Introduction

Information on organismal traits has the potential to identify individuals, populations or species that are vulnerable to environmental changes such as climatic warming and extreme events such as drought (Poff et al., 2010; Tierno de Figueroa et al., 2010; Sandin et al., 2014; Hershkovitz et al., 2015). Trait analysis can also provide insight into mechanisms of responses to environmental change and the consequences of such responses (Poff, 1997; van den Brink et al., 2011; Verberk et al., 2013). Furthermore, consideration of traits facilitates generality and global comparisons because biogeographic barriers have less influence on the trait composition of assemblages than on taxonomic composition (Statzner et al., 2001).

In the aquatic environment, traits expressing preferences and tolerances for environmental factors such as current velocity and temperature may be particularly useful in predicting responses to climatic variation (Stamp et al., 2010; Rosset & Oertli, 2011). However, values of such traits are often unavailable for many taxa, and current trait databases often assign trait values to taxa subjectively according to expert opinion. Moreover, traits are often treated as ordinal or nominal variables with few categories, even if they are really continuous variables. For example, the widely used trait listing by Poff et al. (2006) assigns rheophily and thermal preference to just three categories each.

Ideally, environmental preferences and tolerances would be deduced from experiments under controlled conditions. However, such experiments are technically demanding and time consuming. In the absence of experimental data, information on field distributions in relation to environmental gradients may be used to estimate preferences and tolerances (e.g. Brunke et al., 2001; Carlisle et al., 2007; Rosset & Oertli, 2011; Keck et al., 2016), and may show agreement with laboratory results (Magnuson et al., 1979; Haidekker & Hering, 2008; Horrigan et al., 2007). However, it is also possible that derivation from field data is misleading because the environmental gradient of interest is confounded with other gradients. In addition, species may not occur at their environmental preferences because of factors such as seasonal environmental changes, dispersal constraints, biotic interactions, and trade-offs in requirements for multiple environmental variables.

In this study, I set out to assess the utility of environmental preferences estimated from field data. I derived preferences of aquatic invertebrate taxa for dissolved oxygen concentration, current velocity and temperature, using an extensive field dataset from biological monitoring of streams in the state of New South Wales and the Australian Capital Territory in south-eastern Australia. I then tested the derived values by examining relationships between these traits and both long-term climate and short-term climatic changes between periods of drought and flooding in Australia’s Murray–Darling Basin. I also examined the relationships between estimated environmental preferences and respiratory and dietary traits in order to investigate likely mechanisms underlying the preferences and potential ecological consequences of shifts in the composition of invertebrate assemblages.

Materials and methods

Study areas

The State of New South Wales (NSW: 800,642 km2) and the Australian Capital Territory (ACT: 2358 km2), which is contained within NSW, together extend over 9° of latitude and are climatically and topographically varied. The western two-thirds of NSW are mostly of low relief and arid or semiarid, while the eastern third is mostly humid and hilly or mountainous. In the east, greater precipitation and topographic relief have produced a high drainage density of relatively short, fast-flowing and often perennial rivers. In the west, rivers are longer, farther apart, and naturally intermittent or ephemeral.

The Murray–Darling Basin (MDB: 1,061,469 km2) falls between the latitudes of 24 and 38°S and includes the Australian Capital Territory and parts of four States (New South Wales, Queensland, South Australia and Victoria). Its elevation ranges from sea level to 2228 m, and it contains numerous flowing and standing water bodies, including Australia’s three longest rivers—the Darling (2740 km), Murray (2530 km) and Murrumbidgee (1690 km). The MDB is mostly arid or semiarid, and many of its rivers stop flowing or dry completely during droughts unless sustained by releases from dams. The MDB’s annual average discharge is only 4700 GL, but would be 12,200 GL without water diversions (CSIRO, 2008). It is an important agricultural region, containing about 65% by area of Australia’s irrigated pastures and crops and 39% by value of the nation’s farm production (ABS, 2008).

Annual average air temperature for the MDB has been above the Australian Bureau of Meteorology’s baseline (1961–1990 average) in every year since 1997, although only slightly so during 2010–2012 (Fig. 1). Annual rainfall has been below the baseline since 2001, except during 2010–2012 and in 2016 (Fig. 1). The long period of higher temperatures and reduced rainfall from about 2001 until 2009 has been referred to as the ‘Millennium Drought’ (Van Dijk et al., 2013).

Fig. 1
figure 1

Mean annual air temperature and precipitation for the Murray–Darling Basin from 1991 to 2016. Horizontal lines are long-term (1961–1990) averages, and dark bars are years during which invertebrate sampling occurred. Data from the Australian Bureau of Meteorology

Full size image

Derivation of environmental preferences

Environmental preferences (EPs) of aquatic invertebrate families for dissolved oxygen (hereafter referred to as hydroaerophily), current velocity (rheophily) and temperature (thermophily) were estimated from 8928 bioassessment samples and associated contemporaneous environmental data collected by NSW State agencies. The samples were obtained by kick sampling from riffles and sweep sampling from pool edgewaters in the State of New South Wales and the Australian Capital Territory in the years 1994–2010 (see Chessman, 2012). The analysis was undertaken at family level because this is the usual level of taxonomic resolution for stream bioassessment in Australia. The preference of a family for an environmental variable was estimated by calculating a weighted mean value of the environmental variable at which the taxon occurred (M) as follows:

$$M = \sum\limits_{i = 1}^{n} {Ei.pi \Big/\sum\limits_{i = 1}^{n} {pi} }$$

where Ei is the ith value of the environmental variable and pi is the proportion of samples collected at that value that contained the taxon. Actual current velocity was not available for any of the samples and was therefore assigned unitless values of 0 for pool samples and 1 for riffle samples. Dissolved oxygen, measured with various field meters, was grouped into 16 values (0–15 mg/L, rounded to the nearest whole number) and water temperature, also measured with field meters, into 37 values (2–38°C, again rounded to the nearest whole number).

Testing of derived environmental preferences

The behaviour of the estimated EPs was examined with data from the Sustainable Rivers Audit (SRA: Davies et al., 2010), a monitoring programme that sampled invertebrates from rivers across the MDB from late 2004 until mid-2012. This programme commenced during the Millennium Drought and concluded during a much wetter period when widespread flooding occurred.

For the SRA, individual catchments within the basin were visited on a rotating basis, with each sampled once every 2 years in either autumn–early winter or spring–early summer. Thus, there were four sampling rounds comprising the periods 2004–2006, 2006–2008, 2008–2010 and 2010–2012, with the third round ending as the drought was breaking and the last round falling in the initial post-drought period. Invertebrate sampling sites within each catchment were selected by a stratified random process wherein the catchment was divided into altitudinal zones and sites were chosen randomly from a defined stream network within each zone, but with restrictions on the proximity of sites to one another. Back-up sites were selected to replace sites that were dry when visited. Altogether, 779–816 sites were sampled across the basin in each round. Some 7% of sites were sampled in every round but 89% were sampled only once.

Two invertebrate samples were collected from each site on each sampling occasion. For sites with fast-flowing water, one sample was collected from riffle habitat by kick sampling and one from slow-flowing or still water near the stream edge by sweep sampling, in both cases with a hand net (250 μm mesh) over an estimated distance of 10 m. If flowing riffles were absent, two edgewater samples were collected. Methods varied geographically because individual State agencies followed different protocols; for example, in South Australia, samples were preserved and sub-sampled randomly in the laboratory, whereas in other States, live samples were sub-sampled selectively on site.

Data provided by the Murray–Darling Basin Authority from its SRA database comprised the number of specimens of each invertebrate taxon recorded from each sample. These taxa ranged from genus to phylum, but most were families, and taxonomy was sometimes outdated and inconsistent across the basin. For uniformity in analysis, I grouped data at family level and ignored all specimens not identified at least to family.

I first tested the relationship between the EPs and long-term climate using the WorldClim global climate grids (www.worldclim.org; accessed April 2016). These grids provide climatic variables interpolated between weather stations to a resolution of 30 arc-seconds, or approximately 1 km (Hijmans et al., 2005). Two variables were used for this analysis: mean air temperature (MAT) and mean annual precipitation (MAP), both calculated over the period 1950–2000. I extracted values for all of the SRA sampling sites and used linear and nonlinear regression to relate them to the average hydroaerophily, rheophily and thermophily of the families in each sample. I conducted separate analyses for riffles and edgewaters. I hypothesised that areas with low MAT and high MAP would have cooler, faster-flowing and more aerated streams and would therefore favour families with low thermophily and high hydroaerophily and rheophily. Conversely, areas with high MAT and low MAP would favour families with low hydroaerophily and rheophily and high thermophily.

Next, I assessed how mean hydroaerophily, rheophily and thermophily varied between the SRA sampling rounds. To do so, I first removed the relationships to long-term climate by regressing average hydroaerophily, rheophily and thermophily on MAT and MAP and calculating residuals from these regressions. I then compared residuals between sampling rounds using analysis of variance with post hoc Tukey tests. Consideration of residuals rather than raw values was necessary to correct for a bias whereby sites sampled during drier periods were on average in areas with cooler and wetter long-term climates, because many streams in warmer and low-rainfall areas were dry and could not be sampled during such periods. I hypothesised that the average hydroaerophily and rheophily of the fauna would be greater during wetter periods when stream flow was stronger and more aerated, whereas the average thermophily of the fauna would increase during drier and warmer periods when taxa with higher temperature preferences would be favoured.

I also examined relationships between inferred dissolved oxygen, current and temperature preferences and two respiratory traits of aquatic life-history stages: possession of gills and possession of mechanisms for accessing atmospheric oxygen, such as lungs, snorkels, bubble trapping, plastron respiration or a surface-dwelling habit. Information on respiratory mechanisms was obtained from a wide variety of literature and online sources, including Keilin (1944), Hinton (1976), Thorpe (1950), Schäfer et al. (2011) and ‘Identification and Ecology of Australian Freshwater Invertebrates’ (www.mdfrc.org.au/bugguide/). I used t-tests to compare families with and without each of these two respiratory adaptations. I hypothesised that gill-breathing families would be more reliant on dissolved oxygen and therefore would have greater hydroaerophily and rheophily than families accessing atmospheric oxygen.

Finally, I used t-tests to compare mean hydroaerophily, rheophily and thermophily between families with and without the following dietary components: animal material, fine organic deposits, periphyton, plankton and seston, and plant material. Dietary information was obtained from a wide range of literature sources, including Chessman (1986), Schäfer et al. (2011) and ‘Identification and Ecology of Australian Freshwater Invertebrates’ (www.mdfrc.org.au/bugguide/). More than one dietary component per family was possible. This analysis was used to infer possible consequences of shifts in environmental preferences at the assemblage level for ecosystem processes. This analysis was exploratory because in this case I lacked a basis to construct a priori hypotheses.

Results

The ranges of estimated EPs were 2.5–11.0 for hydroaerophily, 0.0–1.0 for rheophily, and 5.0–32.4 for thermophily (Appendix). Pearson correlations among the three traits were 0.57 for hydroaerophily and rheophily, − 0.55 for hydroaerophily and thermophily, and − 0.44 for rheophily and thermophily (P < 0.001 in each case).

Whereas taxon richness (the number of taxa per sample) was only weakly related to long-term mean air temperature and mean precipitation (Fig. 2), the average hydroaerophily, rheophily and thermophily of families in a sample were more strongly correlated with these variables (Figs. 3, 4, 5). My hypotheses were supported in that hydroaerophily and rheophily were positively related to MAP and negatively related to MAT for both pool-edge and riffle habitats, whereas thermophily showed the opposite pattern.

Fig. 2
figure 2

Relationships of taxonomic richness per sample to mean annual air temperature and precipitation (WorldClim 30 s grid) for samples from edgewaters and riffles. Fitted lines are linear regressions

Full size image
Fig. 3
figure 3

Relationships of average hydroaerophily per sample to mean annual air temperature and precipitation (WorldClim 30 s grid) for samples from edgewaters and riffles. Fitted lines are quadratic regressions

Full size image
Fig. 4
figure 4

Relationships of average rheophily per sample to mean annual air temperature and precipitation (WorldClim 30 s grid) for samples from edgewaters and riffles. Fitted lines are quadratic regressions

Full size image
Fig. 5
figure 5

Relationships of average thermophily per sample to mean annual air temperature and precipitation (WorldClim 30 s grid) for samples from edgewaters and riffles. Fitted lines are quadratic regressions

Full size image

Multiple regressions of hydroaerophily, rheophily and thermophily on MAT and MAP had coefficients of determination between 0.34 and 0.67, while taxon richness had coefficients of 0.10–0.13 (P < 0.001 in all cases). Residuals from these regressions generally behaved as predicted. The average residual for richness declined during the drought (first three sampling rounds) and did not recover immediately afterwards (fourth round) (Fig. 6), but residuals for hydroaerophily and rheophily declined during the drought and recovered afterwards, whereas the residual for thermophily increased during the drought and fell thereafter (Fig. 7). These changes were more pronounced for riffle samples than for edgewater samples.

Fig. 6
figure 6

Mean (± SE) residuals from regression of taxon richness per sample on long-term climatic variables. Values are presented for edgewater and riffle samples in each sampling round (1: mid-drought; 2: late drought; 3: end of drought; 4: post-drought). Values with the same letter are not significantly different (Tukey test)

Full size image
Fig. 7
figure 7

Mean (± SE) residuals from regression of average hydroaerophily per sample, average rheophily per sample, and average thermophily per sample on long-term climatic variables. Values are presented for edgewater and riffle samples in each sampling round (1: mid-drought; 2: late drought; 3: end of drought; 4: post-drought). Values with the same letter are not significantly different (Tukey test)

Full size image

As hypothesised, families with gills were on average significantly more hydroaerophilous (t test, P = 0.03) than those without gills (Fig. 8). They were also significantly less thermophilic (P = 0.007). In contrast, families with some form of aerial respiration were, as expected, on average significantly less hydroaerophilous (t test, P = 0.02) than those without aerial respiration (Fig. 8). They were also less rheophilous (P < 0.001) and more thermophilic (P < 0.001).

Fig. 8
figure 8

Average values of hydroaerophily, rheophily and thermophily (± SE) for families with aquatic life-history stages bearing and lacking gills, and with and without means of aerial respiration

Full size image

Consumers of animal material were on average significantly less rheophilous (P = 0.03) and more thermophilous (P < 0.001) than non-consumers of animal material (Fig. 9). Consumers of plant material (i.e. shredders) were on average significantly more hydroaerophilous (P = 0.006) and rheophilous (P = 0.04) and less thermophilous (P < 0.001) than non-consumers of plant material (Fig. 9).

Fig. 9
figure 9

Average values of hydroaerophily, rheophily and thermophily (± SE) for families feeding and not feeding on animals (A), fine organic deposits (FOD), periphyton (Pe), plankton and seston (P/S) and plant material (Pl)

Full size image

Discussion

Several factors restricted the analysis undertaken here. Data constraints limited taxonomic resolution to family level, but finer taxonomic resolution would have been preferable because trait values can vary appreciably among species or genera within a family (Monaghan & Soares, 2013). Further, the SRA did not sample before the Millennium Drought or during its early stages, and therefore initial responses of invertebrate assemblages, which might have been substantial, were excluded from the analysis. Moreover, the last sampling round began very early in the post-drought period and, as a result, the final assemblage change probably did not reflect the full post-drought response. In addition, the estimated hydroaerophily, rheophily and thermophily derived in this study from field distributions could not be directly assessed for accuracy because experimental data on environmental preferences of the study families are lacking. Nonetheless, the use of these traits to assess responses to spatial and temporal climatic variation, and their correlations with respiration mode, were underpinned by several hypotheses. Support for these hypotheses suggested that the derived values are of practical use.

The average hydroaerophily, rheophily and thermophily of families in samples from both edgewaters and riffles were strongly associated with values of long-term climatic variables for study sites and in the expected directions. That is, families present in cooler and wetter areas tended to favour higher concentrations of dissolved oxygen, faster current and lower temperatures than those in warmer and drier areas. These relationships were strong even though the correlations were with the terrestrial climate rather than the aquatic environment. Air temperature probably acted as a reasonable surrogate for water temperature over such a large area and climatic range, and precipitation as a surrogate for stream perenniality. The strong correlations of faunal composition with spatial variation in climate, and marked responses to short-term climatic fluctuations, suggest that the fauna of this region will be sensitive to long-term climatic changes. Temperatures across the Murray–Darling Basin are projected to increase during the 21st century and rainfall to decline (Maxino et al., 2008; Smith & Chandler, 2010). Thus, species with high hydroaerophily and rheophily and low thermophily are likely to be adversely affected and species with the opposite traits are likely to benefit.

Average hydroaerophily, rheophily and thermophily increased or declined as expected during the drought and demonstrated post-drought recovery, whereas taxonomic richness showed no evidence of recovery. Declines of cold-water taxa in warmer years have also been reported in the USA (Stamp et al., 2010), and declines of rheophiles in dry years in Europe (Verkaik et al., 2013; Verdonschot et al., 2015). Post-drought recovery of the trait composition of the MDB assemblages was rapid, suggesting swift recolonisation, but the recovery mechanism is unknown. Notably, both impact and recovery were more pronounced in riffles than in pools, a finding that may reflect the greater environmental change (e.g. in velocity) occurring in the riffle environment.

Variation in estimated environmental preferences could be explained to some degree by respiration mode. As expected, families respiring with gills in aquatic life-history phases were more likely to favour cool, well aerated waters, whereas those able to access atmospheric oxygen were able to occupy warmer waters with lower dissolved oxygen concentrations. Similar or related findings have been reported globally. Thus, Jesus (2008) also found that organisms with branchial respiration occurred at sites with higher dissolved oxygen concentrations whereas species with aerial respiration could occur at sites with low dissolved oxygen concentrations. Horrigan & Baird (2008) observed that water velocity had a negative association with aerial respiration and a positive association with respiration by gills, while Dolédec et al. (2015) found that organisms with gill respiration responded positively and animals with aerial respiration responded negatively to flow restoration in a large river. Bêche & Resh (2007) associated gill respiration with wet years and aerial respiration with dry years. The association of estimated environmental preferences with respiration mode in the present study provides some confidence that the former are meaningful.

Rheophiles and species with low thermal preferences tend to be less common in catchments with more intense land use (Zuellig & Schmidt, 2012). Species with lower temperature preferences and less tolerance of low dissolved oxygen levels are also less common at anthropogenically influenced sites than would be expected in the absence of human impact, whereas species with the opposite traits are more common (Carlisle & Hawkins, 2008; Carlisle et al., 2014). The results presented here support the view that the climatic warming and drying that are forecast for many parts of the globe are likely to result in further declines of species with high hydroaerophily and rheophily and low thermophily (Chessman, 2009, 2015; Hering et al., 2009; Tierno de Figueroa et al., 2010; Rosset & Oertli, 2011; Sandin et al., 2014; Hershkovitz et al., 2015). Thus, estimates of environmental preferences derived from field data should be useful in identifying taxa that are at risk from climate change.

As estimated here, hydroaerophily, rheophily and thermophily were not independent, with strong inter-correlations among these three traits as well as with respiration mode. These associations suggest the existence of a trait syndrome related to adaptation to either rhithron or potamon and lentic environments. Adaptation to the cool, fast-flowing rhithron environment logically involves branchial respiration and preferences for high dissolved oxygen concentrations, fast-flowing water and lower temperatures, while adaptation to the potamon may involve aerial respiration and tolerance of low dissolved oxygen concentrations, stagnant or slow-flowing water and high temperatures.

Dietary variation between families with differing environmental preferences indicates that loss of taxa with particular environmental preferences as a consequence of climate change will have implications for ecosystem functioning. In particular, any loss of drought-sensitive species with high hydroaerophily and rheophily and low thermophily is likely to result in reduced processing of terrestrial plant litter because shredders of plant material tended to have these preference traits. This prediction is consistent with findings of Ledger et al. (2011), who observed a strong decline in shredder abundance in response to simulated drought in mesocosms.

In summary, this study suggests that environmental preferences derived from field data are likely to be useful in identifying species at risk from directional climate change and extreme events such as multi-year droughts, and in predicting consequences of climate-driven community changes for ecosystem processes. Since monitoring data such as used here are widely available from agency bioassessment programs, the approach tested here should be capable of widespread implementation globally.