Assessing biodiversity by remote sensing in mountainous terrain: the potential of LiDAR to predict forest beetle assemblages


*Corresponding author. E-mail:


1. Effective biodiversity management can only be implemented if data are available on assemblage–environment relationships. The level of detail needs to be relevant to the scale of planning and decision making. A number of remote-sensing methods are available, but there are few studies that link information collected at both landscape and local scales. This is particularly true for arthropods even though these organisms are ecologically very important.

2. We assessed the predictive power of habitat variables measured by airborne laser scanning (light detection and ranging; LiDAR) to model the activity, richness and composition of assemblages of forest-dwelling beetles. We compared the results with data acquired using conventional field methods. We sampled beetles with pitfall traps and flight-interception traps at 171 sampling stations along an elevation gradient in a montane forest.

3. We found a high predictive power of LiDAR-derived variables, which captured most of the predictive power of variables measured in ground surveys. In particular, mean body size and species composition of assemblages showed considerable predictability using LiDAR-derived variables. The differences in the predictability of species richness and diversity of assemblages between trap types can be explained by sample size. We expect predictabilities with R2 of up to 0·6 for samples with 250 individuals on average.

4. The statistical response of beetle data and the ecological interpretability of results showed that airborne laser scanning can be used for cost-effective mapping (LiDAR : field survey : beetles 15 : 100 : 260 € ha−1) of biodiversity even in remote mountain areas and in structurally complex habitats, such as forests.

5.Synthesis and applications. The strong relationship between characteristics of beetle assemblages to variables derived by laser scanning provides an opportunity to link data from local ground surveys of hyperdiverse taxa to data collected remotely at the landscape scale. This will enable conservation managers to evaluate habitats, define hotspots or map activity, richness and composition of assemblages at scales relevant for planning and management. In addition to the large area that can be sampled remotely, the grain of the data allows a single tree to be identified, which opens up the possibility of planning management actions at local scales.


Biodiversity is diminishing at an accelerating pace (Chapin et al. 2000), and, therefore, data on biodiversity and spatial distribution are urgently needed for conservation planning (Margules & Pressey 2000). However, the lack of taxonomists and funds precludes particularly the mapping of hyperdiverse taxonomical groups (Early & Thomas 2007; Basset et al. 2008; Vierling et al. 2008). Therefore, management strategies concentrate on charismatic vertebrates and only a few conspicuous arthropods (e.g. Buse, Ranius, & Assmann 2008) and ignore most other species (Hammond 1992). However, the diversity of invertebrates is declining even more rapidly than that of plants and vertebrates (Thomas et al. 2004), even though invertebrates are involved in important ecosystem functions (e.g. pollination and decomposition; Schädler & Brandl 2005; Klein et al. 2008). We clearly need methods to rapidly, effectively and cheaply assess the value of sites for all species, including arthropods, at a level of detail or grain (c. 1·0–10 000 m2) and at an extent (ten to thousands of hectare) relevant for management and conservation (Duelli & Obrist 1998; McCleary & Mowat 2002).

Forests are often hotspots of invertebrate diversity (Stork 1988; Lawton et al. 1998). The compilation of faunal lists and data from canopy projects have increased our knowledge of insect assemblages living on single tree species (Stork 1987; Basset 2001; Brändle & Brandl 2001, 2006). Patterns of insect assemblages in forests, however, are still poorly understood (Stork, Didham, & Adis 1997; Novotny et al. 2007). The three-dimensional structure of the canopy is one of the essential drivers of arthropod diversity in forests (Rinker et al. 2001). Measuring the complex canopy structure during ground surveys requires considerable time (Hyde et al. 2006), and most forest studies have limited sample sizes and consequently limited statistical power (Nadkarni & Cushing 2002).

Numerous studies have demonstrated the potential of remote sensing to bridge the gap between grain and extent. In entomology, remote sensing is used to monitor insect movement (Osborne et al. 1999; Reynolds & Riley 2002) or areas infested by pest species (Solberg et al. 2006; Wulder et al. 2006). High-resolution multi-spectral imagery has been used to predict ground-dwelling ant and beetle assemblages in Australian forests (Lassau et al. 2005b). Despite such promising studies (Lassau & Hochuli 2008), the application of remote sensing for modelling diversity of insects is still underdeveloped. Here, we demonstrate the potential of Light Detection And Ranging (LiDAR) as a remote-sensing technique (Lefsky et al. 2002; Parker, Harding, & Berger 2004) to model assemblage composition, diversity and activity of forest beetles. Although several review articles on the potential of LiDAR in habitat modelling have appeared, the number of applications in real situations is still low, and the studies concentrate on birds (Bradbury et al. 2005; Vierling et al. 2008).

Among forest insects, beetles are the most prominent group (Grove & Stork 2000) with considerable impact on ecosystem functions (e.g. Grove & Stork 2000; Dennis, Aspinall, & Gordon 2002). The forest canopy is an important driver for the assemblages of beetles. It provides the habitat for many species, and it determines the density, height and composition of the herb layer, which is important for feeding (Böhme 2005). LiDAR provides proxy variables of the structural diversity of the forest, including the density of the canopy layer and forest gaps (Lefsky et al. 2002). LiDAR also measures the elevation of a site, which is a proxy for precipitation, temperature and vegetation types (Körner 2007). To examine the potential of variables extracted from LiDAR data sampled from a helicopter for statistical modelling of activity, diversity and composition of beetle assemblages in forests, we selected an area with a broad range of canopy structures. For comparison, we also collected environmental data to characterize habitats from ground surveys. Our analyses have three objectives: (i) to evaluate the predictive power of LiDAR in modelling beetle assemblages in comparison to ground-based measurements, (ii) to evaluate the ecological interpretation of the relationships between single parameters of LiDAR and beetle community characteristics and (iii) to compare the response of the activity of several beetle feeding guilds to LiDAR variables.

Materials and methods

Study area

Our study was conducted in the Bavarian Forest National Park in south-eastern Germany, a mountainous forest area between 650 and 1400 m a.s.l. (Müller et al. 2008b). Owing to a no-take forestry policy (everything was left completely unmanaged) in the core zone of the National Park, wind blows and bark beetle infestations have influenced the canopy structure considerably during the last 25 years (Müller et al. 2008b). Forests form a patchy mosaic of habitats (see Fig. S1) differing in structure in a way that is not systematically related to altitude (see also Lomolino 2001). We sampled habitat characteristics as well as beetle assemblages along four transects (T1–T4) from 5·7 to 8·6 km in autumn 2006 and 2007 (total length of 29·3 km; Fig. 1).

Figure 1.

 Study area and sampled transects T1–T4. The map of Germany shows the location of the National Park. The map of the park shows the topography with low (dark) to high (pale) elevations. The 171 sampling stations are shown as white dots. The inset at the top shows an example of the digital surface model generated by LiDAR, with 0·1 ha circles and 1·0 ha squares.

Sampling of beetles

Along the transects, we selected 171 sampling stations in 2007 (T1: 35, T2: 44, T3:42 and T4: 50), with a minimum distance between two stations of 100 m. This distance is sufficient to ensure negligible spatial autocorrelation between biodiversity samples (Fig. S3; Baker & Barmutta 2006). At each station centre, we installed one flight-interception trap of a type widely used to survey forest beetle communities (Fig. S2; Grove 2000). Each trap was installed 1 m above the ground between two trees; a 1 L sampling jar containing 3% copper vitriol solution for preservation of specimens was placed at the bottom of the trap funnel (for details, see Müller et al. 2008b). We also installed one 0·5-L pitfall trap under each flight-interception trap to sample ground dwellers. Flight-interception traps were set up from May to September, pitfall traps only in May, July and September 2007 and emptied monthly.

Almost complete information is available on the diet, distribution and body size of beetle species in Central Europe (Freude, Harde, & Lohse 1964–1983; Böhme 2005), which enabled us to identify almost all specimens to the species level and to obtain trait data (Table S1). As dependent variables, we used the total number of individuals as well as the number of particular feeding guilds. These are measures of beetle activity at a site (Goßner 2004). Individual feeding guilds were analysed only for guilds with at least 2000 individuals within a trap type. This criterion was met by zoophagous, phytophagous, xylophagous and mycetophagous feeding guilds in flight-interception traps, and zoophagous and phytophagous feeding guilds in pitfall traps. We log10 transformed the number of individuals in each trap for further analyses. We estimated (species) richness according to Gotelli & Colwell (2001) as the residuals of a linear model of log10 (number of species) vs. the log10 (individuals). We calculated the diversity in each trap using the Simpson index as a measure independent from sample size (Lande 1996) and the weighted mean body size (in mm, weighted by relative abundance). To characterize the entire assemblage, we used direct ordinations of a species-abundance matrix with square-root transformation of the number of individuals. For further justification of the choice of dependent variables, see Fig S4.

Habitat variables – LiDAR and ground surveys

We used LiDAR and field measurements of abiotic and biotic data to characterize habitats. Our LiDAR data were collected in May 2007 in full waveform after foliation with an average point density of 25 m−2 from 400 m above-ground, resulting in a footprint size of 25 cm (for details, see Appendix S1 and Müller et al. 2009). One to 11 discrete points were generated per laser shot. With these data, a digital surface model (Fig. 1, inset) and a digital terrain model were calculated. The digital terrain model provided the mean altitude around a trap station. For the digital surface model of the canopy, the raw data points were sorted into a rectangular array of cells (0·25 × 0·25 m2) using only the point with the highest value for further calculations. Based on these data, we calculated the SD of canopy height and the maximum tree height at each sampling station. The SD is an index for vertical variation of the canopy height. This is influenced by the mixture of tall and short trees but is also influenced by tree species combination, with generally higher values in plots with conifers than in plots dominated by beech (cf. Müller et al. 2009). The maximum tree height provides information about the availability of at least one tall and old tree in a plot, which is a surrogate for habitat continuity (Ohlson et al. 1997). We also calculated the penetration rate of the laser echo (in %; number of laser echoes reaching 2 m above-ground divided by the number of laser echoes measured at 50 m above-ground). This is a proxy of microclimatic conditions (e.g. light and temperature) in the lower stratum of forests. All LiDAR variables were calculated for areas of 0·1 ha circles and 1·0 ha squares surrounding the trap stations. Preliminary analyses, however, showed that the variables were correlated across the two scales (r > 0·7); therefore, we used the 1·0 ha data set for the present analyses. Altogether we derived four environmental variables from LiDAR: (i) altitude, (ii) SD of the canopy height, (iii) maximum tree height and (iv) penetration rate.

For comparison, we sampled abiotic and biotic variables during a ground survey (see also Meyer et al. 2001). In 2006, at each site we sampled the vegetation in the herb layer, shrub layer (up to 5 m), tree layer 1 (>5 to 20 m) and tree layer 2 (>20 m) within a circular area 200 m2 around each trap station, and estimated the cover according to Londo (1976). From these raw data, we derived the following five biotic variables: (i) broadleaf tree cover, i.e. the sum of the cover of all individual broadleaf trees in all vegetation layers; (ii) coniferous tree cover, calculated similarly; (iii) number of plant species in the herb layer; (iv) the total cover of the herb layer; and (v) height of the herb layer. We did not include the number of species in the tree layers because the montane forest is dominated by only two to three species. In 2007, we also measured the following abiotic variables (details in Bässler et al. 2008): (i) pH value of the humus layer, (ii) ground moisture on a rank scale of 0–10, (iii) mean annual temperature derived from data loggers along the altitudinal gradient (Bässler et al. 2008) and (iv) dead wood surface (m2) estimated from all pieces of dead wood within a 0·1 ha circle around each trap station, based on a ground survey of dead wood in 2006 (for details, see Table S2).

Statistical methods

We used canonical correlation analysis to investigate the relationships between our environmental data sets (Hotelling 1936; Krzanowksi 2004). We used the adjusted R2 and partial R2 from variance partitioning as implemented in ‘vegan’ (Oksanen et al. 2006) to evaluate the predictive power of LiDAR, and we used the abiotic as well as biotic environmental data sets to define the characteristics of the assemblages of beetles. These analyses extract the unique and joint contribution of each data set for predicting species richness, diversity or community composition (Borcard, Legendre, & Drapeau 1992). We used bootstrapping to estimate the variability of the variance components. Furthermore, we evaluated the percentage of explained variation in the redundancy analyses by randomizing species across plots; this provides a distribution of the expected explained variation when species distributions are independent from the environment.

To study the influence of the variation of individuals on predictive power, we ordered all samples by increasing number of individuals. We calculated the R2, considering all three environmental data sets for the ordered samples starting with 1–50, 2–51 and so on up to the maximum of 652 individuals in flight-interception traps and 260 in pitfall traps. A window size of 50 was a compromise between sample size and variation between windows.

Finally, we tested the dependent variables against the individual LiDAR variables and estimated the parameters. Predictive power and parameter estimates are two different statistical points of view: predictive power could have the same numerical value, but the signs may be opposite. We applied multiple linear regression models to analyse the relationships of assemblage characteristics and the activity of major feeding guilds with the variables derived from LiDAR (Quinn & Keough 2002). Finally, we checked the residuals of models for spatial independence (Quinn & Keough 2002). However, our models showed no spatial autocorrelation of residuals even at very small distances (see Fig. S3), and, therefore, we ignored space in the statistical analyses presented.


Altogether, we collected 50 910 individuals at the 171 sampling stations, representing 782 species (Fig. 2a and b). Only three individuals of Gabrius sp. (Staphylinidae) and one individual of Epurea sp. (Nitidulidae) could not be identified to the species level and were excluded from subsequent analyses. In both types of traps, we found a curvilinear relationship between the number of individuals and the number of species within a trap (Fig. 2c and d), which became linear after log-transformation (Fig. S4). The body sizes of species showed much more variability, and the mean was higher in the pitfall traps than in the flight-interception traps (Fig. 2e). The rank–abundance relationship was steeper for the pitfall traps than for the flight-interception traps (Fig. 2f). In both trap types, zoophagous species were the most species-rich guild (Fig. 2a). Necrophagous beetles, with 19 species, were not species rich, but high numbers of individuals were present, especially in pitfall traps (Fig. 2b). However, most of these individuals belong to five species (Figs S5 and S6). As necrophagous beetles were attracted to small mammals accidentally caught in the traps, we decided to exclude this guild in our analysis (for further justification, see Figs S5 and S6). The final data set comprised 24 287 individuals in flight-interception traps and 12 216 in pitfall traps.

Figure 2.

 Distribution of (a) species and (b) individuals across the feeding guilds in the two types of traps. Relationship between the number of sampled individuals and the number of species within (c) the flight-interception traps and (d) the pitfall traps. (e) Distribution of mean body size and (f) rank–abundance plots of the individuals sampled in the two types of traps.

Canonical correlation analysis between the environmental variables showed clear correlations between the first roots (Fig. 3). Furthermore, the shared information represents a substantial part of the total variance within each data set. This holds true even after excluding the integrative variable altitude from the LiDAR data set (see Appendix S2). Apparently, all three data sets extracted similar environmental gradient patterns from the forest sites.

Figure 3.

 Correlations between the first roots of a canonical correlation analysis extracted to maximize the correlation between the data sets of (a) abiotic and LiDAR-derived variables and (b) biotic and LiDAR-derived variables. The insets show the decrease of correlation coefficients for the first four roots of each pair of data sets.

In general, R2 values and partial R2 values were lower for pitfall traps than for flight-interception traps for each of the three independent environmental data sets used to predict activity, richness and diversity of beetles (Table 1). For the assemblages sampled with the flight-interception traps, all variables explained 15–44% of the variance. LiDAR variables contributed >60% to the total predictive power (Table 1). One notable exception to the observation that the predictability of assemblage characteristics is better for samples of flight-interception traps than for pitfall traps was the weighted mean body size. It showed the opposite pattern with higher R2 for pitfall trap (27·1%) than for flight-interception trap (14·7%). LiDAR variables contributed almost 90% to the explained variance in pitfall traps. For species composition sampled with the pitfall traps, the explained variance was almost 22% (95% confidence band: 18·5–25·2), which is much higher than expected by chance (Fig. 4). LiDAR variables contributed 78% to the explained variance. For the assemblage collected with flight-interception traps, the explained variance was around 25% (95% confidence band: 20·4–29·8), and LiDAR contributed 82% to the explained variance (Fig. 4).

Table 1.   Partitioning of the explained variance for the three data sets of predictors: LiDAR, biotic and abiotic
 Pitfall trapsFlight-interception traps
Total R2Total 95% CILiDAR R2Biotic R2Abiotic R2Total R2Total 95% CILiDAR R2Biotic R2Abiotic R2
  1. For the total explained variance, the 95% confidence intervals (CI) based on bootstraps are given. Note, that ‘varpart’-functions frequently give negative estimates of variation. We give the adjusted R2, which adjusts the number of explanatory terms in a model. Adjusted R2 can be negative for any fraction, while unadjusted R2 of testable fractions will always be non-negative (Oksanen et al. 2006). Such negative values indicate variability in the data large enough to produce a negative estimate, even though the true value is zero or positive. LiDAR, light detection and ranging.

Individuals8·9−0·3 to 19·917·7−8·054·243·830·5–55·799·875·854·8
Richness3·0−8·7 to 10·799·599·05·926·413·6–39·489·130·148·4
Diversity (Simpson)3·7−6·5 to 19·899·054·3−25·723·814·3–34·094·860·566·5
Body size27·1 14·3–39·687·660·619·014·73·9–27·266·831·014·0
Figure 4.

 Variance partitioning of explained variance using three different environmental data sets: variables derived from LiDAR data, biotic and abiotic variables derived from ground surveys, and data from beetle assemblages captured with 171 pitfall traps and 171 flight-interception traps after excluding necrophagous species. Each histogram shows the distribution and the mean (bar) of explained variance components after 10 000 bootstraps. The open histogram shows the result of random distribution of species among plots.

Using a window of 50 samples, we tested how sample sizes influenced predictability (Fig. 5). The broad pattern of an increase of R2 for richness, body size and community with increasing number of individuals was similar for pitfall traps and flight-interception traps, and therefore the low predictability of certain characteristics for pitfall traps is in part due to the lower number of individuals sampled in these traps. Only the Simpson index was, as expected, independent from sample size (see Lande 1996). For traps with a mean of more than 250 individuals, R2 for species richness and body size reached values of around 40–50%. Note also that despite the low predictability of species richness from pitfalls traps, for a given sample size, the predictability of mean body size and assemblage composition is much better for pitfall traps than for flight-interception traps.

Figure 5.

 Adjusted R2 values for sliding windows of 50 traps moved across the samples sorted from small to large sample sizes and plotted against the mean number of individuals in the 50 traps. R2 was calculated for samples 1–50, 2–51 and so on. Note that the calculation of the adjusted R2 by variance partitioning as implemented in ‘vegan’ can result in negative values. Such negative values indicate variability in the data large enough to produce a negative estimate, even though the true value is zero or positive.

Our linear models showed several significant responses to the LiDAR-derived parameters (Table 2). Species richness in pitfall traps increased with altitude. Individuals, species richness and diversity in flight-interception traps increased, but body size decreased with an increase in the SD. The number of individuals and body size in pitfall traps decreased, but the number of individuals in flight-interception traps increased with an increase in canopy openness. The analysis of the activity of major feeding guilds showed a general decrease of phytophagous and mycetophagous species with an increase in altitude in both trap types. The activity of almost all feeding guilds increased with the increase in the laser penetration rate (Table 2).

Table 2.   Results of multiple linear models using different characteristics for beetle assemblages and diversity
 TrapAltitudeSD of vegetation heightMaximum tree heightPenetration ratio
  1. To allow for direct comparisons of the estimators, the predictors were standardized to a mean of 0 and a variance of 1. For the estimated values, the SE are given.

IndividualsFlight trap  0·21 ± 0·070·005−0·28 ± 0·100·0100·38 ± 0·06<0·001
Pitfall trap0·026 ± 0·090·008      
RichnessFlight trap−0·08 ± 0·030·0020·12 ± 0·03<0·001    
Pitfall trap  0·10 ± 0·030·006    
Diversity (Simpson)Flight trap−0·14 ± 0·04<0·0010·16 ± 0·040·001    
Pitfall trap0·13 ± 0·040·011      
Body sizeFlight trap0·47 ± 0·150·003−0·60 ± 0·18<0·0010·68 ± 0·250·008  
Pitfall trap0·91 ± 0·330·008−0·81 ± 0·390·0361·78 ± 0·550·002−0·93 ± 0·350·009
Individuals zoophagousFlight trap  0·32 ± 0·08<0·001−0·30 ± 0·110·010·30 ± 0·07<0·001
Pitfall trap        
Individuals phytophagousFlight trap      0·37 ± 0·11<0·001
Pitfall trap−0·21 ± 0·090·026    0·26 ± 0·100·019
Individuals mycetophagousFlight trap−0·65 ± 0·09<0·0010·38 ± 0·10<0·001−0·51 ± 0·14<0·001  
Individuals xylophagousFlight trap      0·35 ± 0·11<0·001


Our results showed that LiDAR provides useful variables with which to model diversity–habitat relationships with astonishingly few parameters. The main advantage of LiDAR and related methods is that they allow sampling of habitat characteristics with a high resolution at large spatial scales, providing statistically well-behaved data (see skewness and kurtosis in Table S2). Furthermore, remote-sensing techniques such as LiDAR are cost-effective. During our study, we invested c. 15 € ha−1 for the LiDAR data, which included the flight and sampling from the helicopter (9 € ha−1) and processing of the basic data (6 € ha−1). The costs of collecting the ground-based habitat parameters were c. 100 € ha−1, while sampling, sorting and determination of the beetles cost another 260 € per plot (equipment 40 € ha−1, emptying 60 € ha−1, sorting in laboratory 60 € ha−1 and identification costs 100 € ha−1). Furthermore, the costs per ha for LiDAR decreases with the area covered during a campaign because of certain fixed costs. Therefore, the advantage of LiDAR compared with ground surveys increases with the extent of a study.

The high costs of collecting hyperdiverse taxa data limit the number of traps or plots that can be sampled during a survey. Published studies in tropical, temperate or boreal forests used between 80 and 240 flight-interception traps and up to 480 pitfall traps (Grove 2002; Martikainen & Kouki 2003; Müller, Bußler, & Kneib 2008a). Sometimes 5–20 traps were set up at one sampling location. Assuming that 240 flight-interception traps are the maximum number of traps that can be handled, a survey across the Bavarian Forest National Park (24 000 ha) would entail 1 trap per 100 ha. This grain of sampling is much too coarse to map relevant spatial patterns of species richness or community composition relevant for the management of the park. Planning is usually carried out at scales of 5–50 ha, while actual management activities, such as felling operations, occur at scales between 0·3 and 1·0 ha (Meyer et al. 2001; Burschel & Huss 2003). Clearly, surveys of hyperdiverse insect taxa are too expensive for mapping with the grain and extent needed for management planning. Remote-sensing techniques such as LiDAR provide the means to at least estimate activity, richness or composition of assemblages with the appropriate grain and extent. Nevertheless, ground surveys of biodiversity provide the baseline to derive the parameters needed to model assemblages and their characteristics. Therefore, remote sensing is not an alternative to field surveys, but is rather a valuable technique that allows point data to be transferred to a broader spatial scale.

As LiDAR variables were generated according to the technical possibilities, these variables have no direct and obvious meaning for beetle assemblages. However, phenomenological models are of enormous heuristic value, particularly at larger spatial scales (Vierling et al. 2008). Available remote-sensing methods and in particular LiDAR offer a broad array of techniques (Lefsky et al. 2002; Turner et al. 2003; Bradbury et al. 2005; Hinsley et al. 2006; Goetz et al. 2007). To discuss the value of LiDAR-derived habitat variables for predicting beetle assemblages in more detail, we need to evaluate three important issues: (i) the sampling methods, (ii) the predictive power of our models and (iii) the findings in the context of autecology.

The two trap systems used in our study are standard methods for sampling beetle assemblages (Baker & Barmutta 2006; Grove 2000), but in most studies, more traps were used per site (Martikainen, Kouki, & Heikkala 2006). Resource constraints forced us to use only one trap per site, and the sample size of pitfall traps was often too low for the analysis of certain community characteristics (Fig. 5). We therefore suggest that for calibrating LiDAR variables for monitoring insect communities in temperate forests mean sample sizes of 250 individuals should be used.

Rather unexpectedly, the predictability of mean body size and community composition was not influenced by sample size (individuals) to any degree. Furthermore, in these two cases, the assemblages sampled with pitfall traps had a better predictability than the assemblages sampled with flight-interception traps. These findings have two implications. First, for applied investigations, body size and ordinations may provide reliable patterns of assemblages even with small samples sizes (Basset et al. 2008). Secondly, characteristics of assemblages sampled with pitfall traps are more tightly correlated to environmental variables and are therefore more structured than samples of flight-interception traps. The predictability of community characteristics such as richness depends also on ‘equilibrium’ conditions. If the assemblages of beetles on the ground are not in equilibrium with the local conditions as well as with the regional pool, actual environmental conditions may have nothing to do with actual species composition.

The predictive power of most of our models varied between 10 and 40%. With mean sample sizes of 250 individuals, the predictive power may be almost 60% (Fig. 5). Other studies using R2 to quantify the predictability of certain characteristics of insect assemblages have found similar values. For ants and beetles in Australian forests, the predictive power was between 30% and 50% for richness and abundance (Lassau et al. 2005a; Lassau & Hochuli 2008; Lassau et al. 2005b). Note also that the percentage of explained variation dropped to zero when we randomized species across plots (Fig. 4). Despite the sometimes low sample size, sufficient variation in the composition of assemblages was explained by our environmental variables.

Our findings can be placed in the context of autecology using the extensive literature. Based on numerous studies of disturbances in forests, we expected a decrease in body size with the laser penetration ratio, and larger species in closed forests (Peters 1983; Nee & Lawton 1996; Siemann, Tilman, & Haarstad 1996; Weller & Ganzhorn 2004). In temperate and boreal biomes, several xylophagous species need open stands with sunlight and dead wood for feeding and mating (Bouget & Duelli 2004). Therefore, we expected an increase in activity of xylophagous species with an increase of the laser penetration ratio. Last but not least, we expected the activity of mycetophagous and phytophagous species to decrease with altitude because of a reduced availability of hosts. These expectations matched the multiple linear regression models (Table 2), which shows that LiDAR variables are not only suitable for phenomenological models, but can also provide ecological information.


Insects and especially beetles with a body size of < 2 cm act on small spatial scales in most cases, and rugosity may be important for foraging (Kaspari & Weiser 1999). Nevertheless, the canopy structure has important influences on certain habitat characteristics. Therefore, characteristics of species assemblages in forests depend on the structure of the canopy. The high proportion of explained variance by LiDAR-derived variables compared with ground measurements shows the high potential of data from remote sensing for modelling biodiversity in forests at broad spatial scales, with costs around 5–10% of ground survey costs (Lefsky et al. 2002). However, reliable biodiversity data are needed to calibrate and validate the statistical models. After such a modelling exercise, LiDAR provides simple but ecologically meaningful variables for a rapid extrapolation of activity, richness and composition of assemblages across large areas. This enables conservationists to evaluate habitat over large areas and to define diversity hotspots as well as to monitor environmental changes for regional and even countrywide management plans.


The study was supported by the Bavarian State Ministry of the Environment, Public Health and Consumer Protection. We are grateful to Sarah König, Ute Augenstein, Thomas Wagner, Heinz Bußler and Boris Büche for help in the field, sorting the material and identification. We thank Karen A. Brune for linguistic revision of the manuscript and Marco Heurich for providing information about the LiDAR data. Kerri Vierling, Robert Ricklefs, Martin Goßner, Marc Cadotte, Jos Barlow, Raphael Didham and an anonymous reviewer gave valuable comments on an earlier version of the manuscript.