Concepts of scale

We covered two concepts of spatial scale: (i) spatial extent or size of the study area and (ii) spatial resolution or wavelength of variability of the variables [25]. Ours is a large scale study from both conceptual points of view, i.e. large extent (Germany, area approximately 357,000 km²) and large (coarse) resolution (approximately 18 km (10 arcminutes) based on the variability of the BIs). We use the terms “large scale” and “scale” for both concepts. When we refer to the scale of Germany, we mean extent; but when we refer to the scale of the relation, e.g. variability and pattern, we mean resolution. Moreover, when we refer to the scale of relationship, we refer to both concepts.

Data and processing

Aquatic insect data.

We used governmental biomonitoring data on macroinvertebrates from 4,752 stream sites (i.e. stream reaches with a maximum of 20 meters length) sampled during 2006–2007 that covered the whole spatial extent of Germany (Fig 1). Data coverage in terms of number of sites was lower (approximately 1% of the total number of sites) in the southeast and northwest (in proximity to the North and Baltic sea) than in the other regions. The biomonitoring data was produced following a standardized protocol, where a pooled sample was taken from all major habitat types in a stream site [32]. Samples were collected from the middle and small sized streams in each ecoregion of Germany. For more details on biological sampling, subsampling and sorting see AQEM CONSORTIUM, Rolauffs et al. and Biss et al. [32–34]. Given the semi-quantitative nature of macroinvertebrate data, the data were originally reported as abundance classes where the classes approximated log-transformed abundance data following the classes of the saprobic index (for the description of the abundance classes see AQEM CONSORTIUM and Rolauffs et al. [32,33]). Thus, the abundance classes varied on a scale of zero to seven; zero meaning no abundance, i.e. absence and seven the highest abundance [33]. Overall, abundance classes for 2,099 stream macroinvertebrates were available. Abundance data were preferred over presence-absence data because more powerful hypothesis tests are available for abundance data in spatial pattern analysis of assemblage compositions and studies of turnover rates [35].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Distribution of the 4,752 stream sites sampled by the German national bio-monitoring program during 2006–2007.

Spatial reference system is WGS 1984.

https://doi.org/10.1371/journal.pone.0130025.g001

We examined the homogeneity of taxonomic resolution and found that the organisms were reported at different taxonomic levels (from class to species). We took the subset of 1,901 organisms (91%) that were reported at genus (660) and species (1,241) levels. From this subset, we selected the aquatic insect orders, namely Diptera (True flies), Ephemeroptera (Mayflies), Odonata (Dragonflies and Damselflies), Plecoptera (Stoneflies) and Trichoptera (Caddisflies) (Table 1). Aerial dispersers are more suitable for large scale analyses than exclusive aquatic dispersers, because they can disperse through the landscape and are not limited to the stream network [24,29]. Moreover, these orders were also used in previous large scale studies to indicate climate vulnerability [4,18–21] and information for the selected traits were available for all organisms in these orders. This resulted in 782 species [and 395 genera] that comprised 384 (216) dipterans, 101 (39) ephemeropterans, 42 (33) odonates, 52 (36) plecopterans and 203 (71) trichopterans. Next, in case that a taxon was identified at genus level for more than 1% of stream sites, we converted all species belonging to this genus to genus level. This was the case for 73% of the species and was done to avoid artifacts from potential spatial pattern linked to the taxonomic resolutions, for instance mainly genus level identification in regions with low data coverage.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Explained variances and spatial autocorrelation in the traits of each order and full data by the bioclimatic indices.

https://doi.org/10.1371/journal.pone.0130025.t001

Calculation of assemblage trait composition.

The biomonitoring data were linked to the trait data using the codes of “The development and testing of an integrated assessment system for the ecological quality of streams and rivers throughout Europe using benthic macroinvertebrates” (AQEM) project to avoid discrepancies in naming conventions [38]. Each of the species was assigned with the traits using their corresponding percentage membership states that were multiplied with the absolute abundance classes of the species for a site to compute relative abundance classes for the traits (Fig 2). To assign trait information to genera, we calculated the median of the related species level information following Schmidt-Kloiber and Nijboer [39] except for maximal body size where genus level information were available in the Tachet database for all genera. Subsequently, the assemblage trait composition, i.e. abundance weighted trait (AWT) was calculated following the procedure described in [40] and as outlined in Fig 2. The AWT was calculated as a measure of assemblage trait composition because it is the most frequently used metric to assess the relationship between assemblage traits and environmental variables [41,42]. Note that we use the term assemblage trait composition to improve readability, although the assemblage data was restricted to aquatic insects, and hence does not represent the complete macroinvertebrate assemblage. The calculation resulted in annual averaged abundance-weighted traits (AWT) for each insect order (Fig A in S1 File) and for the combined (full) data (Figs 3 and 4) for each stream site. The calculation was omitted for the dispersal capacity of ephemeropterans and plecopterans because the grouping feature consisted of only one trait (low dispersal). However, they were included in the calculation for the full data.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Conversion steps from abundance classes of the selected aquatic insects to trait compositional (annual averaged abundance weighted traits) data.

https://doi.org/10.1371/journal.pone.0130025.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Annual averaged abundance weighted traits across 4,752 stream sites in Germany for the biological traits of the full data.

The figure sub-captions and panel captions indicate names of grouping features and traits, respectively. The gray dots indicate the zero abundance, i.e. trait absence.

https://doi.org/10.1371/journal.pone.0130025.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Annual averaged abundance weighted traits across 4,752 stream sites in Germany for the ecological traits of the full data.

The figure sub-captions and panel captions indicate names of grouping features and traits, respectively. The gray dots indicate the zero abundance, i.e. trait absence.

https://doi.org/10.1371/journal.pone.0130025.g004

Bioclimatic indices and altitude data.

The 35 bioclimatic indices (BI, denoted as “Bio01” to “Bio35”, see Table 2 for details) for temperature, precipitation, radiation and moisture were collected from the global climatologies for bioclimatic modeling (CliMond) database (www.climond.org) [30]. A previous study showed that these BIs can provide an approximation of climate impact on assemblage patterns, despite the omission of confounding endogenous factors such as biotic interactions, evolutionary change and dispersal potential [43]. The scale of variability was determined by the spatial resolution of the BI raster, which is 10 arc-minutes (approximately 18 km). The digital elevation model (giving altitude over mean sea level) for Germany was collected from the ASTER GDEM on one arc-second (approximately 30 m) resolution [44]. The altitude raster was resampled to the resolution of the BI rasters to extract altitude information for each BI raster cell.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Explained variances and spatial autocorrelation by the individual bioclimatic indices in the traits and orders with the highest climate response and potential for changing distribution pattern.

https://doi.org/10.1371/journal.pone.0130025.t002

Although no clear gradient in the BIs was found for Germany, lower temperatures and higher precipitation were mostly observed in the southern regions, whereas higher temperatures and lower precipitation were mostly observed in the northern regions (Fig B in S1 File). For example, observed ranges of the annual mean temperature (Bio01) and annual precipitation (Bio12) are 2 to 5°C and 7 to 12°C, and 1200 to 1600 mm and 600 to 800 mm in the southern and northern regions, respectively. The northern and southern regions are portrayed as flat (zero to 250 m above sea level) and mountainous (600 to 1800 m above sea level), respectively (Fig C in S1 File). The BIs showed significant (p < 0.001) spatial autocorrelation, i.e. average Moran's I = 0.28 (Table B in S2 File). Significant spatial gradients were also observed for the BIs (Table B in S2 File). Longitudinal (North—South) and altitudinal (high—low) gradients were both stronger than the latitudinal gradient (East—West) for most of the BIs. Longitude and altitude of the BI cells showed significantly high correlation (r = -0.8, p < 0.001) with each other and thus indicates that the dominant climatic variation along the North—South (longitudinal) gradient on the scale of Germany (also observed in Fig B in S1 File) may be attributed to topography, i.e. altitude (low—high).

Analyses of the spatial relationship between traits and climate

The spatial relationship between the aggregated AWT per BI cell and the BIs was analyzed in four steps (Fig F in S1 File). First, we checked for spatial autocorrelations in the AWT (Table C in S2 File). The spatial autocorrelation was analyzed using Global Moran's I (see Bonada et al. [24] for details on computation), where great circle distances among BI cell pairs were given as weights based on the simplified assumption that the selected species disperse symmetrically during their terrestrial life stage [29]. The computations of spatial autocorrelations were done using the R package “ape” [51].

Second, zero-or-one inflated beta regression models were fitted with the AWT as response and 35 principal component scores of the BIs as explanatory variables [52]. This was done to identify the traits and insect orders with the highest climate response. We used zero-or-one inflated beta regression because the response variables were proportional data and included many zeros and ones [53]. The models were fitted for the AWT of each order and the full data and the adjusted R²s were calculated to identify the explained variance by the BIs. The zero-or-one-inflated beta regression model fitting was done using the R package “gamlss” [54].

In the third step, we checked for spatial autocorrelation in the residuals of the trait-climate models using Moran's I as outlined above. The Moran's I values for the residuals of the trait-climate models were subtracted from the complete Moran's I values for the AWT (computed at the first step). Thus, the percentage of trait spatial autocorrelation that is associated with the BIs was identified. This was done to identify the traits and orders that show the highest potential for changing their distributional pattern, i.e. redistribution under future climate change.

In a final step, the zero-or-one inflated beta regression models were re-fitted with the AWT for the previously identified traits and orders with the highest climate response and potential for redistribution as response variables and 35 BIs (original values) separately as explanatory variables. The BIs with the highest explanatory power in terms of R² were identified for the traits and insect orders with the highest climate response. To identify the BIs explaining the highest amount of spatial autocorrelation in the traits and in the insect orders with the highest potential for redistribution, we computed the Moran's I in the residuals of the trait-individual BI models and subtracted them from the complete Moran's I computed at the first step.

Which of the climate-associated traits and organism groups show the highest response to climate and highest potential for changing distribution pattern under future climate change?

We quantified the amount of large scale variability and spatial autocorrelation in the assumed climate-associated traits from six grouping features and five aquatic insect orders of the freshwater assemblages that is explained by 35 global BIs. The BIs explained 19% of the large scale variability in the AWT of the full data on average (Table 1). Traits of the temperature preference grouping feature were the most responsive (32% on average) to the BIs, and the insects with very cold temperature preference (50%) showed the highest response. Among the insect orders, Ephemeroptera and Plecoptera (16%) showed the highest response to the BIs on average, and the ephemeropterans with very cold temperature preference (33%) showed the highest response in particular (Table 1).

The highest response of the traits of the temperature preference grouping feature, particularly of the very cold and cold preference may be due to traits of temperature preference grouping feature being the product of several underlying climate-associated biological traits [19,55,56]. For example, cold temperature preference of the selected aquatic insects in our study was significantly related to low dispersal capacity, large body size (>4 cm), low reproductive capacity (semivoltine) and resistance to drought (egg diapause) (Table D in S2 File), and together they explained 55% of the variability in cold temperature preference. Likewise, warm temperature preference of the insects was related to high dispersal capacity, small body size (≤0.5 cm), high reproductive capacity (multivoltine) and resistance to drought (adult diapause) (Table D in S2 File), and together they explained 48% of the variability in warm temperature preference. These findings are in agreement with other studies on the association of traits with climate change. For example, insects with low dispersal are often characterized by a restricted temperature (cold) niche and hence are more affected by change in temperature regimes, e.g. contractions of alpine regions than the insects with high dispersal ability [18–20,57]. Large-bodied insects generally lack efficient respiration and thus have high ectotherm oxygen demand and hence typically inhabit streams with high oxygen supply, i.e. cold water streams [15,58,59]. Hence, we argue that the highest response of the traits of temperature preference grouping feature to the BIs in our study rather follows from the response of several underlying climate-biological traits relationships. Thus, we envisage an adverse effect of global warming on the insects inhabiting cold water streams in Germany because their biological and ecological niche will be contracted. This prediction is in line with Poff et al. [5], where temperature has been shown to be mostly accountable for the differences in the sensitivity of stream macroinvertebrate traits across geographic space and also with Lawrence et al. and Stamp et al. [15,56] where major declines in macroinvertebrates that inhabit cold water streams were reported as a result of climate change.

The differences in the response of insect orders observed in our study are related to their biological and ecological traits (Tables A and D in S2 File) [4,18–20,58]. Although European ephemeropterans were found to be generally tolerant to climate change [4], we observed the highest BI response in the German ephemeropterans with very cold temperature preference (Table 1). This indicates that ephemeropterans inhabiting very cold water streams in Germany are also vulnerable to climate change because of shrinking ecological niche [60]. Plecopterans showed equally high response as ephemeropterans because they show high membership state for the very cold and cold preference traits, which showed the highest response to the BIs (Table 1 and Table A in S2 File). Generally, plecopterans have a very narrow environmental tolerance with nymphs living mainly in cold and well-oxygenated running water and adults showing low flight ability [18,59]. Hence, plecopterans have never transitioned to thermally variable lentic water and are thus vulnerable to increasing temperature and severe drought episodes [58]. Thus, we also anticipate an adverse effect of climate change on plecopterans in Germany. Overall, our results indicate that insects with traits such as preference for cold water (due to several underlying traits), and from certain orders, i.e. Ephemeroptera and Plecoptera may indeed be more vulnerable to climate change than others (Table 1). Thus, we suggest that future studies on the vulnerability of macroinvertebrate assemblage traits to climate change should particularly focus on traits and orders exhibiting the strongest signal to climate.

Regarding the potential for changing distribution pattern, i.e. redistribution, on average, 59% of the spatial autocorrelation in the AWT of the full data was associated with the BIs (Table 1). The BIs explained the highest spatial autocorrelation in the temperature preference (81%), particularly in the insects with cold temperature preference (91%) (Table 1). More than 50% of the spatial autocorrelation for the majority (62%) of the traits in the insect orders was associated with the BIs. The BIs explained the highest amount of spatial autocorrelation for the insect order Ephemeroptera (59%) in general, and for the Trichoptera with moderate temperature preference (97%). The amount of large scale variability explained by the BIs (described above) in insect traits and orders showed positive significant correlation (r = 0.5, p < 0.001) with the amount of explained spatial autocorrelation. This indicates that the traits and orders showing higher response to the BIs also exhibit a higher potential for changing spatial distribution pattern under changing BIs and vice-versa. Overall, the spatial distribution pattern, i.e. patchiness in the aquatic insects on large scales mostly originate from their high response to spatially autocorrelated climate that is line with Bonada et al. and Domisch et al. [24,26].

The highest potential for redistribution in the traits of temperature preference grouping feature and insect order Ephemeroptera, and trichopterans preferring moderate temperature also presumably relates to their strong covariation with underlying climate-associated biological and ecological traits as discussed above (Tables A and D in S2 File). For example, trichopterans showed high membership state for the underlying biological traits of the moderate temperature preference, i.e. small body size (< 0.5 cm) and high drought resistance (adult diapause) (Tables A and D in S2 File), and hence moderate temperature preferring trichopterans showed the highest potential for redistribution. The redistribution of traits and orders may occur through local extinction of vulnerable insects and thus range contraction [19], or by expansion of the range of tolerant macroinvertebrates in response to climate change [61,62]. Moreover, given that there is a strong association of the spatial distribution pattern of AWT of the insect orders individually (Fig A in S1 File) and of the full data (Figs 3 and 4) with the longitudinal gradient (which is coherent with the observed longitudinal spatial distribution pattern in the climate sensitive European stream macroinvertebrates [4,19,20]), and the BIs also showed a major longitudinal gradient with high correlation to altitude (Fig B in S1 File and Table B in S2 File), the redistribution may occur along the longitudinal (altitudinal) gradient. For example, a higher proportion of insects (0.4) and ephemeropterans (0.3) with cold temperature preference were observed in the cooler southern mountainous regions than in the warmer flat North of Germany (Fig 4 and Fig A in S1 File) that may shrink their distribution range. By contrast, trichopterans with moderate temperature preference that predominantly (0.5) occur in the warmer flat northern regions than in the cooler South may extend their range from North to South because more streams will be suitable for their habitat due to increasing temperature. A similar phenomenon was observed in Hering et al. [19] where most of the European trichopterans were suggested to benefit from increasing stream temperature (78%) and decreasing current (77%). Overall, climate change may alter the trait distribution pattern especially with respect to temperature preference and for the insect order Ephemeroptera, Plecoptera, and for trichopterans with moderate temperature preference in Germany, though adaptations may occur and ameliorate the ecological effects.

The explained variability and spatial autocorrelation for the traits and orders by the BIs in our study are similar (with a few exceptions) to previous studies using aerial and exclusive aquatic dispersers on comparable spatial scales [5,24]. A study dealing with the Mediterranean basin found that climate and environmental variables together explained < 19% variability for the same insect orders (except Diptera) [24]. Moreover, a lower percentage (< 30%) of spatial autocorrelation was associated with climate and other environmental variables than in our study, and in many cases significant spatial autocorrelation remained in the residuals. This discrepancy may be explained by the fact that the study considered only two climate variables (average precipitation and temperature) whereas we considered 35 BIs. The 35 BIs used in our study better captured the climate gradient in Germany and consequently are associated with higher variability and spatial autocorrelation in the AWTM. The use of different biological endpoints, i.e. taxonomic richness in [24] and trait abundance in our study may also explain this discrepancy. In another study on the catchment scale, climate and hydrological variables together explained a similar (19%) trait variability [5] although this study was conducted on a largely different set of traits of macroinvertebrates. Overall, the differences between the studies presumably relate to the traits, organism groups and the number (dimension) of climate variables used as input in models [27].

The inclusion of other environmental drivers such as geology and stream size may decrease the amount of trait variability and spatial autocorrelation that can be attributed to the BIs, especially if drivers exhibit collinearity with the BIs. Nevertheless, other environmental drivers explained much lower taxonomic and trait variation than climate in previous studies [5,24]. Moreover, in our study, the BIs explained more than half of the spatial autocorrelation for the majority of traits, and no statistically significant (all p ≥ 0.08) spatial autocorrelation was observed in the residuals of the trait-climate models (Table 1). This indicates that the remaining trait variability and spatial autocorrelation that can be explained by other environmental drivers are either statistically insignificant or have already been captured by climate, and thus these drivers are of lower importance for the traits under scrutiny [27].

The results may bear some uncertainty regarding the northwestern and southeastern regions of Germany, which were represented by a relatively lower number of stream sites and in turn a lower coverage of BI raster cells than other regions (Fig 1 and Fig D in S2 File). However, previous studies on comparable spatial scales successfully captured macroinvertebrate trait and taxonomic variabilities and their relationships with climate and other environmental drivers, despite relying on less stream sites (lower density) [5,12,24]. Thus, we suggest that our results are sufficiently robust on the scale of Germany, though more stream sites may be required for smaller scale studies in some regions.

Which of the climatic aspects show the strongest relationship with the traits and organism groups showing the highest response and potential for redistribution?

The explained variance and spatial autocorrelation in the most responsive traits and orders by individual BIs was on average 50% lower than by the combined BIs (Table 2). The BIs precipitation of the driest week (18%) and radiation seasonality (17%) exhibited the strongest relationship with insects preferring very cold temperature (Table 2). Precipitation and moisture indices, i.e. annual moisture index and precipitation of the driest week (both 14%), and moisture seasonality, moisture of the wettest and driest quarter (all 13%) explained the highest variance in the very cold preferring ephemeropterans. The radiation seasonality (46%), and radiation (65%) and mean temperature (64%) of the driest quarter explained the highest amount of spatial autocorrelation in the cold temperature preferring insects and moderate temperature preferring trichopterans, respectively (Table 2). Overall, these results suggest that aquatic insects in Germany may mainly be affected in response to potential changes in seasonal radiation and moisture.

In the coming decades, the winter and summer temperatures are highly likely to increase, with the strongest increase predicted for the South of Germany [60]. Moreover, winter precipitation has been predicted to increase with a larger increase in the North. By contrast, summer precipitation has been predicted to decrease in Germany with the strongest decrease in the South [60]. Thus, we anticipate an increase in winter radiation and decrease in summer moisture for the South of Germany where the majority of very cold and cold temperature preferring insects occur (Fig 4), particularly the very cold and cold temperature preferring ephemeropterans and plecopterans (Fig A in S1 File). Thus the increasing winter radiation and decreasing summer moisture may drive climate change effects on insects in general and ephemeropterans and plecopterans in particular that prefer cold water streams in Germany, and may eventually shrink their distribution range. These findings are in line with [13,15], where cold preferring stream macroinvertebrates were shown to be the most adversely affected by increasing winter temperature and decreasing summer precipitation. However, insects may also adapt to increasing temperature and decreasing precipitation [7,8]. For example, adaptations such as decreasing body size [61] and color lightening of adults [62] have been observed in insects. Trichopterans with moderate temperature living in the flat North of Germany (Fig 4 and Fig A in S1 File) may benefit from increasing radiation and recolonize upstream [19], and thus extend their distribution range from the North to the South. Overall, we anticipate a substantial change in the aquatic insect distribution pattern along the longitudinal gradient in Germany because of increasing seasonal radiation and decreasing moisture, especially in ephemeropterans and plecopterans with very cold and cold temperature preference and trichopterans with moderate temperature preference.