Modelling macroinvertebrate and fish biotic indices

1 downloads 0 Views 1MB Size Report
EPT and hydromorphological conditions; LIFE and percentage of salmo- nid biomass). ...... Manual for the Application of the AQEM System. A Comprehensive Method .... Atlas y Libro Rojo de los Peces Continentales de España. Madrid, Spain.
Science of the Total Environment 577 (2017) 308–318

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Modelling macroinvertebrate and fish biotic indices: From reaches to entire river networks Mario Álvarez-Cabria ⁎, Alexia M. González-Ferreras, Francisco J. Peñas, José Barquín Environmental Hydraulics Institute, Universidad de Cantabria, Avda. Isabel Torres, 15, Parque Científico y Tecnológico de Cantabria, 39011 Santander, Spain

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

• Invertebrate and fish indices were predicted across whole river networks (110,000 km2). • Models were done with the Random Forest analysis using 25 predictor variables. • Invertebrate and fish metrics showed similar spatial patterns. • All indices were mainly driven by the same environmental factors. • Indices were higher in the Atlantic region than in the Mediterranean one.

a r t i c l e

i n f o

Article history: Received 15 June 2016 Received in revised form 12 September 2016 Accepted 25 October 2016 Available online 29 October 2016 Editor: D. Barcelo Keywords: IASPT EPT Life Salmonid biomass Ebro River Cantabrian region

a b s t r a c t We modelled three macroinvertebrate (IASPT, EPT number of families and LIFE) and one fish (percentage of salmonid biomass) biotic indices to river networks draining a large region (110,000 km2) placed in Northern and Eastern Spain. Models were developed using Random Forest and 26 predictor variables (19 predictors to model macroinvertebrate indices and 22 predictors to model the fish index). Predictor variables were related with different environmental characteristics (water quality, physical habitat characteristics, hydrology, topography, geology and human pressures). The importance and effect of predictors on the 4 biotic indices was evaluated with the IncNodePurity index and partial dependence plots, respectively. Results indicated that the spatial variability of macroinvertebrate and fish indices were mostly dependent on the same environmental variables. They decreased in river reaches affected by high mean annual nitrate concentration (N 4 mg/l) and temperature (N12 °C), with low flow water velocity (b0.4 m/s) and aquatic plant communities being dominated by macrophytes. These indices were higher in the Atlantic region than in the Mediterranean. This study provides a continuous image of river biological communities used as indicators, which turns very useful to identify the main sources of change in the ecological status of water bodies and assist both, the integrated catchment management and the identification of river reaches for recovery. © 2016 Elsevier B.V. All rights reserved.

⁎ Corresponding author. E-mail addresses: [email protected] (M. Álvarez-Cabria), [email protected] (A.M. González-Ferreras), [email protected] (F.J. Peñas), [email protected] (J. Barquín).

http://dx.doi.org/10.1016/j.scitotenv.2016.10.186 0048-9697/© 2016 Elsevier B.V. All rights reserved.

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

1. Introduction The ecological status assessment of aquatic ecosystems in Europe, according to the Water Framework Directive (WFD), requires the evaluation of their different components: 1) biological communities, 2) water physico-chemical characteristics and 3) hydromorphological conditions (in order of importance; European Commission, 2000). Macroinvertebrate communities are commonly used to evaluate water quality deterioration, mainly because of their broad response to different water quality stress gradients (Álvarez-Cabria et al., 2010) and the efficient results they provide (Bonada et al., 2006). Fish communities are also commonly used as bioindicators, but in this case mainly in order to analyse the hydromorphological conditions of rivers and streams (flow regime and longitudinal connectivity; e.g. Schmutz et al., 2016). However, in many cases, a wide variety of human impacts are simultaneously affecting fluvial ecosystems (water quality and hydromorphological conditions), changing the structure and composition of biological communities in a different way than a single alteration would, due to concomitant effects (Piggott et al., 2012). Therefore, understanding the effects of multiple anthropogenic stressors is essential to disentangling the response of biological communities and their associated indices. For instance, this is the case of catchments that are severely affected by agriculture. This land use has been identified as highly responsible for water quality deterioration in Europe, because the excessive use of fertilisers increases the risk of eutrophication (Lowicki, 2012). Moreover, agricultural landscapes are usually affected by deforestation, which increases the incidence of solar radiation on soils and the transference of heat to rivers and streams (Binkley and Fisher, 2013). Furthermore, the agricultural activity not only changes water quality (eutrophication and water temperature), but it also causes major changes in the hydromorphological elements of fluvial habitats. Thus, catchments dominated by agricultural land uses are more affected by sediment inputs than forested ones, especially where natural riparian vegetation has been removed or altered (Piggott et al., 2012; Burdon et al., 2013). Moreover, the use of infrastructures for crop irrigation, such as reservoirs, modifies the natural hydrologic and hydraulic properties of rivers and streams, changing the longitudinal distribution of fishes and other aquatic communities. Understanding the mechanisms that drive the variability of biotic indices is sometimes a difficult task, because this variability is influenced by the complex interplay between a wide array of anthropogenic stressors and the natural factors affecting river ecosystems (ÁlvarezCabria et al., 2010). Moreover, the high natural spatial variability of biological communities and the scarcity of data hamper the evaluation of large water bodies and the development of River Catchment Management Plans, which are usually based on the values of biotic indices calculated from biological communities sampled in few sites. In this regard, results from biotic indices representing a given river reach (100 m length) are then used to evaluate the integrity and the ecological status of large, and sometimes highly variable, water bodies (see European Commission, 2000). Carrying out field surveys to estimate the variability of the structure and composition of the biological communities, and the subsequent biotic indices, within the water bodies of a river network is not possible due to the high costs and intensity of such surveys. The absence of spatial continuous data for these biological communities makes it difficult to assess the river ecosystem integrity without making very rough assumptions as to “representativeness” of the selected river reach for surveying within the water body. Data modelling might be an effective solution to complete this lack of data, because it allows predicting the different biotic indices to all river network in relation to different types of anthropogenic impacts and natural factors. In this study, we defined a virtual watershed (sensu Barquín et al., 2015) as a framework to incorporate the environmental and biological information of a large area (110,000 km2) situated in Spain (Europe). We used this digital framework to investigate how natural and

309

anthropogenic factors, acting at catchment and sub-catchment scale, affected the variability of three biotic indices based on macroinvertebrate communities (the Iberian Average Score per Taxon: IASPT, the Ephemeroptera, Plecoptera and Trichoptera number of families: EPT and the Lotic Invertebrate Index for Flow Evaluation: LIFE) and one index based on fish community (the percentage of salmonid biomass). Thus, our main objective was to model these biotic indices for all the reaches in the whole river network. We chose these biotic indices because they represent the structure and composition of two different biological communities and they were proposed and designed to assess the integrity of different fluvial ecosystem components (water quality; IASPT and EPT and hydromorphological conditions; LIFE and percentage of salmonid biomass). The second objective was to identify the most important variables that define the variability of these biotic indices. Most reviewed models in the literature analyse the variation of biotic indices estimated from only one type of biological community; such as diatoms (Teittinen et al., 2015), invertebrates (Erba et al., 2015), macrophytes (Demars et al., 2012) or fishes (González-Ferreras et al., 2016). However, to date we have not found many studies that model the response of biotic indices calculated from different communities in the same study area (but see Clapcott et al., 2016, who modelled biotic indices calculated from fish, invertebrate and algal communities). Thus, we are especially interested in determining whether biotic indices designed to evaluate different impact types (e.g. organic enrichment or low flow) and calculated from different biological communities (macroinvertebrates and fishes), show similar or different spatial patterns, as well as exploring how they are affected by the different natural and anthropogenic factors. 2. Methods 2.1. Study area The study area (110,000 km2) is situated in northern and eastern Spain and includes river catchments managed by three different regional water agencies (Confederación Hidrográfica del Cantábrico: CHC, Agencia Vasca del Agua: URA and Confederación Hidrográfica del Ebro: CHE). The study area can be grouped into two large territories (Fig. 1). 1. The first one (25,000 km2) is managed by the CHC and URA agencies. Rivers and streams from this territory drains into the Atlantic Ocean (Cantabric Sea; Fig. 1). This region contains small-medium sized catchments (from 30 km2 to 5000 km2) with high slopes. In this area climate varies from thermo-temperate Atlantic climate, on the coast, to an oro- and supra-temperate climate in the interior (Rivas-Martínez et al., 2004). Average annual precipitation is 1300 mm. Broad-leaved forest, shrubs and grasslands occupy 80%, while agriculture only reaches 10% of this region. On average, the population density is 100 inhabitants/km2. 2. The second territory (85,000 km2), managed by the CHE agency, is occupied by the Ebro catchment and drains into the Mediterranean Sea. This region has an average of 32 inhabitants/km2, and it is characterised by a Meso-mediterranean and Supra-mediterranean climate, with average annual precipitation of 650 mm (from 300 mm in the central area, to 1700 mm in the Pyrenees mountain range). Snow is abundant in winter and early spring (Bejarano et al., 2010). Agriculture covers 50% of the Ebro catchment. 2.2. Biological and environmental information The biological database was completed with information from the water agencies indicated above. We compiled information from macroinvertebrate communities surveyed in 1077 river reaches, with 436 belonging to CHC, 204 to URA and 437 to CHE. Invertebrates were sampled following the protocols of these 3 water agencies. Samples were taken

310

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

Fig. 1. Study area, including the virtual watershed and the synthetic river network.

in summer (July–October) from 2003 to 2009, with a kick hand net (0.5 mm mesh size). Every sample was composed by 20 (CHC and CHE) and 5 kicks (URA). However, in sites sampled by the URA, the taxonomic list was completed in the field by searching new taxa not included in the sample (5 kicks). In each site kicks were distributed according to the importance of the different microhabitats (e.g. pools, runs, macrophytes or submerged banks roots; AQEM, 2002). Samples were preserved in ethanol (70°) and divided in 3 fractions in the laboratory (b1 mm, 1–5 mm, and N5 mm). Invertebrate abundance was estimated by subsampling at least 100 individuals in each fraction. Finally, the two major fractions (N1 mm) were thoroughly checked for new taxa to evaluate the number of taxa in the community. Invertebrate identification was done to family level, except for Hydracarina, Nematoda, Ostracoda, Copepoda and Oligochaeta. To characterise and evaluate the integrity of these communities in relation to water quality we calculated two biotic indices: IASPT and EPT. The values of these biotic indices decrease with water quality deterioration (Álvarez-Cabria et al., 2010). To assess the effects of hydromorphological alterations over invertebrate communities we calculate the LIFE index, which decreases with flow reduction (Extence et al., 1999). In order to assess the possible differences on macroinvertebrate biotic indices because of the use of various sampling efforts, we have compared the values of the calculated metrics in the 10 sites that were available in our database surveyed in the same year and season with both methods (20 and 5 kicks). The mean IASPT value in these 10 sites was the same using 5 and 20 kicks (5.1 ± 0.3), while the mean LIFE value was quite similar (mean 20 kicks = 6.8 ± 0.2 and mean 5 kicks = 6.5 ± 0.4). However, the number of EPT families was on average 0.24 times higher in sites surveyed with the 20 kicks method. To homogenise the values of the EPT metric, the results of this index were multiplied by 1.24 in all the samples taken with the 5 kick method (a total of 204 URA sites). This approach was selected as all water agencies use composite samples for each site (they do not have information from each of the separate kick samples) and rarefaction curves could not be used instead. The fish community database was developed with information from 590 river reaches, with 157 belonging to CHC, 106 to URA, 225 to CHE and 102 to the Regional Government of Cantabria. All samples were taken in summer (July–October), from 2003 to 2009, using

electrofishing techniques. The length sampled in each site was 10 times the width of the river (see UNE-EN 14011:2003, 1998; in Spanish). Therefore, the sampling area in high order rivers was generally higher than in low order streams, varying from 22 m2 to 4400 m2, in order to maintain the sampling effort constant over wider rivers. This database presents a certain degree of heterogeneity, however, it allows maximising the coverage of field distribution data for the different species of fishes (see Leathwick et al., 2006). To characterise and evaluate how fish communities change in relation to natural and anthropogenic factors, we calculated the percentage of salmonid biomass in relation to the total fish community biomass. Salmonids were only represented by two native species: brown trout (Salmo trutta, presented in the whole study area) and Atlantic salmon (Salmo salar, absent in the Ebro catchment). In order to create a spatial framework to include all the information required for the models (environmental and biological), we developed a virtual watershed using flow directions from a digital elevation model (DEM; 25-m; Benda et al., 2011). This synthetic river network was created with the NestStream software (Miller, 2003). Finally, the virtual watershed was composed of 86,749 river reaches, with lengths between 16 and 800 m (average length of 500 m). An initial set of 37 predictor variables with potential influence on macroinvertebrate and fish communities was selected. Pearson's correlation coefficient between variables was calculated. Variables with correlation N0.7 were not included in the models. Finally, we only used 19 and 22 variables to model biotic indices from macroinvertebrate and fish communities, respectively (Table 1). 1. Water quality variables were measured in 1069 sites and modelled for the whole study area in relation to different environmental attributes (climate, topography, land cover, hydrology, geology and anthropogenic alterations). These models were done following the same approach as the one presented in this study. For a detailed description of how water quality variables were calculated and modelled, please see Álvarez-Cabria et al. (2016). 2. Habitat variables (hydraulic conditions, substrate composition, instream vegetation types, etc.) were characterised in summer in 800 sites using the River Habitat Survey method (RHS; Raven et al.,

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

311

Table 1 Predictor variables used to model biotic indices estimated from macroinvertebrate and fish communities. Predictor variables

Acronym

Macroinvertebrates

Fishes

Mean annual water temperature Mean winter water temperature Mean annual nitrate concentration Mean annual phosphate concentration Flow velocity index. Velocities were measured in summer Abundance of woody debris and leaf litter in the channel Abundance of the different types of aquatic plant communities Substrate size index Channel width High flow events per year (7 times median flow) Flow predictability (sensu Colwell, 1974) Magnitude of the minimum flow (90 days duration) Julian day of minimum flow Julian day of maximum flow Drainage catchment area Synthetic water conductivity index (Inferred from upstream geology) Distance to upstream effluent Distance to downstream effluent Distance to upstream hydromorphological alteration Distance to downstream hydromorphological alteration Total length of the hydromorphological alterations in the river reach Distance to upstream dam Distance to downstream dam Number of weirs and dams in the river reach Distance to downstream dam/gauge/weir Distance to upstream dam/gauge/weir

T_MEAN T_WIN NO3_MEAN PO4_MEAN FI W AV SSI E_WIDTH FRE7 PRED X90LF J_MIN J_MAX AREA_SQKM MN_COND V_DAR V_DAB H_DAR H_DAB H_MTR P_DAR P_DAB P_NTR APP_DAB APP_DAR

X

X X X X X X

3.

4. 5.

6.

1998) and modelled for the whole study area in relation to topographic, geologic, land-use and hydrologic predictors (Barquín et al., 2011). Within this group of variables, we included the aquatic vegetation index (AV), which was calculated using section G of the RHS form (Channel vegetation types). We carried out a Principal Component Analysis (PCA) with this information in order to determine the patterns of correlation between the different vegetation types. The first axis of the PCA was a very good descriptor of the dominance of different aquatic vegetation groups. Positive values of this vegetation index were related to the aquatic vegetation from headwater sites, dominated by mosses and liverworts, while the most negative values were related to the aquatic plant communities form downstream sites, mainly dominated by macrophytes (e.g. Potamogeton spp.). Hydrologic variables were modelled using the mean daily flows from 156 gauge-stations. These gauges were selected according to these requirements: (1) unaltered series (not affected by water impoundment or abstraction) (2) high quality series record (excluding periods of repeated values, non-natural extreme low flows for short time periods, zero flow values in perennial rivers and streams and non-natural flow magnitude rises and falls or large differences between two periods), (3) gaps lower than 30 days/year and (4) series with information of at least 7 years within the period 1976–2006 (for a detailed description see Peñas, 2014). Topographic variables. The drainage catchment area was derived from a 25-m DEM. Geologic variables. The percentage of limestone surface in the catchment and sub-catchment upstream drainage area was derived from the lithostratigraphic and permeability map (1:200,000) developed by the Spanish Geologic and Mining Institute. The base of the calculation of this variable was the percentage area occupied by the original rock classes included in the data layer. These classes were then reclassified into broader ones and then, we assigned them a numerical value based on geological hardness and soil permeability (for details of this approach, please see Peñas, 2014). Anthropogenic pressures. We calculated several variables in relation to the quantity and distance of each river reach in the network to the closest upstream and downstream human alteration: effluents (industrial and urban effluents with loads ≥2000 inhabitant equivalent.

X X X X X X X X X X X X X X X X X

X X X X X X X X X X X X X X

X X X

Sewerage and storm-water effluents were not included), morphological alterations (embankments with lengths ≥100 m) and alterations to river continuity (dams, weirs and gauges with heights ≥0.5 m). All the distances represented by these predictors were calculated using the ESRI's ArcPy Python module (ESRI, 2011). Due to computational limitations, we only identified pressures located upstream of every single reach at a distance range between 0 and 5000 m, while the distance to the nearest pressure located downstream did not have limitations. Finally, we also assessed the total number of effluents, weirs, dams and gauges in the upstream channel and in the river reach itself. We can take the number and distances of effluents as a proxy of urban development, because they are highly correlated in our study area. Thus, point source effluents are indicative of urban pollution, in which heavy metals, drugs, and other chemical compounds could also be important. 2.3. Data analysis We used the Random Forest technique (RF), a non-parametric method widely applied in environmental studies (Snelder et al., 2011; Olson and Hawkins, 2012; Álvarez-Cabria et al., 2016), to predict the values of the biotic indices as a function of the predictors. The RF model comprises an ensemble of individual Classification and Regression Trees (CART; Breiman et al., 1984). CART partitions observations into groups based on a series of splits constructed from the predictors. RF models increase the prediction accuracy of CART by introducing a random variation by growing each tree with a bootstrap sample of the training data and only using a small random sample of the predictors to define the split at each node. The number of trees (t) needs to be sufficiently high to ensure convergence. This value depends on the number of predictors that can be used at each split. Although the value of t can be optimised, we used default values for the number of predictors (the square root of the total number of predictors) and a large number of trees (500) rather than attempting to optimise model performance. In order to assess the model fit we used the fitted R2 between the observed and predicted values of the modelled variables using the entire data set. Moreover, as indicated above, RF fits many CART with a randomised subset of sites and predictor variables. Each CART is used to predict the sites excluded from the data set. These predictions are

312

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

then used to calculate the predictive accuracy of the model giving the predicted R2. In order to define the importance of the predictor variables in the models, we used the IncNodePurity index. This index reports the total decrease in node impurities from splitting on each variable, averaged over all trees. Higher values of this index indicate higher importance of the variables in the regressions (see Kuhn et al., 2008). We also chose a threshold-based approach to visualise the fitted functions using partial dependence plots, which indicate the effect of each predictor on the response variable after taking into account the average effect of all other predictors in the model (Elith et al., 2008). Although graphs are not a perfect representation of the effects caused by each variable, they provide a useful basis to understand the relationship between the response and each predictor variable. All statistical analyses were performed in R (R Development Core Team, 2011). 3. Results 3.1. IASPT model The predicted R2 of the IASPT model was 0.64, showing the highest prediction capacity of the 4 models developed in this study, while the fitted R2 reached 0.94 (Table 2). The modelled mean IASPT value for the whole study area was 5.3, ranging from 3.0 to 7.1. This index had higher values in the Atlantic region (mean: 5.8, highest: 7.1, lowest: 3.4) than in the Mediterranean area (mean: 5.1, highest: 6.5, lowest: 3.0; Fig. 2A). The most important variables predicting the IASPT index were 1) the mean annual nitrate concentration, 2) the aquatic vegetation index and 3) the flow index (Table 2). Following the partial dependence plots, the IASPT value was negatively correlated with NO3_MEAN and AV, while its relationship with FI was positive (Fig. 2B, C and D). IASPT was especially influenced by the concentration of nitrates when the mean annual concentration of this ion increased from 4 mg/l to 10 mg/l (Fig. 2B). IASPT increased with negative values of the AV, indicating higher values of IASPT in rivers dominated by mosses and lichens, than in rivers dominated by macrophytes. Finally, IASPT values increased with summer flow velocity, particularly when this value increased from 0.1 m/s to 0.4 m/s. 3.2. EPT model The predicted R2 of the EPT model was 0.50, while the fitted R2 was 0.94 (Table 2). The geographical distribution of the EPT values was quite similar to the one described previously for the IASPT model. The modelled mean EPT number of families for all the river reaches was 10.5 (from 1.8 to 18.4). Rivers from the Atlantic region had a higher number of EPT families (mean: 12.4, highest: 18.4, lowest: 1.8) than Mediterranean rivers (mean: 9.5, highest: 17.5, lowest: 2.2; Fig. 3A). The most important predictor variables in determining the EPT value were 1) the mean annual nitrate concentration, 2) the aquatic vegetation index and 3) the mean annual water temperature (Table 2). Following the partial dependence plots, EPT was negatively correlated with these three predictors (Fig. 3B, C and D). Patterns followed by EPT in relation to NO3_MEAN and AV were quite similar than those described for IASPT (Figs. 2B, C & 3B, C). Regarding water temperature, the

number of EPT families were more heavily affected when the mean annual water temperature increased from 12 °C to 15 °C (Fig. 3D). 3.3. LIFE model The fitted R2 of this model was 0.93, explaining 50% of the total variability (R2 = 0.50), which was the lowest prediction capacity of the four models presented in this study (Table 2). The modelled mean value of this index was 7.1 for the whole study area. Once again, rivers from the Atlantic region showed higher values of this macroinvertebrate index (mean: 7.5, highest: 8.4, lowest: 5.6) than Mediterranean rivers (mean: 6.9, highest: 8.0, lowest: 5.2). Although the IASPT and EPT indices were designed to assess water quality and the LIFE index was proposed to assess changes in river flow, the IncNodePurity index indicated that LIFE was related with the same predictors as IASPT and EPT. Thus, the three most important variables to predict the LIFE index were 1) the aquatic vegetation index, 2) the mean annual nitrate concentration and 3) the flow index (Table 2). Following the partial dependence plots, LIFE was negatively correlated with AV and NO3_MEAN and positively related with flow velocity (FI; Fig. 4B, C and D). As described above for IASPT and EPT, the LIFE index mainly decreased when AV increased from − 1 to 1 and when NO3_MEAN increased from 4 mg/l to 10 mg/l. On the other hand, higher increments in LIFE occurred when the summer flow velocity increased from 0.1 m/s to 0.5 m/s. 3.4. Percentage of salmonid biomass The fitted R2 of this model was the highest of the four models (0.95; Table 2), while its predicted R2 was 0.56. The modelled mean value of this index was 52% for the whole study area. Again, rivers from the Atlantic region showed higher values of this index (mean: 64%, highest: 99%, lowest: 0%) than Mediterranean rivers (mean: 46%, highest: 98%, lowest: 0%). The variability of this biotic index was influenced by the same predictor variables as described above for the macroinvertebrate biotic indices (Table 2). Thus, the percentage of salmonid biomass was mainly dependent on 1) the mean annual nitrate concentration, 2) the Flow Index and 3) the mean annual water temperature. The biomass of salmonids showed a negative correlation with the NO3_MEAN, mainly when the nitrate mean annual concentration increased from 4 mg/l to 6 mg/l (Fig. 5B), and with the T_MEAN, mainly when increased form 12 °C to 14 °C (Fig. 5D). Finally, the correlation between the percentage of salmonid biomass and the flow velocity in summer was positive, particularly when velocity increased from 0.2 m/s to 0.5 m/s (Fig. 5C). 4. Discussion We modelled four biotic indices used to estimate the ecological status of river ecosystems in entire river networks covering 110,000 km2 in Northern and Eastern Spain. Although these biotic indices were descriptors of the structure and composition of different fluvial communities (invertebrates and fishes), and were designed to evaluate different types of anthropogenic alterations (water quality and flow conditions), our results indicated that the spatial variability of these biotic indices was very similar. Moreover, models also indicated that biotic indices

Table 2 Mean modelled values and predictive accuracy and fit of the models. The three most important predictors and their contribution to the variance explained for each model, following the IncNodePurity Index, are also included (NO3_MEAN: mean annual nitrate concentration, AV: Aquatic vegetation index, FI: Flow index, T_MEAN: mean annual water temperature). Biotic indices

Average value

Predicted R2

Fitted R2

First predictor

Second predictor

Third predictor

IASPT EPT LIFE Salmonid biomass (%)

5.33 10.49 7.14 52%

0.64 0.50 0.50 0.56

0.94 0.94 0.93 0.95

NO3_MEAN; 27% NO3_MEAN; 21% AV; 18% NO3_MEAN; 18%

AV; 13% AV; 13% NO3_MEAN; 15% FI; 14%

FI; 10% T_MEAN; 12% FI; 10% T_MEAN; 13%

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

313

A) B)

C)

D)

Fig. 2. Result of the modelled IASPT (A) and partial dependence plots with the functions for the most important predictors: (B) mean annual nitrate concentration, (C) aquatic vegetation index and (D) flow velocity index.

were mainly influenced by the same environmental variables: (1) mean annual nitrate concentration, (2) mean annual water temperature, (3) low flow water velocity and (4) the aquatic vegetation type. The fit to the training data set was very high in all cases (fitted R2 ranged from 0.93 to 0.95; Table 2). Our models improved the results of other studies where macroinvertebrate indices were modelled with the Random Forest technique (Unwin and Larned, 2014) and with other statistical approaches (e.g. Multiple Regression Analysis, Erba et al., 2015 or Generalised Linear Models, Leps et al., 2015). In these studies, the explained variance of the modelled biotic indices ranged from 0.54 to 0.72, from 0.21 to 0.64 and from 0.11 to 0.52, respectively. Thus, our fitted R2 values demonstrate the capacity of our models to learn the information in the training data set and to identify the main relationships between biotic indices and the environmental variables. However, these results are not a good indicator of the prediction capacity of the models, and thus, we used the predicted R2 to assess their accuracy. The predictive performance of our models was higher than other macroinvertebrate indices models developed with Boosted Regression Trees, in which the prediction capacity ranged from 0.22 to 0.54 (Pilière et al., 2014). In fact, our results were better than expected, because they improved the performance of a previous study in which we modelled the spatial and seasonal variability of water quality in the same study area, using the same approach and analytical techniques (Álvarez-Cabria et al., 2016). In Álvarez-Cabria et al. (2016) the predicted R2 was N0.5 in just three of the fifteen proposed models, while in the present study all the models presented a predictive R2 ≥ 0.5. The improvement in the predictive capacity of the current models may be partly explained by the larger size of the data set (1077 sites for macroinvertebrates and 590 for fishes) in comparison to our previous study (297 sites for water temperature, 267 for nitrates and 250 for phosphates). Similar conclusions have been reported by Fernández et

al. (2014), indicating that Random Forest models perform better when using an elevated number of training sites, like other machine learning techniques (e.g. neuronal networks; Andonie, 2010). 4.1. Macroinvertebrate biotic indices Our results confirmed that the structure and composition of macroinvertebrate communities in the selected study area was mainly shaped by water physico-chemical conditions (nitrates and water temperature) and by physical habitat characteristics (water velocity and the structure and composition of aquatic plant communities). As indicated above, the three modelled macroinvertebrate biotic indices were affected by the same environmental predictors, although they were designed to evaluate different types of anthropogenic impacts: organic pollution (IASPT and EPT) and changes in the natural flow regime and in the hydraulic characteristics (LIFE). Understanding why the variability of these biotic indices is mainly dependent on the same environmental variables requires looking at the biological traits that characterise the functionality of the different macroinvertebrate taxa. IASPT is a biotic index calculated from the IBMWP index (Alba-Tercedor and Sánchez-Ortega, 1988) which is the Spanish version of the Biological Monitoring Working Party index (BMWP; Hellawell, 1978). The IBMWP allocates scores to the different macroinvertebrate families in relation to their tolerance to organic pollution; from 1 (very high tolerance) to 10 (very low tolerance/sensitive). On the other hand, the LIFE index also allocates scores to the different macroinvertebrate families in relation to their preferential flow velocity. In this regard, the LIFE index classifies macroinvertebrates in six groups, from group 1 (GI; families primarily associated with rapid flows; N1 m/s), to group 6 (GVI; families frequently associated with lentic or drying sites; for a more detailed description of these six groups, please see Extence et al., 1999). However, in our database we

314

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

A) B)

C)

D)

Fig. 3. Result of the modelled for the EPT number of families (A) and partial dependence plots with the functions for the most important predictors: (B) mean annual nitrate concentration, (C) aquatic vegetation index and (D) mean annual water temperature.

did not have macroinvertebrates from GVI (e.g. Chirocephalidae). All the insect families from our database belonging to the group V (GV; Taxa primarily associated with standing waters) breathe atmospheric oxygen through different mechanisms, such as siphons (e.g. Nepidae), spiracles (e.g. Culicidae) or aerial vesicles (e.g. Hygrobiidae; see Merritt and Cummins, 1996; Tachet et al., 2000) and have a IBMWP score between 1 and 3 (very tolerant to organic pollution). Thus, these aquatic macroinvertebrates can live in rivers and streams affected by organic enrichment and low-dissolved oxygen concentrations in the water column (Álvarez-Cabria et al., 2011). This could be the case of river reaches affected by high nitrate concentrations in our study area. Several studies have reported a negative correlation between oxygen and nutrient concentration (e.g. phosphates and nitrates; Elósegui and Pozo, 1994; Arimoro et al., 2015). This relationship mainly occurs because the increment of these chemical compounds enhance the development of instream producer communities, such as phytoplankton and macrophytes, which can drastically reduce the oxygen concentration during non-light hours, due to breathing processes are higher than the oxygen production (Elósegui and Pozo, 1994). On the other hand, insect families from the LIFE GI (i.e. Heptageniidae, Capniidae, Chloroperlidae, Perlodidae, Perlidae, Rhyacophilidae, Odeontoceridae, Philopotamidae, and Goeridae) breathe oxygen dissolved in the water column through gills. This breathing mechanism makes it difficult for them to develop in environments with low concentrations of dissolved oxygen and high concentrations of organic compounds. Given that tolerant taxa to organic pollution are primarily associated to low flows and standing waters, and sensitive taxa are predominant in rheophilic and well-oxygenated environments, the modelled spatial variation of IASPT and LIFE was quite similar and influenced by the same environmental variables. Moreover, all the families included within the LIFE GI are ephemeropterans, plecopterans and trichopterans, and thus, the EPT index showed the same spatial patterns

as the IASPT and the LIFE indexes. Following this approach, macroinvertebrate biotic indices were lower in Mediterranean catchments, probably because in this region the flow reduction in summer is more pronounced than in Atlantic catchments. During the dry season (summer), in many Mediterranean streams the relative area of pools and other lentic habitats increases notably. Odonata, Coleoptera, Heteroptera and Diptera are usually the most abundant insects in this type of habitats, however, pools do not host rheophilic macroinvertebrates (i.e. EPT), due to more lentic environmental conditions, characterised by slow flows and standing waters, higher temperatures, lower concentration of dissolved oxygen, etc. (Sánchez-Montoya et al., 2009). Regarding predictor variables, partial dependence plots indicate that IASPT, EPT and LIFE values decreased abruptly when the mean annual nitrate concentration exceeded the value of 4 mg/l. Although this concentration of nitrates does not seem to be high enough to cause mortality of sensitive taxa (see Camargo et al., 2005), it may act as an early signal that promotes their migration by drift processes (Camargo and Ward, 1992, 1995). The concentration of nitrates in the study area was mainly related to agricultural uses in the upstream drainage catchment (see Álvarez-Cabria et al., 2016). Thus, the relationship between macroinvertebrate biotic indices and nitrate concentration could be reflecting an underlying relationship between these biotic indices and the predominant land-use in the catchment. As described by Piggott et al. (2012) & (2015) agricultural streams are under the influence of multiple stressors, which can be synthetized in: (1) Nutrient enrichment, produced by the use of fertilisers, (2) Higher load of fine sediments on riverbed, caused by soil erosion and (3) Increment in water temperature, related with deforestation, particularly in the riparian area. There may be interactions and synergies between the different stressors that increase their individual effect on river biological communities (e.g. diatoms, macroinvertebrates) and processes (e.g. leaf litter decomposition). Therefore, we believe that the decline of macroinvertebrate

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

A)

B)

C)

D)

315

Fig. 4. Result of the model for the LIFE index (A) and partial dependence plots with the functions for the most important predictors: (B) aquatic vegetation index, (C) mean annual nitrate concentration and (D) flow velocity index.

biotic indices associated to nitrate concentration could be due to these types of interactions and synergies, commonly related with agricultural landscapes. Another important predictor identified by the models was the aquatic plant community structure and composition. Macroinvertebrate biotic indices were positively correlated with river reaches dominated by mosses and lichens, while these biotic indices decreased in reaches dominated by submerged macrophytes (e.g. Potamogeton spp., Ranunculus spp.). Changes in fluvial plant communities are highly dependent on the position in catchment. For example, Manolaki and Papastergiadou (2013) described bryophyte and pteridophyte as indicator taxa of fast flowing sites from headwaters of Mediterranean catchments, while several macrophytes, such as Potamogeton pectinatus, were indicators of lower catchment parts. Finally, these authors described 4 different aquatic vegetation groups in mid-sites sections. The segregation of these four groups was mainly due to hydraulic conditions (fast/slow flowing waters), water quality (high/low nutrient concentration) and sediment size (fine/stony). Altered mid-sites, with high nutrient concentration, or affected by water abstraction, with dominance of slow flows and fine sediments (mud and sands), were dominated by typical plant communities from downstream sites, while mosses and liverworts remained in non-altered mid-sites (see also Ot'ahel'ova et al., 2007). The relationship between macroinvertebrate biotic indices and aquatic plant communities seems to derived from the fact that both communities, macroinvertebrates and aquatic plants, change their natural distribution in the catchment affected in a similar way by human alterations (water quality and hydromophological impacts; Álvarez-Cabria et al., 2010). Thus, macroinvertebrate taxa and aquatic plants characteristics from fast-flowing and well-oxygenated sites (e.g. EPT taxa and mosses/lichens) are replaced by taxa adapted to lentic environments with a higher nutrient load and lesser dissolved oxygen concentration (e.g. dipterans and macrophytes). In this regard, models

also recognised the flow velocity (FI) as one of the four most important predictors in the spatial variation of the modelled macroinvertebrate biotic indices (Table 2). 4.2. Fish biotic index; percentage of salmonid biomass The percentage of salmonid biomass was mainly affected by the same environmental variables as those seen in macroinvertebrate biotic indices (MN_NO3, MN_TEMP and FI). Following the results of partial dependence plots, salmonids and macroinvertebrates responded similarly to the increase in the concentration of nitrates, with an important decline in the four biotic indices when nitrates exceed the mean annual concentration of 4 mg/l (Figs. 2B, 3B, 4C and 5B). Salmonids also decreased in warmer and more lentic rivers (mean annual water temperature N 12 °C; summer flow velocity b 0.4 m/s). These types of rivers are more abundant in the Mediterranean region, where cyprinids (e.g. Cyprinus carpio) and alien species (e.g. Esox lucius) increase their importance in fish communities in comparison with Atlantic rivers and streams (Doadrio, 2002). Moreover, the natural distribution range of one of the two autochthonous salmonid species from our study area, the Atlantic salmon (S. salar), does not include the Mediterranean biogeographical region of Spain, and only inhabits catchments draining into the Cantabrian Sea (Atlantic Ocean; González-Ferreras et al., 2016). The importance of water temperature and flow velocity on salmonid populations has been largely reported (e.g. Filipe et al., 2013; Jonsson and Jonsson, 2011). In addition, both predictors, water velocity and temperature, are inter-related, because standing waters and lentic environments remain warmer than more rheophilic rivers and streams (Caissie, 2006). Following the revision carried out by Fenkes et al. (2016), flow reduction and thermal stress lead to important changes in the migration timing and reproductive success of salmon. For instance, thermal stress is related with damages in the sperm's DNA and

316

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

A)

C)

B)

D)

Fig. 5. Result of the modelled percentage of salmonid biomass (A) and partial dependence plots with the functions for the most important predictors: (B) mean annual nitrate concentration, (C) flow velocity index and (D) mean annual water temperature.

decreases the sperm longevity of salmonids. Thermal stress also generates energy loss on both, salmon and trout, because the oxygen demand increases at warmer temperatures (e.g. cardiorespiratory system; Fenkes et al., 2016). Energy loss is also related with increments in the post-reproductive mortality rates on salmonids. Another important effect of water temperature on salmonids is related with the embryological development. Thus, Jonsson and Jonsson (2011) reported temperature limits for salmon embryo development between 0 and 16 °C, and between 0 and 14 °C for trout embryo development. 4.3. Large scale implications In this study, we selected various biological communities and biotic indices to evaluate the ecological integrity of river ecosystems in a large area (110,000 km2), however, the spatial patterns followed by these biotic indices were quite similar and influenced in all cases by a similar subset of environmental predictors: nitrate concentration, water temperature and water velocity. Surprisingly, the variability of the modelled biotic indices was not influenced by urban or industrial sewage outflows. Distance to sewage outflows was one of the most important predictors in determining phosphate concentration in the studied area (Álvarez-Cabria et al., 2016). This could indicate that the investment in the construction of new and better water treatment plants during the latest decades in order to reduce the impact of point-source effluents (urban and industrial), might have been effective on reducing the effects on the river biota. We believe that this result is also related to the inclusion of river reaches that are from contrasting climates (Atlantic and Mediterranean) and catchments with very different land use configuration (agriculture, forestry, pasture) in the data set. Thus, these results illustrate how the effects of global change (associated to climate-hydrological and land use change drivers) may cause major

changes in the studied biological communities and in the biotic indices derived from them. The effects of these global stressors must be addressed from a watershed perspective (i.e. integrated catchment management). For instance, changes in catchment land uses, with replacement of forested areas by high-yield agriculture, is a determinant factor in the increment of water temperature and nitrate concentration by run-off processes (Álvarez-Cabria et al., 2016). In this regard, priorisation of river reaches for recovery, following a catchment perspective (Benda et al. 2011), in conjunction with the implementation of green infrastructures (e.g. riparian belts, hillside vegetation, or floodplain recovery) might be a key management strategy in these Mediterranean and Atlantic catchments to adapt to climate change and to meet the EU environmental standards (i.e. WFD & Habitat Directive). Several studies have indicated that enhancing vegetation at catchment and reach scale could be implemented to (1) reduce nutrient runoff processes, (2) buffer water temperature variability, (3) enhance the river ecosystem habitability and (4) increase the quantity and quality of feeding resources for aquatic organisms, such as the concentration of coarse particulate organic matter (e.g. Binkley and Fisher, 2013; Piggott et al., 2015; Álvarez-Cabria et al., 2016; Bay et al., 2016). 5. Conclusions Our models have revealed as a useful tool for the estimation of the ecological status of entire river networks. They are capable to analyse the response of biotic indices against multiple stress factors, acting at catchment and reach scales, and to quantify the importance of those factors (natural and anthropogenic) in determining the structure and composition of biological communities and the variability of the associated biotic indices. Moreover, the use of virtual watersheds and synthetic river networks to develop our models, allows the partition of large

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

river networks in small river segments (b1 km length), reducing the spatial uncertainty of the water body evaluation approach (N 10 km2 catchments surface). The results of our models indicated that river biological communities are mainly affected by processes that take place at a catchment scale. Thus, the solutions for river ecosystem impairment must be taken from an integrated catchment management approach, and the implementation of recovery measures that help resolve the impacts of different human pressures and their concomitant effects. Acknowledgements This study was partly funded with the RIVERLANDS (Ref. BIA201233572) and HYDRA (Ref. BIA2015-71197) projects, granted by the Spanish Ministerio de Economía y Competitividad (MEC). José Barquín is supported by a Ramon y Cajal Grant (ref: RYC-2011-08313; MEC). Alexia Mª González-Ferreras is supported by a PhD Grant (Ref: BES2013-065770; MEC). We would like to thank the CHC, the CHE, the URA and the Government of Cantabria for providing macroinvertebrate, fish and water quality databases. References Alba-Tercedor, J., Sánchez-Ortega, A., 1988. Un método rápido y simple para evaluar la calidad biológica de las aguas corrientes basado en el de Hellawell (1978). 4. Limnetica, pp. 51–56. Álvarez-Cabria, M., Barquín, J., Juanes, J.A., 2010. Spatial and seasonal variability of macroinvertebrate metrics: do macroinvertebrate communities track river health? Ecol. Indic. 10:370–379. http://dx.doi.org/10.1016/j.ecolind.2009.06.018. Álvarez-Cabria, M., Barquín, J., Juanes, J.A., 2011. Microdistribution patterns of macroinvertebrate communities upstream and downstream of organic effluents. Water Res. 45:1501–1511. http://dx.doi.org/10.1016/j.watres.2010.11.028. Álvarez-Cabria, M., Barquín, J., Peñas, F.J., 2016. Modelling the spatial and seasonal variability of water quality for entire river networks: relationships with natural and anthropogenic factors. Sci. Total Environ. 545-546:152–162. http://dx.doi.org/10.1016/ j.scitotenv.2015.12.109. Andonie, R., 2010. Extreme data mining: inference from small datasets. Int. J. Comput. Commun. Control 5:280–291. http://dx.doi.org/10.15837/ijccc.2010.3.2481. AQEM, 2002. Manual for the Application of the AQEM System. A Comprehensive Method to Asses European Streams Using Benthic Macroinvertebrates. Developed for the Purpose of the Water Framework Directive Version 1.0. February 2002. Arimoro, F.O., Odume, O.N., Uhunoma, S.I., Edegbene, A.O., 2015. Anthropogenic impact on water chemistry and benthic macroinvertebrate associated changes in a southern Nigeria stream. Environ. Monit. Assess. 187:14. http://dx.doi.org/10.1007/s10661014-4251-2. Barquín, J., Snelder, T.H., Booker, D., Álvarez-Cabria, M., Peñas, F.J., Fernández, D., 2011. Modelling physical characteristics of habitats from river reaches to entire river networks in northern Spain. Seventh Symposium for European Freshwater Sciences (SEFS), Gerona, Spain. Barquín, J., Benda, L.E., Villa, F., Brown, L.E., Bonada, N., Vieites, D.R., Battin, T.J., Olden, J.L., Hughes, S.J., Gray, C., Woodward, G., 2015. Coupling virtual watersheds with ecosystem services assessment: a 21st century platform to support river research and management. WIREs Water. http://dx.doi.org/10.1002/wat2.1106. Bay, Y., Jiang, B., Alatalo, J.M., Zhuang, C.W., Wang, X.Y., Cui, L.J., Xu, W.H., 2016. Impacts of land management on ecosystem service delivery in the Baiyangdian river basin. Environ. Earth Sci. 75:258. http://dx.doi.org/10.1007/s12665-015-4831-7. Bejarano, M.D., Marchamalo, M., Garcia de Jalón, D., González del Tánago, M., 2010. Flow regime patterns and their controlling factors in the Ebro basin (Spain). J. Hydrol. 385: 323–335. http://dx.doi.org/10.1016/j.jhydrol.2010.03.001. Benda, L., Miller, D., Barquín, J., 2011. Creating a catchment scale perspective for river restoration. Hydrol. Earth Syst. Sci. 15:2995–3015. http://dx.doi.org/10.5194/hess-152995-2011. Binkley, D., Fisher, R.F., 2013. Ecology and Management of Forest Soils. Wiley-Blackwell (347 pp). Bonada, N., Prat, N., Resh, V.H., Statzner, B., 2006. Developments in aquatic insect biomonitoring: a comparative analysis of recent approaches. Annu. Rev. Entomol. 51: 495–523. http://dx.doi.org/10.1146/annurev.ento.51.110104.151124. Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.I., 1984. Classification and Regression Trees. Wadsworth, Belmont. Burdon, F.J., McIntosh, A.R., Harding, J.S., 2013. Habitat loss drives threshold response of benthic invertebrate communities to deposited sediment in agricultural streams. Ecol. Appl. 23:1036–1047. http://dx.doi.org/10.1890/12-1190.1. Caissie, D., 2006. The thermal regime of rivers: a review. Freshw. Biol. 51:1389–1406. http://dx.doi.org/10.1111/j.1365-2427.2006.01597.x. Camargo, J.A., Ward, J.V., 1992. Short-term toxicity of sodium nitrate (NaNO3) to non-target freshwater invertebrates. Chemosphere 24:23–28. http://dx.doi.org/10.1016/ 0045-6535(92)90563-7. Camargo, J.A., Ward, J.V., 1995. Nitrate (NO3-N) toxicity to aquatic life: a proposal of safe concentrations for two species of Nearctic freshwater invertebrates. Chemosphere 31:3211–3216. http://dx.doi.org/10.1016/0045-6535(95)00182-8.

317

Camargo, J.A., Alonso, A., Salamanca, A., 2005. Nitrate toxicity to aquatic animals: a review with new data for freshwater invertebrates. Chemosphere 58:1255–1267. http://dx. doi.org/10.1016/j.chemosphere.2004.10.044. Clapcott, J.E., Goodwin, E.O., Harding, J.S., 2016. Identifying catchment-scale predictors of coal mining impacts on New Zealand stream communities. Environ. Manag. 57: 711–721. http://dx.doi.org/10.1007/s00267-015-0627-5. Colwell, R.K., 1974. Predictability, constancy, and contingency of periodic phenomena. Ecology 55:1148–1153. http://dx.doi.org/10.2307/1940366. Demars, B.O.L., Potts, J.M., Tremolieres, M., Thiebaut, G., Gougelin, N., Nordmann, V., 2012. River macrophyte indices: not the holy grail! Freshw. Biol. 57:1745–1759. http://dx. doi.org/10.1111/j.1365-2427.2012.02834.x. Doadrio, I., 2002. Atlas y Libro Rojo de los Peces Continentales de España. Madrid, Spain (374 pp). Elith, J., Leathwick, J.R., Hastie, T., 2008. A working guide to boosted regression trees. J. Anim. Ecol. 77:802–813. http://dx.doi.org/10.1111/j.1365-2656.2008.01390.x. Elósegui, A., Pozo, J., 1994. Spatial versus temporal variability in the physical and chemical characteristics of the Agüera stream (Northern Spain). Acta Oecol. 15, 543–559. Erba, S., Pace, G., Demartini, D., Di Pasquale, D., Dörflinger, G., Buffagni, A., 2015. Land use at the reach scale as a major determinant for benthic invertebrate community in Mediterranean rivers of Cyprus. Ecol. Indic. 48:477–491. http://dx.doi.org/10.1016/j. ecolind.2014.09.010. ESRI, 2011. ArcGIS Desktop: Release 10. Environmental Systems Research Institute, Redlands, CA. European Commission, 2000. Directive 2000/60/EC of the European Parliament and of the Council-Establishing a Framework for Community Action in the Field of Water Policy, Brussels, Belgium. Extence, C.A., Balbi, D.M., Chadd, R.P., 1999. River flow indexing using British benthic macroinvertebrates: a framework for setting hydroecological objectives. Regul. Rivers Res. Manag. 15, 543–574 (DOI: 10.1002/(SICI)1099-1646(199911/12)15:6b545:: AID-RRR561N3.0.CO;2-W). Fenkes, M., Shiels, H.A., Fitzpatrick, J.L., Nudds, R.L., 2016. The potential impacts of migratory difficulty, including warmer waters and altered flow conditions, on the reproductive success of salmonid fishes. Comp. Biochem. Physiol. A Mol. Integr. Physiol. 193:11–21. http://dx.doi.org/10.1016/j.cbpa.2015.11.012. Fernández, D., Barquín, J., Álvarez-Cabria, M., Peñas, F.J., 2014. Land-use coverage as an indicator of riparian quality. Ecol. Indic. 41:165–174. http://dx.doi.org/10. 1016/j.ecolind.2014.02.008. Filipe, A.F., Markovic, D., Pletterbauer, F., Tisseuil, C., De Wever, A., Schmutz, S., Bonada, N., Freyhof, J., 2013. Forecasting fish distribution along stream networks: brown trout (Salmo trutta) in Europe. Divers. Distrib. 19:1059–1071. http://dx.doi.org/10.1111/ddi. 12086. González-Ferreras, A.M., Barquín, J., Peñas, F.J., 2016. Integration of habitat models to predict fish distributions in several watersheds of Northern Spain. J. Appl. Ichthyol. 32: 204–216. http://dx.doi.org/10.1111/jai.13024. Hellawell, J.M., 1978. Biological Surveillance of Rivers. Water Research Centre (332 pp). Jonsson, B., Jonsson, N., 2011. Ecology of Atlantic Salmon and Brown Trout. Habitat as a Template for Life Stories. Springer (680 pp). Kuhn, S., Egert, B., Neumann, S., Steineck, C., 2008. Building blocks for automated elucidation of metabolites: machine learning methods for NMR prediction. BMC Bioinf. 9: 400. http://dx.doi.org/10.1186/1471-2105-9-400. Leathwick, J.R., Elith, J., Hastie, T., 2006. Comparative performance of generalized additive models and multivariate adaptive regression splines for statistical modelling of species distributions. Ecol. Model. 199:188–196. http://dx.doi.org/10.1016/j.ecolmodel. 2006.05.022. Leps, M., Tonkin, J.D., Dahmb, V., Haase, P., Sundermann, A., 2015. Disentangling environmental drivers of benthic invertebrate assemblages: the role of spatial scale and riverscape heterogeneity in a multiple stressor environment. Sci. Total Environ. 536:546–556. http://dx.doi.org/10.1016/j.scitotenv.2015.07.083. Lowicki, D., 2012. Prediction of flowing water pollution on the basis of landscape metrics as a tool supporting delimitation of nitrate vulnerable zones. Ecol. Indic. 23:27–33. http://dx.doi.org/10.1016/j.ecolind.2012.03.004. Manolaki, P., Papastergiadou, E., 2013. The impact of environmental factors on the distribution pattern of aquatic macrophytes in a middle-sized Mediterranean stream. Aquat. Bot. 104:34–46. http://dx.doi.org/10.1016/j.aquabot.2012.09.009. Merritt, R.W., Cummins, K.W., 1996. An Introduction to the Aquatic Insects of North America. Kendall/Hunnt Publishing Company, Dubuque, USA. Miller, D., 2003. Programs for DEM Analysis. Earth System Institute, Seattle, WA (38 pp). Olson, J.R., Hawkins, C.P., 2012. Predicting natural base-flow stream water chemistry in the western United States. Water Resour. Res. 48:1–19. http://dx.doi.org/10.1029/ 2011WR011088. Ot'ahel'ova, H., Valachovic, M., Hrivnak, R., 2007. The impact of environmental factors on the distribution pattern of aquatic plants along the Danube River corridor (Slovakia). Limnologica 37:290–302. http://dx.doi.org/10.1016/j.limno.2007.07.003. Peñas, F.J., 2014. Classification of the natural flow regime and prediction of hydroecological characteristics in the northern third of the Iberian Peninsula. Thesis. University of Cantabria, Santander, Spain. Piggott, J.J., Lange, K., Townsend, C.R., Matthaei, C.D., 2012. Multiple stressors in agricultural streams: a mesocosm study of interactions among raised water temperature, sediment addition and nutrient enrichment. PLoS One 7 (11), e49873. http://dx.doi. org/10.1371/journal.pone.0049873. Piggott, J.J., Townsend, C.R., Matthaei, C.D., 2015. Climate warming and agricultural stressors interact to determine stream macroinvertebrate community dynamics. Glob. Chang. Biol. 21:1887–1906. http://dx.doi.org/10.1111/gcb.12861. Pilière, A., Schipper, A.M., Breure, T.M., Posthuma, L., de Zwart, D., Dyer, S.D., Huijbregts, M.A.J., 2014. Unraveling the relationships between freshwater invertebrate assemblages and

318

M. Álvarez-Cabria et al. / Science of the Total Environment 577 (2017) 308–318

interacting environmental factors. Freshwater Sci. 33:1148–1158. http://dx.doi.org/10. 1086/677898. R Development Core Team, 2011. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. (ISBN 3-900051-07-0). http://www.R-project.org/. Raven, P.J., Holmes, N.T.H., Dawson, F.H., Everard, M., 1998. Quality assessment using river habitat survey data. Aquat. Conserv. Mar. Freshwat. Ecosyst. 8, 477–499 (DOI: 10.1002/(SICI)1099-0755(199807/08)8:4b477::AID-AQC299N3.0.CO;2-K). Rivas-Martínez, S., Penas, A., Díez, T.E., 2004. Bioclimatic Map of Europe, Bioclimates. University of León, Cartographic Service, León, Spain. Sánchez-Montoya, M.M., Vidal-Abarca, M.R., Puntí, T., Poquet, J.M., Prat, N., Rieradevall, M., Alba-Tercedor, J., Zamora-Muñoz, C., Toro, M., Robles, S., Álvarez, M., Suárez, M.L., 2009. Defining criteria to select reference sites in Mediterranean streams. Hydrobiologia 619:39–54. http://dx.doi.org/10.1007/s10750-008-9580-0. Schmutz, S., Jurajda, P., Kaufmann, S., Lorenz, A.W., Muhar, S., Paillex, A., Poppe, M., Wolter, C., 2016. Response of fish assemblages to hydromorphological restoration

in central and northern European rivers. Hydrobiologia 769:67–78. http://dx.doi. org/10.1007/s10750-015-2354-6. Snelder, T.H., Lamoroux, N., Pella, H., 2011. Empirical modelling of large scale patterns in river bed surface grain size. Geomorphology 127:189–197. http://dx.doi.org/10.1016/ j.geomorph.2010.12.015. Tachet, H., Richoux, P., Bournaud, M., Usseglio-Polatera, P., 2000. Inverte'bre's d´ eau douce syste'matique, biologie, e´ cologie. CNRS Editions, Paris, France (587 pp). Teittinen, A., Taka, M., Ruth, O., Soininen, J., 2015. Variation in stream diatom communities in relation to water quality and catchment variables in a boreal, urbanized region. Sci. Total Environ. 530–531:279–289. http://dx.doi.org/10.1016/j.scitotenv.2015.05. 101. UNE-EN 14011:2003. Calidad del agua. Muestreo de peces con electricidad (Water quality - Sampling of fish with electricity), 1998. Unwin, M., Larned, S.T., 2014. Statistical Models, Indicators and Trend Analyses for Reporting National-Scale River Water Quality. NIWA, Christchurch.

Suggest Documents