The study was conducted in the Ningxia region (northern China), between 36°N and 38°N and between 105°E and 108°E, in three types of grassland ecosystems: desert, typical and meadow steppes (Kang et al. 2007). A map of the study area with photographs of the three grassland ecosystems is given in Tsafack et al. (2019b). The desert steppe was characterized by a vegetation mainly represented by drought-tolerant species such as Agropyron mongolicum, Artemisia desertorum, Artemisia blepharolepi, and Stipa spp. The typical steppe included natural patches of grass (Stipa bungeana, S. grandis, Artemisia frigida, Thymus mongolicus, Heteropappus altaicus, and Potentilla acaulis) in association with cut grasses used as fire belts and crops. Vegetation in the meadow steppe was mainly represented by Festuca brachyphylla, S. bungeana, A. frigida, and Achnatherum splendens.

We selected three main sampling areas representing these three types of steppes. To reflect within-ecosystem variability of the typical and meadow steppe habitats, on the basis of vegetation characteristics, we identified three habitat types (that we refer to as sectors 1, 2, and 3) within the typical steppe, and two habitat types (sectors 1 and 2) in meadow steppe. Data were gathered from 90 sampling sites distributed as follows: 15 sites in the desert steppe, 45 sites in the typical steppe (15 sites in each sector), and 30 sites in the meadow steppe (again 15 sites in each sector). In each sector, sites were selected haphazardly (i.e., without any regular spatial arrangement) and separated by at least 150 m to avoid, or at least reduce, possible autocorrelation.

At each sampling site, five pitfall traps (separated by at least five meters from each other) were installed. Pitfall traps consisted of 7.15 cm-diameter plastic cups, sunk in the ground with the cup-lip level with the soil surface, and filled with 60 ml of a mixture of tap water and vinegar (8%), sugar (4%), and 70% alcohol (4%). Sampling was done from May to September 2017. During the sampling period, pitfall traps were placed in the sites once a month in mid-month, and left in the field for 72 h prior to collection. Traps were composed of two buckets, with the smaller inserted into the larger. At each sampling session, the smallest were extracted to collect the trapped beetles and then paced again in the largest, which were left dug into the soil. This ensured that trap position remained exactly the same over the sampling period and disturbance reduced to minimum. Thus, we collected 25 samples (5 pitfall traps × 5 sampling dates) for each site, for a total of 2,250 samples. This spatially intensive sampling allowed us to obtain extremely detailed data about species distribution at fine scale. Further details about study area and data collection can be found in Tsafack et al. (2019b, 2020a). For each site, we pooled the catches by period and calculated species abundance as the average among the five sampling dates. Carabids were identified to species based on keys and museum specimens with the aid of a taxonomist expert in carabid beetles (Prof. H. Liang, see Acknowledgments). Voucher specimens are preserved in the collections of the School of Agriculture, Ningxia University.

We calculated the Pianka index (Pianka 1973) of niche overlap to express spatial overlap between species pairs, using EcoSimR 1.00 (Gotelli et al. 2015):

$O_{jk}=O_{kj}=\frac{{\sum }_i^n{p}_{ij}{p}_{ik}}{\sqrt{{\sum }_i^n{p}_{ij}^2{\sum }_i^n{p}_{ik}^2}}$

where O_jk is Pianka’s index of niche overlap between species j and k, p_ij is the proportion of the ith resource used by species j, p_ik is the proportion of the ith resource used by species k, and n is the total number of resources. In our case, the resource is the habitat space (sampling sites), the use of which is assumed to be expressed by species abundances. The O_jk index is a symmetrical modification of the asymmetric index proposed by MacArthur and Levins (1967) for estimating competition coefficients from field data about resource utilization. It is an analogue of a correlation coefficient and ranges from zero (complete dissimilarity in resource utilization between two species) to one (complete identity in resource utilization). Pianka’s symmetric measure is more convenient than the original MacArthur and Levins’ asymmetric form (May 1974). Moreover, there is a general agreement that overlap measures cannot be used as true competition coefficients (Hurlbert 1978; Abrams 1980; Holt 1987) because there is an inverse relationship between competition and niche overlap (Pianka 1976, 1978).

Following widespread criticism of quantifying competitive interactions based on overlap indices (e.g., Connell 1980), more modern analyses of niche overlap are meant as description of interspecific interactions sensu lato, unless competitive interactions are known to occur a priori (Holt 2009; Lombardo and Mjelde 2014). Thus, following the current views in ecology, we have not automatically implied competition from O_jk values, but used them merely as measures of niche overlap (see Colwell and Futuyma 1971; Pianka 1974 for the distinction between overlap and competition). However, negative (competitive-like) mutual influences of two coexisting species may be suspected when overlap over limited key resources is high (Chesson 2011) but, on the long term, competitive exclusion should lead to low niche overlap. Co-existence does not imply competition, but we believe that our analyses can provide useful insights into potential competition processes.

To assess if the mean values of niche overlap differed from those expected by chance, we compared the observed average values of spatial niche overlap (i.e., the averages of the niche overlap values between each species pair) with the expected averages obtained from simulated null-assemblages (pseudo-communities) constructed with two alterative sets of rules (see below). In each case, we calculated 10,000 null-assemblages, a number of permutations considered enough to avoid algorithm biases in calculations (Lehsten and Harmand 2006).

Null-assemblages were simulated using Monte Carlo randomization algorithms that assign resource use values (in our case, number of individuals from different sampling sites) to each species. The choice of an appropriate model to construct null-assemblages is a critical issue. Lawlor (1980) developed four randomization algorithms that differ in whether utilizations are reshuffled or replaced by a random number, and in whether the zeros in the matrix are retained or not. These algorithms are referred to as RA1, RA2, RA3, and RA4. Both retaining/relaxing niche breadths and retaining/reshuffling zeroes have implications for the structure of the null-assemblage and affect the power of the test (Gotelli and Graves 1996). Theoretical and empirical analyses of these algorithms (Winemiller and Pianka 1990; Kobayashi 1991) led to the conclusion that Lawlor’s (1980) RA3 is the best existing algorithm for use in null models of resource overlap. This algorithm tests for community structure by retaining niche breadth (i.e., the amount of specialization) for each species through simulated specialization equal to the observed value, but reshuffles zero states (i.e., by randomly varying the particular resources that were used, in our case space utilization). This destroys any guild structure resulting strictly from the zero structure of the resource utilization matrix.

The RA3 algorithm relies on the equiprobability assumption that each species will use the resource in a site with the same probability or each site will have the same probability to host a species (in other words, differences between rows (species) or columns (sites) are not preserved). The RA3 algorithm tends to overestimate niche overlap if the equiprobability assumption is not met, because more abundant resources will be used by all species even if niche segregation occurs. Thus, for comparative purposes, we also used the RA2 algorithm, which tests for structure in the generalist-specialist nature of the resource utilization matrix by conserving guild structure (zero states are retained, thus preventing species that did not use a certain resource in the field from doing so in simulations), but relaxes niche breadth (i.e., it allows niche breadth to vary, thus assuming a random equiprobable specialization) (Gotelli and Graves 1996). In all cases, equiprobable resource use was a priori assumed in the analyses.

We also calculated species segregation by using the C-score (Stone and Roberts 1990). The C-score for a species pair jk is calculated as:

C_jk=(R_j–SS)(R_k–SS)

where R_j is the row total for species j, R_k is the row total for species k, and SS is the number of samples that contain both j and k. Thus, for any particular species pair, the C-score is a numerical index that ranges from a minimum of 0 (maximally aggregated) to a maximum of R_jR_k (maximally segregated with no shared samples). The matrix-wide C-score is an average of all the pairwise values of C-score for different species, so it reflects both positively and negatively associated species pairs. To establish whether the matrix had an average C-score significantly different from what can be expected from a null model, we compared the observed values with the expected averages obtained from 10,000 simulated null-assemblages using the fixed row-fixed column (FF) algorithm, which preserved row and column totals (Ulrich and Gotelli 2007; Strona et al. 2014).

While niche overlap analysis uses a data matrix with species- and site-specific observed abundances, the C-score searches for non-random structure in the species assemblage data by using a presence/absence data matrix (Gotelli and Graves 1996). High average C-scores indicates a low randomness, i.e., a greater likelihood that the distribution of one species has been directly affected by the presence of other species. Thus, for an assemblage that is competitively structured, the C-score should be significantly larger than a randomly assembled community (Gotelli 2000).

The role of competition in carabid beetles is debated. Interspecific competition in carabid beetle communities was reviewed in detail by Niemelä (1993), who found support for interspecific competition in carabid communities or evidence of resource partitioning in only half of the studies examined.

Laboratory studies have produced contrasting results. For example, laboratory research showed that mortality rates observed when two species of Pterostichus were reared together did not differ from those observed when they were reared separately (Paarmann 1966). However, other research found that, when reared separately, different species of the genus Scarites can show low mortality, but if they are reared together, they show extremely high aggressive behaviors even at low density (Peyrieras 1976). Experimental laboratory studies involving Pterostichus adstrictus Eschscholtz, 1823, and P. melanarius (Illiger, 1798) indicated that intraguild predation and interspecific competition for food among both larvae and adults have the potential to reduce survival and reproductive potential in nature (Currie et al. 1996). Nonetheless, such effects could not be documented in a two-year field experiment (Niemelä et al. 1997).

Field studies have produced contradictory results. Field research performed by den Boer (1979) showed population stability through years in co-occurring congeneric species, a result that contrasts with the idea that competition plays a major role in defining their distribution. These results were obtained by comparing species abundances of adult beetles found in pitfall traps kept at the same places over years. However, it is possible that adults do not compete at all, but that competition structures communities through larval competition (see, for example, Baars and Van Djik 1984). It has been also suggested that species that seem to co-occur might in fact occupy different microhabitats (Abrams 1986; Zalewski et al. 2014). In a study of two syntopic species of the genus Carabus, Lenski (1982) showed that C. sylvosus Say, 1823 attained a larger average body size where C. limbatus, 1825 was less abundant, suggesting the influence of competition (Lenski 1982). Studies based on serological (Sergeeva 1994) and isotopic approaches (Zalewski et al. 2014) have documented high niche overlap in species diets, which indicates that food does not constitute a primary factor of competition between some co-occurring carabid species.

It has been suggested that carabid species with overlapping trophic niches can co-occur if they have different circadian rhythms. For example, Notiophilus biguttatus (Fabricius, 1779) and Nebria brevicollis (Fabricius, 1792) not only have their activity peaks in different seasons (in spring and in autumn, respectively), but also show different circadian patterns, being diurnal and nocturnal, respectively (Dennison and Hodkinson 1983). Most species in a Norwegian carabid assemblage occupied different biotopes or differed in body size, and the only two species that occupied the same biotopes and had the same size showed different seasonal activity patterns (Erikstad et al. 1989). Pearson and Mury (1979) studied potential niche separation in an assemblage of Cicindela species. They showed that prairie-dwelling species had mandibles of different sizes, suggesting that they feed on animals of different sizes. However, they explained overlap in mandible sizes of species living near pond edges, by suggesting that competition was low there because of higher local abundance of animals that prey on these tiger beetles keeping their populations low relative to resource supply.

Based on the high frequency of co-occurrence of congeneric species, some authors (Darlington 1971; Basilewski 1984; den Boer 1986) have argued that competition is not a driver of community structure in carabid beetles. However, other studies have shown that congeneric species do not tend to co-occur, and that co-occurring species tend to differ in size, occupy different habitats, or have different mating periods (Peyrieras 1976; Brandl and Topp 1985; Loreau 1986). Character displacement has been described in Amblystogenium pacificum (Putzeys, 1869) and A. minimum Luff, 1972 (Davies 1987). These two species have very different body size where they co-occur, but where A. pacificum is alone its size is more similar to that of A. minimum (Davies 1987). A similar situation has been recorded in cave-dwelling species communities (Van Zant et al. 1978). In this paper, we have broadened the discussion to investigate potential competition in carabid communities by using null model approaches to test deviation of species coexistence patterns from randomness.

Niche complementarities (i.e., niche differences between species) have long been considered to be important key drivers of species coexistence (e.g. Darwin 1859; Macarthur and Levins 1967; Silvertown 2004; Stubbs and Wilson 2004; Mason and Wilson 2006). Nonetheless, local environmental conditions may blunt their effects, producing patterns in which co-occurring species are more similar than expected by chance (Leibold 1998; Mayfield and Levine 2010). Thus, niche similarities can be also observed, when the focal species are not regularly distributed in a niche space, but aggregated in groups of ecologically similar species, e.g., when resource limitation is not regularly a feature of the environment. Understanding how species sort themselves into communities is essential to understand how species co-occur to maintain biodiversity. Thus, important insights may come from observation and study of non-random patterns in community assembly. Such patterns do not necessarily depend on competition (Zalewski et al. 2014), but on the other hand they are not a simple product of chance and may depend on resource partitioning developed over evolutionary time frames (Behangana and Luiselli 2008). Use of null models helps us to discover non-random patterns and to link them to underlying mechanisms through further study. In fact, our null model approaches, in which the resource is the space designated by capture in a particular site, are useful to suggest potential species interactions for more direct tests. For example, species that are collected in the same place might not interact if they are active in different moments of the day. Yet, comparison of null models constructed by using different rules may disclose patterns that can be used to develop hypotheses about mechanisms. In particular, our use in this study of two different null models to understand the spatial distribution of carabid beetles provided complementary information.

The RA3 algorithm, which is based on the assumption of retaining niche breadth, revealed that the spatial niches of carabid species overlapped significantly and more than expected in both the desert and the typical steppe. These results were obtained when niche breadths were preserved but species were allowed to be modelled as present in sites where they were not found in the observed data set. This indicates that some places are unsuitable for particular species, and that existence of such places likely forces spatial niche overlap in places that are suitable. This result is consistent with the idea that in these ecosystems resources are scarce and structured spatially, so that unsuitable parts of the environment (here, for example, sandy areas without vegetation) may restrict species from finding areas of suitable habitat. This result seems to reflect that mobility of carabids is strongly constrained in desert steppe, as we found in a previous paper (Tsafack et al. 2020b). In that work, we demonstrated that there were significant correlations between carabid activity-density and vegetation indices at a small scale (300 m) in desert steppe, although they disappeared at large scales (buffers zones of 1400–1500 m diameter; Tsafack et al. 2020b). Furthermore, the significant spatial niche overlap observed in the desert steppe is likely related to the low community stability found in this ecosystem (Tsafack et al. 2019a). This instability may produce substantial temporal changes in species composition of local assemblages, which interferes with the evolution of competitive interactions. In contrast, in the typical steppe, spatial niche overlap was significant, but community stability was higher and did not differ from that in meadow steppe (Tsafack et al. 2019a). This result suggests that niche overlap can be high even when communities are stable, if the habitat (i.e., extent of vegetation cover) is homogeneous.

Our analysis was strictly about spatial distribution, and we did not consider explicitly the distribution of vegetation or productivity. However, the desert steppe supports low vegetative productivity that is not uniformly distributed, so we think that spatial segregation patterns likely reflect uneven resource distribution. Thus, that the species aggregation observed in our data is driven by beetle concentration in places where resources are more abundant. This is in agreement with den Boer’s (1986) “coexistence principle”, which states that ecologically closely related species frequently coexist in the same habitats, if interspecific interference is not more important than intraspecific competition.

The carabids of the desert community also showed the highest variance in niche values, i.e., some species pairs show high niche overlap and others show low niche overlap. Thus, we argue that spatial resource partitioning is due to the presence of sites that are favorable to certain species but not to others, which supports our prediction H1. The assumption of resource partitioning can also be supported by the high variance observed in carabid functional diversity in desert steppe (Tsafack et al. 2019b), which suggests that species are associated with different microhabitat or use different resources.

The carabid community of the meadow steppe did not deviate significantly from randomness under the RA3 assumptions, which means that the environment is used in a more uniform way, with no or few spaces unavailable to carabids. This is consistent with the greater landscape homogeneity of this ecosystem as postulated by our prediction H2. A previous study showed that carabid activity-density in the meadow steppe was significantly correlated with vegetation indices at a relatively large scale (buffers zones of 1450 m and 1500 m diameter), thus suggesting that carabid species in this ecosystem are more mobile and can exploit larger areas of suitable habitat (Tsafack et al. 2020b). Thus, differences in dispersal ability are consistent with observed differences in species co-occurrences (Zalewski et al. 2018).

Modelled with the RA2 algorithm (which retains the zero-structure of the matrix, but allows niche breadths to vary), niche overlaps in the beetle assemblage of the desert steppe did not vary significantly from randomness. This indicates that some places are unavailable to some species, but there are no other constraints on space utilization. Although the carabids of this ecosystem divide the shared spatial resource in a random way, some places are apparently unsuitable for particular species. For data from the typical steppe, however, niche overlap values from the RA2 algorithm were lower than expected if determined by chance, suggesting that the environment is spatially structured. In other words, presence of zeroes in the matrix of habitat use are important, and if they are removed niche overlap will increase significantly. Thus, the unsuitability of some sites for particular species contributes to the pattern of spatial utilization among species. Similarly, in the meadow steppe, use of the RA2 algorithm leads to lower niche overlap than expected by chance; however, if the influence of zeroes is removed (RA3), niche overlap does not increase to become significantly higher than by chance. This suggests that places unavailable to particular species are less important to assemblage structure in the meadow steppe than in the other two ecosystems. This interpretation should be offered with caution however, because use of RA2 algorithm is prone to Type I error (false positive) (Gotelli et al. 2015).

If the three sectors of the typical steppe are analyzed separately, neither the RA2 nor RA3 approach revealed non-random community structure. This suggests that spatial structure in the overall analysis arose because the three sectors include different assemblages. This is not surprising as the abundance of even the most widely distributed species (C. vladimirskyi) varied widely in this ecosystem, being trapped commonly in two sectors, but rarely in the third. Thus, spatial niche separation likely results from low overlap in patterns of habitat use among these three assemblages more than spatial segregation within each sector. In other words, spatial segregation was greater between species found in different sectors than between species within the same sector. This is an interesting result showing that spatial niche overlap values should be interpreted with consideration of the scale of analysis. At a broad scale, niche separation appears to result from species segregation reflecting their different distributions, but when the analysis is conducted at a finer scale, species appeared to be no more segregated than expected by chance. These results are consistent with Loreau’s (1986) results about forest carabids, where there was no spatial separation of species because of habitat homogeneity.

The situation is paralleled in the meadow steppe in the second sector, but although overall niche overlap in the first sector was lower than expected. This suggests that the first sector has a high habitat heterogeneity, that facilitates species segregation. This is consistent with the strong correlations that we found between carabid activity density and vegetation indices only at a larger scale (buffers of 1450 m and 1500 m diameters; Tsafack et al. 2020b).

Species that coexist in a certain area are expected to show lower overlap than a randomly assembled set of species (from the same area) if the community is structured by competition because of mutual exclusion. The significant overlap in spatial distribution found for the carabids of the desert and typical steppes using the RA3 algorithm thus suggests that there is little competition (Pianka 1986; Gotelli and Graves 1996), or perhaps strong competition that did not lead to a divergence on the use of space until the present (Gotelli and Graves 1996). As niche overlap was higher than random in the desert and the typical steppes, but not in the meadow steppe, with the RA3 but not with the RA2 algorithm, we believe that the spatial fragmentation of resource distributions causes some degree of niche overlap.

The high average C-scores found in the meadow and typical steppe suggests that the presence of a species is directly affected by the presence of other species, and hence that competition may have some role in defining species assemblages in these ecosystems, whereas the low value observed in the desert suggests, as above, that competition is not involved here.