Indices and predictive models

Ecological Indicators 11 (2011) 379–388

Contents lists available at ScienceDirect

Ecological Indicators journal homepage: www.elsevier.com/locate/ecolind

Original article

Choosing the best method for stream bioassessment using macrophyte communities: Indices and predictive models Francisca C. Aguiar a,∗ , Maria João Feio b , Maria Teresa Ferreira a a b

Forest Research Centre, High Institute of Agronomy, Tapada da Ajuda, 1349-017 Lisboa, Portugal IMAR-CMA and Department of Zoology, University of Coimbra, Largo do Marquês de Pombal, 3004-517 Coimbra, Portugal

a r t i c l e

i n f o

Article history: Received 19 November 2009 Received in revised form 12 May 2010 Accepted 12 June 2010 Keywords: Macrophytes River Bioassessment Quality index Predictive models

a b s t r a c t The bioassessment and monitoring of the ecological status of rivers using macrophytes has gained new momentum since macrophytes were recognised as biological quality elements for the implementation of the European Water Framework Directive (WFD; EU/2000/60). Our objectives were to test the suitability of two predictive modelling approaches to macrophyte communities as a tool for water quality assessment, and to compare their performance with other more common approaches—the use of macrophytes as indicators of the trophic status of rivers and multimetric indices. We used floristic and environmental data that were collected in the spring of 2004 and 2005 from around 400 sites on rivers across mainland Portugal, western Iberia. We build two predictive models: MACPACS (MACrophyte Prediction And Classification System) and MAC (Macrophyte Assessment and Classification) based on RIVPACS and the BEAST methods, respectively. Whereas MACPACS is derived from taxa occurrence data, MAC uses a quantitative measure of taxa abundance. Both models showed good performance in predicting reference sites to the correct group and low rate of misclassification errors. However, they performed differently. MAC depicts a reliable response to the overall human-mediated degradation of fluvial systems, as does the multimetric index (RVI, Riparian Vegetation Index), but MACPACS presented only a poor correlation with the Global Human Disturbance Index and with the nutrients input. The incorporation of abundance data in vegetation predictive models appears to be particularly important to the detection of high levels of degradation. The values for correlations with physical–chemical pressure variables were lower than expected for MTR (Mean Trophic Rank) due to an insufficient number of scoring species found in Portuguese fluvial systems. Our results suggest that the most effective methods for bioassessment in Mediterranean-type rivers are either the RVI or the MAC predictive model. © 2010 Elsevier Ltd. All rights reserved.

1. Introduction The translation of floristic data into ecological-based systems for the assessment of river quality has been a primary challenge for aquatic plant experts and conservation scientists since the late 1980s. Numerous aquatic plant-bioassessment methods have been developed using diverse aspects of plant and vegetation attributes, such as the richness and abundance of species assemblages (Haslam and Wolseley, 1987; Lange and van Zon, 1983; Stromberg et al., 2006), vegetation structure (e.g. González del Tánago and García de Jalón, 2006), species attributes and functional groups (e.g. Brazner et al., 2007; Ferreira et al., 2005; Mack, 2007; Rothrock et al., 2008), and the use of macrophyte species as indicators of trophic status (e.g. Haury et al., 2006; Holmes et al., 1999; Schneider and Melzner,

∗ Corresponding author. Tel.: +351 213653492; fax: +351 213653338. E-mail address: [email protected] (F.C. Aguiar). 1470-160X/$ – see front matter © 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.ecolind.2010.06.006

2003). Also, different types of data treatment have been used to relate floristic changes and human disturbance, including multivariate analysis (e.g. Dodkins et al., 2005; Schaumburg et al., 2004), classification and decision trees (e.g. Cohen et al., 2005), and univariate methods (e.g. Hering et al., 2006). Having said this, most of the operational bioassessment methods using river plants are based on sensitive species (i.e. indicator indices) or on functional groups (i.e. multimetric indices). The new European water legislation—the Water Framework Directive (WFD; European Comission, 2000)—includes macrophytes in the biological quality element of aquatic flora. This has promoted the development of a number of national assessment methods and a more intensive exchange of information between countries and experts than had been the case with previous national and transnational programmes. In Portugal, the plantbased assessment methods (e.g. Aguiar et al., 2004; Espírito-Santo et al., 2000; Ferreira et al., 2004) that were initially proposed were simple and user-friendly, based on field expertise, and lacked a river

380

F.C. Aguiar et al. / Ecological Indicators 11 (2011) 379–388

typological framework and a reference condition approach sensu Reynoldson et al. (1997). Subsequently, multivariate and multimetric approaches were used to overcome these limitations. Ferreira et al. (2002) used the Bray–Curtis multivariate distance and an overall canonical procedure to identify reference conditions and measure deviation due to perturbation; in another study the Iberian Multimetric Plant Index (IMPI; Ferreira et al., 2005), a partial canonical correspondence analysis of the floristic data with anthropogenic variables, was used to account for environmental differences. Both the aforementioned studies were conducted on a small scale and restricted to semi-arid southern rivers. The quality of the results then led IMPI to be applied to nationwide data for the implementation of the WFD in order to obtain a spatial upgraded index (Riparian Vegetation Index, RVI; Aguiar et al., 2009). Predictive models are an alternative approach to indices, and comprise a sequence of statistical steps with the aim of comparing the observed biota at a test site with the expected/predicted biota from a set of sites representing the reference condition for a given area. These models have been developed and applied worldwide in the ecological assessment of streams, mainly in relation to macroinvertebrate communities (e.g. Feio et al., 2007a, 2009a; Kokeˇs et al., 2006; Poquet et al., 2009; Reynoldson et al., 1995; Simpson and Norris, 2000; Wright, 1995), but also with diatoms (Chessman et al., 1999; Feio et al., 2007b, 2009b; Mazor et al., 2006; Philibert et al., 2006), and fishes (Joy and Death, 2002; Mugodo et al., 2006). However, as far as we know little effort is being made to develop predictive modelling with macrophytes for stream-quality assessment. In the light of these studies it is important to analyse the performance of the aforementioned approaches. With the present study we therefore aimed to: (i) test the suitability of two predictive modelling approaches to macrophyte communities as a water-quality assessment tool; and (ii) compare their performance with other more common approaches using river plants—the use of macrophytes as indicators of trophic status and multimetric indices. We applied the methods originally used for macroinvertebrates to develop the MAC (Macrophyte Assessment and Classification) and the MACPACS (MACrophyte Prediction And Classification System), which are respectively based on the BEAST (Reynoldson et al., 1995, 1997) and RIVPACS (Wright, 1995, 2000) methods. We used the Mean Trophic Rank (MTR; Dawson et al., 1999; Holmes et al., 1999), which was originally developed to fulfil the requirements of the Urban and Waste Water Directive in the UK (91/271/EC), and was designed to respond to nutrient enrichment; and as a multimetric approach, we used the Riparian Vegetation Index (Aguiar et al., 2009), a typological-adapted index, based on structural and functional components of the riparian and aquatic vegetation.

2. Materials and methods 2.1. Site selection and sampling Data on macrophyte species composition and disturbance variables was collected from around 400 sites on rivers across mainland Portugal, western Iberia (Fig. 1). Apart from a few mountainous areas and the occidental northern part of the country, Portugal has a Mediterranean climate that is characterized by a strong seasonal and inter-annual variability of rainfall patterns, with mild winters and dry summers. Due to a water deficit in the summer season and to historical features of human occupation, rivers and riparian woods in Iberia have been impacted for millennia, and pristine locations no longer exist. Major human disturbances are related to water diversion and regulation and with deforestation and agricultural land use of the catchment (Hooke, 2006).

Fig. 1. Location of all catchments and main rivers in Portugal with respective reference (a) and test sites (b) and localization of Portugal in Europe.

Both reference and test sites (i.e. impacted sites) were previously selected following a preliminary screening using digital databases from the Water Institute (INAG IP) and the Portuguese Water Resources Information System (http://www.snirh.pt), expert judgement, and prospective field campaigns. The reference sites met the common criteria of: (i) good chemical quality (nitrate, nitrite, phosphates, ammonia, pH, BOD5, COD)—i.e., values allocated to the A or B categories of water for multiple human uses (INAG IP, http://snirh.inag.pt/snirh/ dados sintese/qual ag anual/classificacao.html); (ii) minimal changes in the riparian zone; (iii) no signs of recent changes in the channel morphology and all expected habitats present; (iv) low levels of urbanization and industrial activities in the catchment area; (v) minimum impacts on the natural hydrological regime; and (vi) low levels of fine sediment load. The quality status of the sites was pre-classified using a composite pressure score (Global Human Disturbance Index, GHD), 1–5 ranked, from four variables at the segment level and four variables at the site level (see Table 1), as used in Pont et al. (2006). A river segment is defined as 1 km for small rivers (catchment 99.9% ellipse

Class 1: 1.290–0.754 Class 2: 0.753–0.566 Class 3: 0.564–0.377 Class 4: 0.377–0.189 Class 5: 0.189–0

% of reference-test sites assigned to class 1 (equivalent to reference condition)

93%

86%

382

F.C. Aguiar et al. / Ecological Indicators 11 (2011) 379–388

reach (e.g. woody debris, rocks) in the channel and along the banks, and on trees at no more than 0.5 m above ground level, and were then identified at the Herbarium of the Natural History Museum of the Lisbon University (LISU). The sampling protocol is available at http://dqa.inag.pt/dqa2002/port/docs apoio/nacionais.html. Simultaneously, a similar procedure was performed for the overall fluvial corridor (i.e. including the channel and both margins). Sampled areas ranged from 400 to 2000 m2 . Taxonomic identification of vascular species that could not be recognized in loco was carried out at the João Carvalho and Vasconcellos Herbarium at the School of Agriculture, Lisbon (LISI). Macroalgae were not recorded. 2.2. Macrophyte-based indices and predictive models We applied two macrophyte-based indices—the MTR (Dawson et al., 1999; Holmes et al., 1999), and the Riparian Vegetation Index, RVI (Aguiar et al., 2009)—and developed two predictive models—MAC, and MACPACS. 2.2.1. MTR MTR is an index based on indicator species that was initially developed for the assessment of the trophic status of British rivers, and has recently been applied and tested in several European countries (Szoszkiewicz et al., 2006). The system is based on the occurrence and abundance of indicator macrophyte species (macroalgae, vascular species, and bryophytes). Each indicator species is scored from 1 to 10—Species Trophic Rank, STR—according to its response to eutrophication. High STR values correspond to species that are intolerant of eutrophication, while low values indicate that the plant is either tolerant to polluted waters or has no preference. The cover value of the species is recorded on a nine-point scale to determine the Species Cover Value (SCV). Multiplying the STR of each ‘scoring’ species by its SCV gives a cover value score (CVS). The final MTR result is calculated by dividing the sum of CVS by the sum of SCV, multiplied by 10. The resulting MTR score lies in the range 10–100 (the lower the score, the more eutrophic the site). Methodological details, including the list of indicator species and the respective STR, can be found in Dawson et al. (1999).

Fig. 2. Simplified dendrograms obtained from the hierarchical clustering analysis of the: (a) species occurrence, and (b) species abundance. Alpha-numerical acronyms represent reference groups used for MAC and MACPACS models.

2.2.2. RVI The RVI was developed in the context of the WFD in order to evaluate and monitor the ecological river quality of Portuguese rivers. It is a structural-based index, and uses functional patterns and compositional attributes of the overall fluvial vegetation that reflect ecosystem processes. Plant metrics were selected from various metric categories (e.g. life cycle and propagation, nutritional resources, vegetation structure) to reflect the multiple dimensions of the biological ecosystem and of disturbance. The RVI has different metrics and ranges for the North and South regions. Scoring metrics are estimated or evaluated using abundance cover, proportion or number of species in functional groups (e.g. aliens, nitrophyllous, ruderals), species attributes (e.g. life form, reproduction strategies), and indicator taxa (e.g. Carex elata ssp. reuterana). Metrics were transformed to dimensionless numbers for aggregation and combined in an additive index. The RVI for a site was obtained by the sum of the quality scores of all the metrics, less the total number of metrics for the region. Each metric scored five points if it had Table 3 Max group and respective maximum Indicator Value (IVmax > 0.30) for each method. Max group is the group where the maximum Indicator value occurs (p value < 0.05). Species are ordered by increasing values of IVmax of MAC reference groups. Hydric group: hid—hydrophyte; hel—helophyte; hyg – hygrophytes. w —woody species; e —Iberian endemism. Characteristic species

Hydricgroup

Max group. IVmax MAC

MACPACS

Scirpoides holoschoenus Cyperus longus Oenanthe crocata Sellaginella denticulata Mentha pulegium Lunularia cruciata Solenopsis laurentia Annograma leptophylla Nerium oleander Flueggea tinctoria Pulicaria paludosa

hel hel hel hyg hyg hyg hyg hyg hyg w hygw,e hyg

A. 62.5 A. 54.5 A. 51.4 A. 49.9 A. 49.5 A. 45.4 A. 43.6 A. 42.3 A. 34.1 A. 32.8 A. 32.0

b1. 54.9 b1. 41.4 a1. 28.4 b1. 56.1 b1. 43.5 b1. 40.1 b1. 31.2 b1. 52.5 b1. 54.9 b1. 51.2 b1. 52.9

Carex elata ssp. reuterana Molinia caerulea Viola palustris ssp. palustris Blechnum spicant Galium broteroanum Hyocomium armoricum Erica arborea Scapania undulata Frangula alnus Polytrichum formosum Pellia epiphylla Potentilla erecta

hel hyg hyg hyg hyg hid hygw hid hygw hyg hid hyg

B. 66.6 B. 65.9 B. 64.2 B. 63.7 B. 61.9 B. 59.1 B. 58.5 B. 55.1 B. 52.1 B. 48.4 B. 44.0 B. 42.2

a2. 41.9 a2. 60.1 a2. 57.8 a2. 44.6 a2. 39.3 a2. 47.2 a2. 47.3 a2. 52.4 a2. 41.0 a2. 29.3 a2. 47.3 a2. 35.3

Paspalum distichum Salix salviifolia

hyg hygw,e

C1. 44.3 C1. 40.8

b2. 26.3 b2. 29.9

Athyrium filix-femina Alnus glutinosa Dactylis glomerata Osmunda regalis Polystichum setiferum Brachypodium sylvaticum Sambucus nigra Hedera hibernica Ranunculus repens

hyg hygw hyg hyg hyg hyg hygw hygw hyg

C2. 66.5 C2. 54.1 C2. 41.2 C2. 40.8 C2. 40.7 C2. 37.2 C2. 35.6 C2. 34.3 C2. 34.0

a1. 45.9 a1. 38.7 a1. 26.7 a2. 32.4 a1. 43.7 a1. 35.0 a1. 28.7 a1. 35.3 a1. 43.3

Salix atrocinerea Leersia oryzoides Tradescantia fluminensis Iris pseudacorus Holcus lanatus Lycopus europaeus Juncus effusus Lythrum salicaria Hypericum undulatum Bidens frondosa

hygw hel hyg hel hyg hel hyg hel hyg hyg

C3. 69.8 C3. 59.4 C3. 58.2 C3. 54.9 C3. 49.8 C3. 41.7 C3. 40.4 C3. 39.5 C3. 37.2 –

a2. 40.0 – – – a1. 39.8 – – b2. 37.7 a1. 43.6 b2. 31.8

F.C. Aguiar et al. / Ecological Indicators 11 (2011) 379–388

a value similar to that expected under reference conditions and at minimum-impacted sites, one point if it had a value similar to that expected with a high level of human disturbances, and three points if it had an intermediate value. The resulting RVI score lies in the range 0–40 for the North region and 0–36 for the South region (the lower the score, the more disturbed the site). Ranges of metrics and quality class values can be consulted in Aguiar et al. (2009). In order to permit comparability, both the MTR and RVI values were transformed into Ecological Quality Ratios (EQR), ecological deviation from non-disturbed situations (WFD Annex V). EQR values were positioned in five ecological quality classes, from High (best of 5 classes) to Bad. The High/Good boundary was established using the median value for the reference sites. The four remaining classes were obtained by dividing the interval limited by the High/Good boundary and the lower extremity of the gradient equally, in accord with Wallin et al. (2003). 2.2.3. MACPACS and MAC MACPACS and MAC are predictive models based on RIVPACS (Wright, 1995, 2000) and the BEAST (Reynoldson et al., 1995, 1997) methods, which were originally developed for macroinvertebrate fauna. These models assess the quality of the sites by the distance between the biological assemblages at a given test site and the biota expected under high quality conditions (i.e. reference sites). Both models are multivariate-based and use reference site grouping, followed by discriminant analyses to select the environmental predictors and to estimate the model performance and calculate the probability that each test site belongs to each reference group. The main differences between the two models are the evaluation of test sites and the type of data. Whereas MACPACS uses the presence/absence of taxa, MAC uses a quantitative measure of abundance. The basic outputs of MACPACS are: (1) the probability that each taxon will occur at a given test site; (2) the expected (predicted) taxa list (with a probability of occurrence >0.5); (3) the Observed/Expected (OE) ratio, which indicates the ecological quality of the site and normally varies between 0 (minimum quality) and ≈1 (maximum quality, observed = expected); in a five-class system where class 1 (high quality; equivalent to reference) represents the range of values of reference sites defined by the interval between the 10th and the 90th percentile, as recommended by Simpson and Norris (2000). The lower classes that represent increasing levels of impairment are the same width as class 1, although the width of the last class is usually less, because it is limited by zero. The expected taxa are calculated by weighing up the probability that a test site will belong to each reference group by the frequency of each taxa in the reference

383

group. More details on this type of model can be found in Wright (1995, 2000), Simpson and Norris (2000) and Feio et al. (2009b). The MAC model, on the other hand, compares a test site with the most similar reference group (given by the probabilities of membership) and assesses the difference between a test site and the reference condition by the distance in a MDS-ordination space. In order to define and categorize the distance into a 4-class system, three Gaussian bivariate probability ellipses are applied to the ordination space, centred in the centroid of the reference sites (90%, 99% and 99.9%, Altman, 1978; Owen and Chmielewski, 1985; Reynoldson et al., 2001; Feio et al., 2007a,b). The multivariate procedures and other detailed steps for model building were provided by Reynoldson et al. (2000) and Feio et al. (2007a) among others. We used the floristic data from the channel and inner banks of the reference sites to build these models. However, taxa that were present at less than 5% of reference sites (n = 9) were eliminated in order to improve the model’s performance. For the MAC model, floristic data was 4th root transformed, and a classification analysis using the Bray–Curtis coefficient of dissimilarity was applied. To form groups of reference sites, clustering analysis (UPGMA; Weighted Pair-Group Means with Arithmetic average) was performed with NTSYSpc2 (F. James Rohlf, Exeter Software, Setauket, NY, USA; Rohlf, 2000). The most consistent reference-grouping structure was selected, observing the R values of the Analysis of Similarities (ANOSIM)—a routine procedure available in the PRIMER software (PRIMER 6, PRIMER-E Ltd, Plymouth, UK; Clarke and Gorley, 2005). For the MACPACS model, a similar procedure was followed with the taxa occurrence data, using Jaccard’s similarity coefficient (Sneath and Sokal, 1973) and the SIMQUAL module. Dufrene and Legendre’s (1997) Indicator Species Analysis approach, performed with PCord 4.25 (McCune and Mefford, 1999), was used to check the consistency of groups and to characterize groups based on taxa common to sites within them. Indicator species of a given group of sites are the most frequently and abundantly found at the majority of sites belonging thereto. This analysis combines proportional abundance of a given species in each group and the relative frequency of occurrence of the species in the group, and ranges from 0 to 100%. The maximum indicator value (IVmax ) for the species across groups is saved as a summary of the overall indicator value of that species (characteristic species, hereafter). Formulae and calculation details are given in McCune and Grace (2002). Statistical significance was calculated using Monte Carlo tests of the observed IVmax for each species, based on 1000 randomizations. For both models, 25 environmental variables were used as potential predictors of the reference group biota and supplied

Table 4 Summary of sampling differences between methods and relative evaluation by: 1 estimate of time-consumption, from 0 (quickest) to 1 (slowest) and 2 outputs’ interpretation of results from • (easiest) to ••• (most difficult). RVI

MTR

MAC

MACPACS

Hydrophytes Helophytes Hygrophytes Terrestrial Channel and riverbanks 1102

Hydrophytes Helophytes

Hydrophytes Helophytes Hygrohpytes

Hydrophytes Helophytes Hygrohpytes

Channel 80

Channel and riverbanks 206

Channel and riverbanks 206

Included Species cover

Included Species cover

Removed Species cover

Removed Occurrence

Effort evaluation1 Field sampling Laboratory/office

1 1

0.4 0.5

0.8 0.8

0.4 0.8

Output appraisal2 Type Interpretation

Quality classes •

Quality classes ••

Banding system ••

O/E ratio Quality classes •••

Sampling Groups

Facies Total number of species (present data) Infrequent specie Biomass estimation

Water chemistry: sediment load segment, Riparian and valley condition: morphological condition site BI & HI: ICM Land use: irrigated cropland site

Water chemistry: TN, TP, ammonia-N Riparian and valley condition: riparian zone segment Water chemistry: TN, TP, ortho-phosphates BI & HI: HQA ≥0.30–0.35

Land use: land use segment, urbanisation segment BI & HI: HQA, QBR

Water chemistry: nutrient organic inputs site, nitrates Water chemistry: nitrates, BOD5 , COD ≥0.35–0.40

Riparian and valley condition: morphological condition site Land use: non-irrigated cropland site, urbanisation segment, BI & HI: QBR, IBMWP

BI & HI: GHD, IBMWP

Water chemistry: nitrates, sediment load segment, nutrient organic inputs site, ortho-phosphates, BOD5 , COD Land use: land use segment, non-irrigated cropland site BI & HI: ICM

Riparian and valley condition: riparian zone segment BI & HI: GHD

MACPACS

Water chemistry: nutrient organic inputs site, sediment load segment Riparian and valley condition: riparian zone segment Land use: land use segment BI & HI: ICM

More than one thousand macrophyte species from 124 families were identified, including 177 woody species. Around 600 samples of bryophytes were collected from all the existing channel and bank microhabitats, resulting in 201 mosses and liverworts from 50 families. A high proportion of terrestrial species was found (58.5% of the vascular flora), and more than 75% were found in less than 5% of the surveys. We listed one hundred alien species, all of which were vascular, and 71 endemisms (European, Iberian, and Lusitanian). Some alien species, such as Acacia sp., the knotgrass (Paspalum distichum), the giant reed (Arundo donax), the water hyacinth (Eichhornia crassipes), and the parrotfeather (Myriophyllum aquaticum), displayed a high invasive potential. Endemic species were mainly found on riverbanks and were frequently site-specific. Emergent species (helophytes) and aquatic plants (hydrophytes) represented only 12% of the total taxa. From the dataset of 50 helophytes and hydrophytes, which were present in more than 5% of sites, we detected a low richness in per-site terms, at both reference (average = 8.1 ± 4.0) and test sites (average = 9.7± 3.8). The most frequent species (present at more than 40% of sites) were Apium nodiflorum, C. elata ssp. reuterana, Cyperus eragrostis, Cyperus longus, Lythrum salicaria, Lycopus europaeus,

BI & HI: GHD

3.1. Floristic overview

≥0.40–0.50

3. Results

≥0.50

The comparison and evaluation of the performance of the methods included a qualitative evaluation, by expert judgement of the effort spent completing field and laboratory tasks and on the output analysis, and the methods’ response to the various indicators of disturbance (Table 1, disturbance variables) using Spearman Rank Correlations, and the correspondence between increasing levels of impairment and indices and model classes through box-plot analysis. For comparison purposes, the predictive bands and classes obtained in the MAC and MACPACS models were made correspondent to the quality classes used in the WFD. We used STATISTICA version 7 (Statsoft Inc., 2004) in these analyses.

MAC

2.4. Comparison between methods

MTR

Fourteen reference sites spread across the country were left out of the model construction and were used to validate the models’ responses by testing whether the latter assigned them to the reference classes. In the case of the MACPACS model we also followed the method of Linke et al. (2005) proposed for the RIVPACS/AUSRIVAS models, under which a model is considered accurate if the regression line of Observed versus Expected values passes through or close to the origin (between 1.5 and −1.5) and has a slope close to the unit (between 0.85 and 1.15).

RVI

2.3. Model test and validation

Spearman Rank Correlation

to the Stepwise Discriminant Analysis with Jackknife crossvalidation. Each environmental variable was previously subjected to a Kolmogorov–Smirnov normality test, and the non-normal variables were log, arcsine, or square-root transformed. If normal distribution was not achieved, more powerful transformations were applied until the best approximation to normality was reached, as prescribed in Zamora-Munõz and Alba-Tercedor (1996) (Table 2). A complete Discriminant Analysis was then performed on both models, using the selected environmental predictors. All the analytical steps except for the initial reference site grouping procedures were performed using the RIO software (http://rio-imar.uc.pt/).

Water chemistry: nutrient organic inputs site BI & HI: GHD

F.C. Aguiar et al. / Ecological Indicators 11 (2011) 379–388 Table 5 Spearman Rank Correlations (p < 0.05) between the disturbance variables and composite indices and the ecological classes attained by each method. BI & HI: Biological and Hydromorphological Quality Indices (see Table 1 for acronyms description).

384

F.C. Aguiar et al. / Ecological Indicators 11 (2011) 379–388

Oenanthe crocata, and Scirpoides holoschoenus. 3.2. Model building Predictive models were based on 206 species associated with the fluvial system (aquatic, emergent, and hygrophilous species), which were found in more than 5% of the sites. Table 2 summarises the characteristics of the MAC and MACPACS predictive models,

385

and Table 3 presents the species that characterize each reference group in both models. For MAC, five groups of sites were obtained by clustering the species occurrences at the 183 reference sites (Fig. 2a), and validated using the ANOSIM procedure (Global R > 0.6, p < 0.001). However, two small groups of sites (n = 3) were not included in the ensuing analysis, because of their low representativity in the study area (