; Schulze et al. (2001) found evidence that, for some butterflies, distribution was influenced by host plant distribution. Atalaya and Vitobamba may represent the lower boundary limit for the range of some montane plants and the upper boundary limit for lowland plant species, and therefore the range limit for the moths that depend on those plants. Overlap between lowland and montane faunas could explain the mid-elevational peak in species richness (Holloway et al. 1990 referenced in Beck et al. 2006b, but see Beck and Chey 2008). To test this specifically, plant species richness and sphingid host plant data are needed for these sites.

The question remains as to what factors drive species richness. As mentioned previously, hypotheses of indirect, trophic cascade effects (host plant diversity) (Beck and Chey 2008) and a combination of direct and indirect effects (McCain 2007) have been proposed as means by which water-energy dynamics limit species richness. Addressing the determinants of species' ranges may also prove insightful to understanding diversity patterns. Longino and Colwell (2011) raise the question as to why mobile species are not constantly expanding their range. They consider a hypothesis in which the upper limits of a species' range is constrained by its thermoregulatory adaptations while lower range limits are determined by competitive interactions (references within Longino and Colwell (2011) provide more information and support for this hypothesis). Gaston and Chown (1999) showed that the thermal minimum temperature of beetles decreased strongly with elevation while the thermal maximum only decreased slightly. This suggests that high elevation species are physiologically able to inhabit lower elevations, but some other factor, such as competition, is restricting their range, further supporting the hypothesis in Longino and Colwell (2011). The extent to which this can explain sphingid ranges depends on the intensity of interspecific competition. Kitching and Cadiou (2000) describe hawkmoth larvae in the Guanacaste Conservation Area, Costa Rica as being host specific, only feeding on a specific genus or several genera within a single family. Depending on the amount of host plant overlap, this would suggest interspecific competition to be either relatively strong or weak. If there is competition between lowland and montane communities, the larger number of lowland centered species with ranges extending into montane habitats would suggest that lowland sphingids are competitively dominant. Unfortunately, this data set is unable to answer these questions. Even with detailed data on plant and moth richness, larval host plants, rainfall, and temperature, the complex nature of these interactions will likely make it difficult to identify the mechanisms driving moth species richness.

Community composition changed between the five sites on a species but not on a subfamily level. Other researchers (Kemp et al. 1990 for grasshoppers, Boggs and Murphy 1997 for butterflies, Kumar et al. 2009 for paper wasps, Longino and Colwell 2011 for ants) have also observed similar species level community composition changes. Boggs and Murphy (1997) noted that weather conditions, such as temperature and precipitation, affect the availability of food resources and therefore the distribution of species. Kemp et al. (1990) found that habitat type largely influenced species presence, and habitat type, precipitation, and elevation were important in structuring community composition. This explanation fits with the hypothesis that vegetation type and water-energy dynamics are important determinants of species richness.

The analysis of community composition at the subfamily level shows this to be too broad a taxonomic level for describing sphingid communities at a small scale. Even at the generic level it is likely that many of the community composition changes would have been missed, since most genera are found throughout the altitudinal gradient. Analyzing data at the species level seems to be the most accurate means of describing community composition dynamics. Analysis of sphingids at the subfamily level does reveal, however, that although species, and even some genera, are restricted to particular habitats, none of the subfamilies appear to specialize in any one habitat or elevational range. Beck et al. (2006a) investigated the relative contributions of sphingid subfamilies to faunas of Southeast Asia and Malesia, and found that forest cover was a strong, though not significant, predictor of Smerinthinae and Macroglossinae. This rainforest effect was strongest for Smerinthinae, which they attributed to life history characteristics. Beck et al. (2006b) showed disturbance had a significant effect on subfamily frequency (negative for Smerinthinae and positive for Macroglossinae) while there was only a trend between elevation and subfamily frequency. These studies suggest that disturbance is the main driving force behind proportional abundances of sphingid subfamilies. The small proportion of Smerinthinae, large proportion of Macroglossinae, and similar subfamily frequencies between all five sites in this study suggests that the five sampling locations have faced comparable levels of disturbance in the past.

There appear to be two distinct moth communities: a lowland community which extends up to Vitobamba, and a montane community which extends down to Vitobamba. Differences in these communities were due solely to changes in species presence rather than a combination of changes in the proportional abundance or richness of subfamilies. Ignatov et al. (2011) also described a distinct split between lowland and montane sphingid communities. Although cosmopolitan species, by definition, were found throughout the gradient, they too followed a pattern similar to the community divisions. For example, Callionima acuta (Rothschild & Jordan, 1910), Enyo ocypete (Linnaeus, 1758), and Eumorpha phorbas (Cramer, 1775), all had low elevation centered populations with only a few individuals collected in the montane region, while Enyo lugubris (Linnaeus, 1771) showed a montane-centered population with few individuals from the lowlands. A difference in plant communities is the most intuitive explanation for this pattern. Below Vitobamba the dominant vegetation type is lowland forest, while areas above Vitobamba are dominated by montane forest. Ignatov et al. (2011) reported that many of the lowland genera feed on plants that are most abundant and diverse at lower elevations; species of Eumorpha, for example, specialize on “Vitaceae, represented by the genus Cissus in the study region, which is largely absent or less diverse at the cloud forest site [Wayquecha].” Since hawkmoths tend to be host plant specialists (Kitching and Cadiou 2000), it should be expected that these two forest types support different moth communities.

Endemic species contributed little to single site species richness, or single site abundance (Table 3). This seems contradictory to the findings of Ignatov et al. (2011), who reported a high number of endemic species (up to 23.4% at one site). This decrease is mainly a result of many of the endemics being reclassified after they were collected at a second site. For example, Adhemarius sexoculata (Grote, 1865), Erinnyis crameri (Schaus, 1898), Euryglottis aper, Euryglottis dognini Rothschild, 1896, and Xylophanes mariae Haxaire, 2013 were classified as endemics from Wayquecha by Ignatov et al. (2011), but were collected at Culebrayoc in this study, placing them in the montane classification. If montane ecosystems are thought to be hotspots for endemism (Meyers et al. 2000, Orme et al. 2005), why is the number of endemic species so low? Endemic as used in this paper, to describe species that were only collected at a single site, is somewhat misleading. This terminology was adopted to allow more accurate comparisons with the results from Ignatov et al. (2011). In their study on overall species richness, threatened species, and endemic species, Orme et al. (2005) defined endemic as the “25% of species with the smallest geographical breeding ranges.” The narrow definition of “endemic” as used in this paper likely underestimates the true number of endemic hawkmoth species. In future papers it may be more constructive to use a more traditional definition of the term.

Estimated species richness at Los Amigos, calculated from the corrected data, was lower than observed species richness from all 11 months of sampling. Reexamination of the raw data revealed that all the species which were lost in the correction process were only collected between November and May, the rainy season. Data used for analysis only came from the dry season. This raises the potential concern that, when data from only one season is used, non-parametric richness estimators cannot account for species turnover between seasons. Gotelli et al. (2010), using different methods from this paper, commented that few existing models account for imperfect detection, such as only sampling during the wet or dry season. This suggests that species richness was underestimated at all five sites. If it is the case that non-parametric estimators do not account for temporal turnover, researchers may face problems when trying to estimate species richness from single season samples.

One other potential problem is the small number of sample sites and the lack of replicates. Such a coarse-grained view makes it difficult to determine the exact elevation of peak species richness. Longino and Colwell (2011) faced a similar problem when studying litter ants in Costa Rica. A true mid-elevation peak (~1,400 m) seems unlikely, and in fact may not be expected (Beck and Chey 2008). A finer altitudinal resolution with replicates at some elevations will give a clearer picture of the elevation-species richness relationship, and allow for more rigorous hypothesis testing of determinants of species richness.