This research was conducted at Tiputini Biodiversity Station (TBS) (0°37'55"S, 76°08'39"W; 190–270 m elevation a.s.l.) (Fig. 1A, B) located in the Ecuadorian Amazon, within the province of Orellana. TBS consists of 650 ha of moist tropical rainforest and borders the megadiverse Yasuní National Park (Bass et al. 2010; Myster 2016). The forest surrounding TBS has experienced relatively minimal anthropogenic disturbance compared to other parts of the Amazon. The area has average annual temperatures of 24–27 °C and high yearly rainfall of ~ 3200 mm. The climate is considered aseasonal compared to other areas in Amazonia, as no month receives < 100 mm of rain. However, there are seasonal patterns in precipitation with rainier months (‘wetter season’) from April through June and September–October and drier periods (‘drier season’) in August and November–January (Pitman 2000; Blake et al. 2011; Myster 2016).

Figure 1. 

Maps of the field sites A country boundaries of South America with the Amazon Basin indicated by the heavier outline B Ecuador with the locations of TBS as a red circle and the boundaries of Yasuní National Park shaded C a DEM of the study area. The 24 sampling sites are indicated, with blue squares representing FP forest sites and green circles for TF forest. Latitude and longitude (in DD) shown along x and y axes. Maps generated through R packages: ‘raster’ (Hijmans 2015), ‘sp’ (Pebesma and Bivand 2005; Bivand et al. 2013), ‘GISTools’ (Brunsdon and Chen 2014), and ‘maps’ (Becker et al. 2015).

The area is relatively flat but several ridges, ~ 25–50 m higher than watercourses, add slight topographic variation. Approximately 90% of the landscape surrounding TBS is non-flooded TF which occurs along the slopes and tops of ridges. Rivers dissect these large blocks of TF forest creating narrow bands of FP forest which comprise ~ 2% of the study area (Pitman 2000; Pitman et al. 2001; Guevara et al. 2012; Pitman et al. 2014; Myster 2016). The palm, Iriartea deltoidea Ruiz & Pavn is a dominant tree species in both FP and TF forests near TBS (Pitman et al. 2001; Myster 2014), which are described in more detail by Burnham et al. (2001), Pitman et al. (2001), and Myster (2014, 2016).

Floodplain forest occurs within a relatively narrow area along the Tiputini River, which has been described as a mixed-water river (contributions of white and black water), or other times as white water (várzea), with a relatively high sediment load (Pitman 2000; Myster 2016). The height and volume of the Tiputini River is most strongly influenced by rainfall and snowmelt in the Andes Mountains (Prance 1979; Pitman 2000; Burnham 2002). See Appendix A for information about mean annual river heights for the years relevant to this study. Forests are inundated by the river for various lengths of time depending on elevation and distance from the river (Burnham et al. 2001; Pitman et al. 2001; Mertl et al. 2009). Previous studies have estimated that the floodplains around TBS are inundated (< 3 m) 1–3 times a year with inundations usually lasting less than a week (Pitman 2000; Burnham et al. 2001; Leimbeck and Balslev 2001; Burnham 2002; Mertl et al. 2009; Myster 2016).

Tiputini Biodiversity Station is located on a geologically young landform consisting of fluvial deposits (red clays and alluvium) originating from the Andes Mountains and are rich in exchangeable bases (Hoorn et al. 2010). Soils of TF have been classified as Ultisols which are clayey, acidic, and high in iron and aluminum (Pitman 2000; Sombroek 2000; Tuomisto et al. 2003). FP soils are less acidic, lower in iron and aluminum but higher in other nutrients (Ca, Mg, K, Na, P) and organic carbon compared to TF soils (Kapos et al. 1990; Duivenvoorden and Lips 1995; Pitman 2000).

Carabid beetles were sampled during the wetter months of June and July in 2011 and 2012 and during the transition to the drier season in October and November of 2011. Two sampling techniques, flight intercept traps (FITs) and hand sampling, were deployed sites in the forest understory, 12 in each of the two forest types (Fig. 1C). Both methods were employed in the rainier seasons but for logistical reasons only FITs were employed in the transitional season in 2011. The number of FIT trapping days varied slightly between forest types due to flooding events in the 2011 wet season, ranging from 33–49 days in the 2011 wetter season, 30–31 days in the 2012 wetter season and 29 days in the 2011 transitional season.

Each FIT consisted of PVC tubes, screen netting, a plastic awning (to prevent rain and debris from entering the trays) and plastic collection trays. Screening was stretched between two 1 meter PVC tube, with a 1 m support PVC tube across the top of the trap to create an ~ 1 m² flight barrier. See Riley et al. (2016) for additional details on FIT materials and implementation. Plastic trays were placed underneath the flight barrier to collect specimens and filled with water and biodegradable soap. Sampling trays were emptied every other day by filtering the contents through a metal strainer with mesh openings < 1 mm.

Since the majority of Carabidae are active at night, hand sampling occurred between 8:30 pm and 10:30 pm with the use of headlamps. This included active searching of the ground (leaf litter, etc.), tree trunks and vegetation (up to eye level), and decaying logs. A total of seven hand sampling events was performed at each of the 24 sampling sites durng two field seasons, four at each site in 2011 and three in 2012. Each sampling event consisted of 0.5 hour of search effort, mostly along a trail, within 100 meters of the FIT for each site. The first author participated in every hand sampling event with the assistance of one or two trained volunteers. Immediately after capture, carabids were preserved in 95% ethanol.

The main goal of this study was to compare richness, diversity, and assemblages of Carabidae between FP and TF forest types; therefore, data from both FITs and hand collecting were simply pooled by each sampling site for the analysis presented in this paper. However, we recognize that collection techniques operate differently and strongly influence the composition of fauna sampled (Gadagkar et al. 1990; Kitching et al. 2001; Missa et al. 2009). In a separate paper, Riley et al. (2016), we compared numbers of individuals, species, diversity, and community-level analyses between hand sampling and FITs for the two forest types, as well as reported taxa-specific results.

Carabid beetles were sorted using a stereo microscope and identified to genera and morphospecies. Since a species-level taxonomic key is not available for the study region, a genus-level key for French Guyana Carabidae (Erwin et al. 2012) was used, with subsequent identification to morphospecies and taxonomic confirmation by TE^†. Only two species, a widespread Neotropical tiger beetle Odontocheila cayenensis Lam. and the hiletine, Eucamaragnathus batesi Chaudoir, could be confidently referred to described species. In addition, the single oodine specimen collected could be confidently referred to Oodinus but could not be confidently determined as belonging to either O. piceous Motschulsky, O. amazonas Chaudoir, or O. limbellus Chaudoir. Otherwise, identifications were restricted to morphospecies. The use of morphospecies for Neotropical arthropods has been successful in ecological studies and is the most feasible option for an extraordinarily speciose region, with the majority of species not described by science (Oliver and Beattie 1996; Stork et al. 2008; Grimbacher and Stork 2009). Since the vast majority of the data set consists of morphospecies, we will use the term morphospecies unless specifically referencing one of the described species mentioned above. Voucher specimens representing taxa collected in this study will be deposited at Museo de Historia Natural, La Escuela Polytecnica Nacional (EPNC) in Quito, Ecuador. Abbreviations used in this paper are as follows:

ABL apparent body length;

ACE abundance-based coverage estimator, ACE;

DEM digital elevation map;

FITs flight intercept traps;

FP floodplain;

1/D inverse Simpson index, a Hill diversity number;

IndVal an indicator statistic, developed by Dufrene and Legendre (1997);

ISA Indicator Species Analysis;

N number of individuals;

NMDS nonmetric multidimensional scaling;

Pentb the most commonly collected cicindelid Pentacomia morphospecies;

S number of morphospecies (i.e., species richness);

S.E. calculated standard error;

E_1/D Simpson’s evenness;

TBS Tiputini Biodiversity Station;

TF terra firme.

As a precursor to our statistical comparisons of FP and TF forests, the data set was tested for spatial autocorrelation. First, a multivariate Mantel correlogram was constructed (Mantel 1967; Oden and Sokal 1986; Sokal 1986; Borcard and Legendre 2012; Legendre and Legendre 2012) using the ‘vegan’ package in R (Oksanen et al. 2016; R Core Team 2016). Inputs for this analysis were the Bray-Curtis dissimilarity for species assemblages and the Cartesian coordinates of the sampling sites as determined using R package ‘SoDA’ (Chambers 2013). If a linear trend was detected, species data were Hellinger-transformed and detrended prior analysis to satisfy normality and the second-order stationarity condition (Borcard et al. 2011). Sturge’s rule was used to determine the number of distance classes (Borcard et al. 2011). Following 5,000 permutations the significance of the normalized Mantel statistic for the distance classes were evaluated using Holm correction for multiple tests (Holm 1979; Borcard et al. 2011; Goslee and Urban 2013). Autocorrelation analyses were completed using R package ‘vegan’ (Oksanen et al. 2016; R Core Team 2016). To correct for a linear trend (F = 3.08, df = 2, P < 0.01), data were Hellinger-transformed and detrended prior to construction of the Mantel correlogram. There were six distance classes, with the smallest ranging from 0 to 32.5 meters; however, no significant spatial correlations were detected so all sample sites were treated as independent sites in all statistical analyses.

Data from both sampling techniques were simply pooled by sampling site to determine the N, S and 1/D (Simpson 1949; Jost 2006; Chao et al. 2014). To detect changes at higher taxonomic levels, the numbers of individuals and species for each carabid tribe were also analyzed. For the analyses listed above, values for FP and TF sites were compared using a two-sample t-test with equal variance unless assumptions of the test were not met. If the assumption of homogeneity was violated, we used two-sample t-tests with a Welch correction. When data failed to meet the assumption of normality, a Wilcoxon rank sum test was used.

Mean body length was compared between forest types using ABL. This is the length from the extreme anterior point of the mandible to apex of elytra and is a standard measurement used in many ecological studies to provide a reliable estimate of overall size for Carabidae (e.g., Erwin and Kavanaugh 1981). For each ground beetle morphospecies, all intact specimens (i.e., the head, thorax, and abdomen were undamaged) were measured and, due to their greater abundance in the sample, at least 21% of the specimens for each tiger beetle (Cicindelini) morphospecies were measured. For each sampling site, mean ABL was determined by averaging the measurements of all individuals collected over the entire study period. The mean site ABL values for FP and TF sites were compared using a two-sample t-test with equal variance.

To account for differences in numbers of individuals collected between forest types and under-sampling bias, sample size was standardized by rarefaction which determines the expected number of morphospecies from a random subsample of individuals from the overall data set (Gotelli and Colwell 2001, 2011). Rarefaction curves for each forest type were constructed in ‘iNEXT’ using R software (Hsieh et al. 2015), using 500 randomizations to determine 95% unconditional confidence intervals (Chao et al. 2014). The x-axis was scaled to individuals to show the increases in number of morphospecies as more individuals were collected (Gotelli and Colwell 2001, 2011; Colwell et al. 2004, 2012). To estimate the number of undetected morphospecies, rarefaction curves were extrapolated by a factor of two from the smallest sample size between forest types (Colwell et al. 2012). The abundance-based non-parametric species richness estimator, Chao1 was used to extrapolate the rarefaction curves. Chao1 has been recommended as the lower bound of asymptotic richness (Chao 1984; Hortal et al. 2006; Colwell et al. 2012).

We attempted to account for estimator bias, by calculating two additional non-parametric diversity estimators with standard errors (randomizations = 1000) using the R package ‘vegan’ (Oksanen et al. 2016): the first order jackknife, Jack1 (Burnham and Overton 1979; Heltshe and Forrester 1983) and the abundance-based coverage estimator, ACE (Chao and Lee 1992; Chao and Yang 1993; Lee and Chao 1994). Estimators were selected based on high performance reported in previous studies (Chazdon et al. 1998; Brose and Martinez 2004; Walther and Moore 2005; Hortal et al. 2006; Gotelli and Colwell 2011), as well as the species abundance distribution and evenness of the data, all of which affect estimator biases. Since ACE accounts for the frequency of rare species, it is more appropriate for datasets with a high proportion of rare species as is common for studies examining carabid assemblages (Chao and Lee 1992; Chazdon et al. 1998; Belaoussoff et al. 2003). Jack1 has performed well in a variety of studies (Colwell and Coddington 1994; Walther and Moore 2005; Hortal et al. 2006; Basualdo 2011) with low bias for data with higher evenness and more mobile organisms (Palmer and Dixon 1990; Brose and Martinez 2004). Results from the three estimators were averaged to provide conservative values relating to sample completeness.

Simpson’s evenness (E_1/D) values (Williams 1964; Smith and Wilson 1996; Payne et al. 2005) were calculated for each sampling site and the significance of the difference was determined by a two-sample t-test with equal variance. Log of relative abundance was plotted against morphospecies rank to compare morphospecies abundance distributions between FP and TF forests. In order to examine morphospecies commonness in the sample, morphospecies were assigned to three rarity categories based on the proportion of their abundance to the overall catch (relative abundance) over the entire sampling period with both forest types combined. Morphospecies were classified as ‘dominant’ if their relative abundance contributed > 10.0% of the total number of individuals collected. ‘Common’ morphospecies contributed 1.0–9.99% to the overall catch, and morphospecies were classified as ‘rare’ if they contributed < 1.0% to the total catch. Differences between FP and TF forests in numbers of individuals and morphospecies in these three rarity categories were analyzed through two-sample t-tests or, when t-test assumptions were not met, nonparametric Wilcoxon tests.

We used a generalized linear model with Poisson error distribution and a log link to test the hypothesis that common morphospecies should be collected from a higher number of sample sites. Significance was determined by analysis of deviance using the Wald chi-square test statistic. Linear modeling was completed using R packages ‘lme4’, ‘car’, and ‘multcomp’ (Hothorn et al. 2008; Fox and Weisberg 2010; Bates et al. 2015; R Core Team 2016). We determined whether the proportion of common and rare morphospecies were similar among tribes using a chi-square test. Dominant and common morphospecies were combined for this analysis (now termed ‘abundant’) to meet the assumptions of the test. Expected numbers for each category were determined by multiplying the total number of morphospecies collected in the tribe by the overall proportion of morphospecies in each rarity category (e.g., 125 of 143 morphospecies were rare, 87.4%).

Morphospecies assemblages for FP and TF were also compared using NMDS (Minchin 1987; Oksanen 2015). Bray-Curtis (abundance-based) and Jaccard (presence/absence) dissimilarity values were used as the distance measures. Non-metric multidimensional scaling (NMDS) is a robust unconstrained ordination method which makes no assumptions about the underlying nature of species distributions. Since the abundance mean-variance relationships were similar between FP and TF forests, untransformed abundance values of all morphospecies were used as the inputs for community-level analyses (Minchin 1987; Clarke 1993; McCune and Grace 2002; Nekola et al. 2008). To ensure solution stability, we used 100 runs for all NMDS analyses with random start points (Minchin 1987). Statistical significance between FP and TF distance matrices was determined using adonis, a robust permutational multivariate analysis of variance (permutations = 999) (Legendre and Anderson 1999; Anderson 2001; McArdle and Anderson 2001; Oksanen 2015). Dispersions within forest types were tested for multivariate homogeneity using function betadisper and TukeyHSD for the post-hoc test (Anderson 2006; Anderson et al. 2006; Oksanen 2015). All community-based analyses completed using R package ‘vegan’ and ‘MASS’ (Venables and Ripley 2002; Oksanen et al. 2016; R Core Team 2016) .

We sought to identify morphospecies that drove differences in community morphospecies assemblages using ISA. The IndVal statistic, developed by Dufrene and Legendre (1997), is insensitive to community-level beta diversity and does not consider absences as negative preferences (Dufrene and Legendre 1997; De Cáceres and Legendre 2009). Only morphospecies collected at least three sites (i.e., 25% of the number of sites in each forest type) over the entire sampling period were included in the ISA. Indicator value and statistical significance were determined using function multipatt in R package ‘indicspecies’ (permutations = 999). We used the Holm correction for multiple testing to reach experiment wide conclusions (Holm 1979; De Cáceres and Legendre 2009).

Results from our study indicate that richness, diversity, and morphospecies composition of understory Carabidae from eastern Ecuadorian lowland rainforests are influenced by forest type. As expected, the simple comparison between hand and FIT collections presented here suggests the two methods sampled different portions of the carabid fauna in each forest type, similar to results reported in other tropical arthropod studies (Gadagkar et al. 1990; Kitching et al. 2001; Missa et al. 2009; Riley et al. 2016). Employing both collection techniques sampled a greater portion of the overall carabid community than either collection technique alone, providing a more comprehensive sample of Carabidae in FP and TF forests.

The number of individual carabids collected in the understories of FP and TF forests did not differ significantly. In contrast, Lamarre et al. (2016) found Carabidae, including cicindelids, were strikingly more abundant in seasonally flooded forests than in TF forests in French Guiana and Peru, with ~ 91% of cicindelids and 75% of all remaining carabids collected in flooded forests. This disparity may be due to several factors, including differences between the studies in species composition, sampling protocols, sample completeness, and the time period and duration of sampling periods. Other studies in Central Amazonia found higher tiger beetle richness in TF forests (Adis et al. 1998; Zerm et al. 2001; Adis and Junk 2002). In our study, all four Cicindelini species occurred in both FP and TF forests but significantly more cicindelid individuals were collected from TF forest. Some studies from other parts of the Amazon reported higher tiger beetle abundance in TF forests (Zerm et al. 2001; Adis and Junk 2002), in contrast higher abundance has also been reported in FP forests (Pearson and Derr 1986; Lamarre et al. 2016). Thus, the data are insufficient to suggest a general abundance pattern for Cicindelini occurring in these two Amazonian forest types.

Studies of invertebrate assemblages as a whole have generally found lower abundances within the understory of FP forests compared to TF forests (Pearson and Derr 1986; Mertl et al. 2009; Lamarre et al. 2016). Since the flooding cycle causes seasonal movements and shifts in the biotic communities of riparian areas, the time of year will likely affect the relative number of individuals occurring in the understories of these two forest types (Irmler 1979a; Adis 1981; Erwin and Adis 1982; Adis and Junk 2002). Abundance patterns of adult Carabidae vary in space and time, and are affected by factors such as flooding duration, frequency, and intensity (Adis 1981; Bonn 2000; Bonn et al. 2002; Lambeets et al. 2008, 2009).

Although mean morphospecies richness differed only marginally (P = 0.06), all other measures of richness and diversity were significantly higher in FP forest compared to TF forest, corroborating results of some previous studies (e.g., Erwin and Adis 1982; Pearson 1984; Pearson and Derr 1986). The higher species richness and diversity of FP forest seems to reflect the higher number of rare morphospecies sampled there than in TF forest. Increases in Carabidae richness after flooding events has been recorded for Central Amazonian forests (Adis 1981), as well as for temperate forests (Bonn 2000; Ellis et al. 2001; Bonn et al. 2002; Lambeets et al. 2008, 2009). After inundation, the litter layer and ground vegetation of FP habitats at TBS are covered by a thin layer of mud and sediments (pers. obs.; Mertl et al. 2009). This may result in higher nutrient content or resource availability when waters recede that promotes features like higher prey abundance that are attractive to particular carabid species (Irmler 1979a, b; Duivenvoorden and Lips 1995; Bonn 2000; Adis and Junk 2002; Haugaasen and Peres 2005; Adis et al. 2010). Disturbance may increase the number of unoccupied microhabitats in post-flood areas, generating a reciprocal increase alpha diversity over time (Salo et al. 1986; Erwin 1991b; Kalliola et al. 1991; Zerm and Adis 2001a; Wittmann et al. 2006). Lastly, the disturbance of flooding could also moderate competitive exclusion, resulting in higher diversity in floodplain forests by promoting higher levels of coexistence (Levin and Paine 1974; Connell 1978; Erwin and Adis 1982; Zerm and Adis 2001a, b).

Differences in carabid beetle richness and diversity patterns among localities within Amazonia may reflect variation among sites, particularly because erosion, deposition and changes in river channels continuously drive spatial and temporal changes in flooded forests (Kalliola et al. 1991; Junk 2000). Faunal richness and diversity have been strongly tied to length, frequency and severity of flooding events (Bonn et al. 2002), and these vary among Amazonian rivers depending on factors such as flooding cycles, topography and weather patterns (Junk et al. 1989; Junk 2000; Tuomisto et al. 2003; Adis et al. 2010; Junk and Piedade 2010). Temporarily flooded forests with less severe and frequent flooding regimes, such as those near TBS, will likely show different patterns in richness and diversity than those subjected to more intense flooding regimes.

Erwin (1979) and Erwin and Adis (1982) previously suggested that a large proportion of carabid species inhabit riverine and wetland habitats in Amazonia. A long history of carabid evolution in tropical wetland habitats (Erwin 1979, 1985), is consistent with adaptation of many species to changing environmental conditions in riparian zones (Adis 1982). Carabid traits that promote persistence during unfavorable periods of the flooding cycle include: synchrony of life cycles with patterns of disturbance, reproductive dormancy, and submersion tolerance (Erwin 1979; Adis 1982; Erwin and Adis 1982; Paarmann et al. 1982; Erwin 1985; Adis et al. 1986; Amorim et al. 1997; Zerm and Adis 2001b, 2003). Although Amazonian wetland and floodplain habitats have experienced repeated reductions and expansions through changes in paleoclimate (e.g., Cheng et al. 2013), they have doubtlessly continued to be available supporting persistence and radiation of species with adaptions to flooding cycles (Erwin and Adis 1982; Kubitzki 1989; Irion et al. 1995, 1997, 2010; Müller et al. 1995; Haffer and Prance 2001; Adis and Junk 2002; Hoorn et al. 2010; Wittmann et al. 2010).

Carabids have patchy distributions, particularly in tropical rainforests, making the average abundance of many species appear low and requiring higher collection effort to obtain a representative sample of the fauna (Erwin 1979; Paarmann et al. 2002; Erwin et al. 2005). Sample completeness, based on the average of three nonparametric diversity estimators, was greater in FP forests (~ 74%) than TF forests (~ 55%). This, in turn, influences the observed richness and diversity values for both forest types. A possible contribution to the lower observed richness and sample completeness in TF forests was the lower efficiency of hand sampling at TF sites compared to FP sites. While the exact reason for the lower efficiency is unknown, possible explanations include an increased vertical stratification of Carabidae in TF forest and higher aggregation of carabid spatial distributions in FP forests. Additionally, non-cicindelid carabid individuals collected from TF sites were, on average, smaller relative to those collected at FP sites (Fig. 4) and such differences may also negatively influence the collectability of individuals in TF forests (see Riley et al. 2016 for further discussion).

Based on extrapolation of rarefaction curves, many morphospecies remained uncollected in TF forests. Although the three diversity estimators we employed gave somewhat different answers, the differences between FP and TF assemblages were relatively small. Nonetheless, we predict that the observed difference in morphospecies richness between FP and TF sites will decrease with greater sampling effort. Within Amazonia, TF forests are more extensive than FP forests, and thus given standard expectations of species-area relationships, richness and diversity of TF forests are expected to be greater than for FP forests across the region (Rosenzweig 1995; ter Steege et al. 2000; Pitman et al. 2001; Burnham 2002; Tuomisto et al. 2003; Myster 2014). Compared to TF forests, FP forests are limited to narrow areas along the banks of rivers and lakes and, at the landscape level, are more fragmented (ter Steege et al. 2000; Pitman et al. 2001).

Differences between FP and TF forests occurred not only at the morphospecies level but also at the tribal level. There were higher numbers of Scaritini and Lachnophorini individuals and/or morphospecies in FP forest and higher numbers of Pentagonicini and Cicindelini individuals and/or morphospecies in TF forest. Although life histories are unknown for the vast majority of Amazonian carabid species, general life history information is known for many of the sampled genera and tribes. Species within Scaritini and Lachnophorini are typically ground-dwelling whereas those of the Pentagonicini and Cicindelini are more likely to be sampled on vegetation (per. obs.; Erwin et al. 2012). Many Lachnophorini and Scaritini species occur in wetter habitats which may explain why they were collected in greater numbers and diversity in FP forests in this study (Erwin et al. 2012; Erwin and Zamorano 2014). Indicator species analysis (ISA) showed two genera, Peruphorticus (Lachnophorini) and Paratachys (Bembidiini), had one morphospecies characteristic of FP and another morphospecies characteristic of TF forest. Differences in microhabitat associations among tribes and genera likely attributed to observed differences in tribal-level richness and abundances, as well as variations in morphospecies composition between forest types.

Rank abundance curves have rarely, if ever, been reported for tropical Carabidae. In temperate and boreal forests, carabid communities typically have a few dominant species with the majority of species being rare (Belaoussoff et al. 2003), similar to the assemblages for FP and TF forests in this study. Dominant and common morphospecies comprised 72.2% of the total number of individuals collected but only 13.3% of the total observed morphospecies. As with many tropical insect assemblages, the contribution of rare species was especially high (Novotný and Basset 2000), with rarer species accounting for a greater proportion of the carabid assemblages than in temperate communities. In our study, singletons contributed slightly less than half of the overall morphospecies richness (43.8%), which is more than a canopy fogging study in a nearby Ecuadorian TF forest where 28.6% were singletons (Lucky et al. 2002).

Chi-square analyses examining rarity categories among carabid tribes showed abundant morphospecies were significantly more likely to be within the Bembidiini, Cicindelini, Lachnophorini, Loxandrini, and Pentagonicini, with fewer morphospecies of Lebiini collected than expected. Although Lebiini was by far the most morphospecies rich tribe (S = 35), only two morphospecies were classified as common while the remaining 33 morphospecies occurred at low numbers. In contrast, all four of the Cicindelini morphospecies were classified as abundant. Common morphospecies were sampled from a higher number of sampling sites within the study area, suggesting that they have wider spatial distributions than less frequently encountered morphospecies. This agrees with previously reported positive relationships between local population abundance and the occurrence at a greater proportion of sample sites (Hanski 1982; Brown 1984; Gaston and Lawton 1988; Niemelä and Spence 1994).

Analyses indicate distinct carabid morphospecies assemblages occur in FP and TF forests, and this is further supported by the number of characteristic morphospecies for each forest type. In our study, only 32 of 143 morphospecies were collected in both forest types, suggesting FP forests maintain unique assemblages compared to neighboring non-flooded TF forests. Near Manaus, Brazil, Zerm et al. (2001) reported only two of 25 tiger beetle species occurring at both FP and TF forest sites. Unique species assemblages for FP and TF forests has been recorded for tropical vegetation (ter Steege et al. 2000; Bredin et al. 2020), birds (Remsen and Parker 1983; van Lieshout et al. 2016), understory arthropods (Adis and Junk 2002; Lamarre et al. 2016), canopy arthropods (Erwin 1983b; Adis et al. 2010), spiders (Höffer 1997), and understory tiger beetles (Zerm et al. 2001). Differences in species assemblages may relate to factors such as flooding regime adaptions, differences in plant composition, forest structure, and resource availability between FP and TF forests. The spatial and temporal distribution of tiger beetle guilds have been shown to be influenced by specific microhabitats within a given habitat (such as open areas, forest type, etc.) (Zerm and Adis 2001a; Zerm et al. 2001; Adis and Junk 2002). Lamarre et al. (2016) reported significant associations for seasonally FP forest or TF forests for several arthropod families, with carabid beetles demonstrating a particularly strong association with seasonally flooded forests.

Since flooded forests oscillate between aquatic and terrestrial phases and there is high variability in flooding events among rivers, it is difficult to adequately measure the overall biodiversity and community composition of flooded forests as a whole (Junk et al. 1989; Junk 2000). Floodplain communities are a mix of migrants, visiting species and residents, the latter likely with highly specialized adaptions for inundation events (Adis 1997; Junk 1997; Adis and Junk 2002). Thus, presence of both resident and migrant species in FP forest may have contributed to the higher observed richness, diversity, and the higher number of rare morphospecies collected in the FP forest. Carabids seeking refuge from flooding events may move into the canopy or sub-canopy strata via tree trunks (Irmler 1973, 1979a; Adis 1981; Adis and Messner 1997), move horizontally following the waterline (Irmler 1979a; Adis and Messner 1997; Zerm and Adis 2003), or fly to non-flooded uplands (Irmler 1973; Adis 1982; Erwin and Adis 1982; Adis et al. 1986). Thus, selection for escape from flooding may help explain why the majority of carabid beetles living in tropical forests are fully winged and presumably capable of flight (Erwin 1979). Other specializations to flooding regimes have also been reported for adults and possibly larvae, including retreat to subsoil layers (e.g., a flightless Stratiotes species), aestivation and submersion tolerance (Adis 1982, 1997; Adis et al. 1986; Rothenbucher and Schaefer 2006; Lambeets et al. 2009).

Terborgh and Andresen (1998) suggest that vegetation of floodplain forests may be more closely allied taxonomically to adjacent non-flooded forests rather than to other flooded forests. Conversely, specific adaptions to life in flooded forests could promote convergence in floodplain community composition, even at broad geographic scales (e.g., Lamarre et al. 2016). In the present study, morphospecies in two genera (Paratachys and Peruphorticus) were significantly associated with either FP or TF forests suggesting that there are evolutionary pressures to specialize with respect to forest type. Or, it could be a combination of both factors. For Amazonian frogs and mammals, Gascon et al. (2000) showed habitat type (FP vs. TF forest) and geographic distance were the most significant predictors of community similarity.