Skip to main content
Dryad logo

Data from: As time goes by: 20 years of changes in the aquatic macroinvertebrate metacommunity of Mediterranean river networks


Cañedo-Argüelles, Miguel et al. (2021), Data from: As time goes by: 20 years of changes in the aquatic macroinvertebrate metacommunity of Mediterranean river networks, Dryad, Dataset,


Aim: to analyse temporal metacommunity dynamics in river networks in relation to hydrological conditions and dispersal. Location: 15 river reaches from the Llobregat, Besòs and Foix catchments in the North-Eastern Iberian Peninsula. Taxon: aquatic macroinvertebrates belonging to 99 different families. Methods: we sampled aquatic macroinvertebrate communities during spring in 20 consecutive years. We built two environmental distances (one related with water chemistry and another one with river flow regime) and two spatial distances (network distance and topographic distance). Then we used Mantel tests (accounting for spatial autocorrelation) to relate macroinvertebrate dissimilarity with environmental and spatial distances. Additionally, we determined the dry and wet years using the Standardised Precipitation Index (SPI) and we classified macroinvertebrate families based on their ability to fly and to drift. Finally, we ran a linear regression model including the correlation value (r) of each Mantel test as response variable and distance type (environmental or spatial), SPI, dispersal mode, their pairwise interactions and a three-way interaction as predictor variables. Results: metacommunity organization varied over time and it was significantly affected by precipitation, which can be related to river network connectivity. The environmental filters, mainly the flow regime, were generally more important than the spatial filters in explaining community dissimilarity over the study period. However, this depended on the dispersal abilities of the organisms. Network fragmentation due to flow intermittence during the dry years significantly reduced the dispersal capacity of strong aerial dispersers, leading to spatially structured metacommunities. For strong drift dispersers, community dissimilarity patterns were generally best explained by environmental filters regardless of SPI. Main conclusions: a significant temporal variation in metacommunity organization can be expected in highly dynamic systems (e.g. Mediterranean rivers) and it might depend on the dispersal modes and abilities of the organisms, since they determine the response to changes in environmental and landscape filters.


We studied 15 river reaches from the Llobregat, Besòs and Foix catchments in the North-Eastern Iberian Peninsula. The annual precipitation ranged from 500 to 1400 mm, and the average annual temperature from 7 to 15 °C. The mean flow ranged from 0 to 12,000 l/s, thereby including both perennial (i.e. surface flow is maintained throughout the year) and temporary (i.e. surface flow ceases during dry periods) river reaches. All river reaches were under low human pressure as they fulfilled most of the 20 criteria to meet reference conditions in Mediterranean rivers. These criteria include a wide range of human uses and disturbances on rivers and streams (e.g., diffuse sources of pollution, invasive species, land use intensity, riparian vegetation, river geomorphology, habitat conditions and hydrological alterations) and some general aspects of naturalness, and they have already been used to assess the impact of stressors in Mediterranean rivers.

At each site and sampling date, we recorded water temperature, conductivity, oxygen (concentration and percentage of saturation) and pH using a multi-parametric digital probe YSI® Pro Plus. We calculated river flow (l/s) measuring the river section (width x depth) and water velocity with a digital anemometer Schiltknecht® MiniAir2. Additionally, a water sample was collected, filtered through glass fibre filters (GF/F; Whatman, Maidstone, UK), transported to the laboratory in ice, and frozen for water chemistry analysis. At the laboratory, we analyzed major anions (chloride, sulphates, nitrites and nitrates) by high-pressure liquid chromatography and estimated soluble reactive phosphorus and ammonium concentrations using standard colorimetric methods.

We collected aquatic macroinvertebrates using a circular hand net of 250 µm mesh size. Sampling consisted of an initial 3-min kick sample from all available habitats. We examined the initial kick sample in the field and we collected successive samples until no additional macroinvertebrate families were found. Since samples were originally collected for biomonitoring purposes, family-level resolution was used. However, families should be good surrogates of species-level assemblage patterns. Samples were preserved in 10% formaldehyde or 70% ethanol solution following the ECOSTRIMED protocol. We sorted the samples and identified macroinvertebrates to family level in the laboratory. The abundance of each occurring family was quantified and ranked as follows: 1 for 1 to 3 individuals, 2 for 4 to 10 individuals, 3 for 11 to 100 individuals and 4 for more than 100 individuals. This ranking followed the application of biological indices for assessing water quality, which was the initial purpose of this database.

We collected all samples during the spring season (mostly in May) in 20 consecutive years (1997-2017). 

Standardised Precipitation Index (SPI)

We determined the dry and wet years using the Standardised Precipitation Index (SPI). SPI represents the standardized deviation from a reference series. Positive and negative SPI values indicate precipitation greater or lower than the mean of the reference series, respectively. It is usually computed using a three-, nine- or 12-month span, but we chose the three-month SPI  (i.e. accumulated precipitation of the last three months before each sampling date) as it reflects short- to medium-term soil moisture characteristics and it has been successfully used as a surrogate of river flow in our study region. Therefore, we assumed that lower SPI values (dry years) corresponded to a higher probability of dry river reaches, whereas higher SPI values (wet years) corresponded to a higher probability of high flows (supporting information). We calculated the three-month SPI for each year from 1997 to 2017 using available monthly rainfall data measured in six weather stations of the Meteorological Service of Catalonia from 1950 to 2017.

According to the SPI values, the study periods could be classified as extremely dry (SPI ≤ -2), severely dry (-2 < SPI ≤ -1.5), moderately dry (-1.5 < SPI ≤ -1), neutral (-1 < SPI < 1), moderately wet (1 ≤ SPI < 1.5), severely wet (1.5 ≤ SPI < 2), extremely wet (2 ≤ SPI).

The SPI values are standardised and normalised with the following formula:

SPI = (Pi−Pm ) ÷ S

where Pi is the accumulated precipitation of a defined time span (i), Pm is the mean of rainfall of the period analysed, and Sp is the standard deviation of the series of precipitation from the same period.

Classification of the stream flow regime

We used the TREHS (Temporary Rivers Ecological and Hydrological Status) open access software to classify each site as perennial or temporary according to their natural hydrological regime . We calculated the coefficients Mf (% of months in a year with flow), Mp (% of months in a year with isolated pools) and Md (% of months in a year with a dry riverbed), which integrate information about the flow regime of the rivers over the years. According to this information 8 river reaches were permanent and 7 were temporary.

Classifying macroinvertebrates into dispersal groups

We classified macroinvertebrate families according to their ability to fly and to drift using information from a set of biological traits and expert opinion as described in the DISPERSE database. For the ability to fly, we considered traits that indicate dispersal mode (passive/active in the aquatic and the aerial environment), but also other related traits from the larvae and adult stages (e.g., female wing length, adult life span). Female wing length included 8 categories from <5 mm to >50 mm, whereas adult life span included 4 categories from < 1 week to > 1 year. For the ability to drift, we considered propensity to drift, which included three categories from rare to frequent intentional drift. Finally, considering these traits and expert opinion, we assigned an affinity to fly and to drift between 1 and 3 (1 for low affinity to 3 for strong affinity) to each family, and created two dispersal groups: strong flyers (i.e. strong fly affinity) and strong drifters (i.e. medium or strong drift affinity). This affinity accounted for intrafamily trait variability (i.e. associated to different genera or species) following the fuzzy coding approach. We considered taxa with medium drift affinity as strong drifters because only three families had the maximum drift affinity. The strong flyers group comprised 23 families (mainly Odonata, Trichoptera, Heteroptera and some Coleoptera and Diptera) and the strong drifters group comprised 20 families (mainly Ephemeroptera, Trichoptera, Plecoptera and some Diptera, plus the Crustacea Gammaridae).

Calculation of distances

We built a Euclidean matrix (water chemistry distance matrix) based on log-transformed and standardised values (mean=0, SD=1) of river flow, oxygen, ammonium, nitrites, nitrates, soluble reactive phosphorus, sulphates and chloride to characterise local environmental conditions. We also built a hydrological dissimilarity matrix through a Gower index based on the hydrological regime (Mf, Mp, Md), river flow of each sample and river flow of the previous summer. The latter intended to include information on preceding hydrological conditions that can be relevant for aquatic macroinvertebrates (i.e. the macroinvertebrate community can be different if it comes from a wet of dry year). This matrix expressed the differences in local hydrological conditions between sites and it was termed “flow regime difference”.

The dispersal-based distances included two physical distances (geographical and network distance) and a landscape resistance distance (topographic distance), which were used to approximate different potential pathways used by organisms to disperse between sites. The geographical distance considers the straight line between sites and was calculated as the Euclidean distance between geographical coordinates. The network distance measures the minimum distance between pairs of sites by following the river network. It was measured using a digitized stream network map (ECRIN database EcrRiv.mdb) and the Network Analyst extension in ARCGIS 10.0. Since the geographical and the network distance were highly correlated, we only considered the latter in our analyses, because it better represents potential dispersal distances for aquatic organisms. The topographic distance considers dispersal through the landscape taking into account the topography and it was calculated using a 200-m DEM and the program CIRCUITSCAPE. CIRCUITSCAPE uses electronic circuit theory to calculate the resistance of the landscape to movements of individuals between each pair of sites based on map pixel values (elevation in our case). Because our sites were located in different catchments we were unable to calculate the network distance between all the sites. Thus, we assigned the maximum network distance value found in our dataset (182.15 km) to pairwise comparisons between sites of different catchments.

Statistical analyses

We calculated alpha diversity (local number of families) for each site and year, and beta (pairwise Bray-curtis dissimilarity distance) and gamma (total number of families) diversity for each year as broad descriptors of macroinvertebrate variation across years.

We used traditional Mantel tests for spatial distances (topographic and network distance) and Mantel tests corrected by spatial autocorrelation through Moran spectral randomization (MSR) for environmental distances (water chemistry and flow regime difference) (999 runs for each test) to identify the relative effect of environmental and spatial filters on community composition in each year. The MSR correction was able to remove the spurious spatial dependence from our environmental distances, providing a correlation value that reflected the net environmental importance. Each Mantel test included the Bray-Curtis dissimilarity based on macroinvertebrate abundances as response variable and one environmental (water chemistry, hydrological) or spatial (topographic, network) distance as predictor. We also ran separated Mantel tests for community dissimilarity exclusively including taxa either with strong aerial dispersal or strong drift dispersal to assess the role of dispersal mode in modulating the importance of environmental and spatial filters. To estimate the relative importance of environmental and spatial distances in explaining community dissimilarity, and given that Mantel tests only allowed testing one distance at a time, we summed the Mantel statistic r of environmental (water chemistry + flow regime difference) and spatial (topographic + network) components. We replaced negative Mantel statistics by zero to avoid artificial distortions resulting from a lack of independence between the two studied matrices. Before summing Mantel r-values, we tested the collinearity between distances. Each pair of distances (water chemistry vs. flow regime, and topographic vs. network) was only weakly correlated (Pearson r: -0.20 to 0.25), suggesting that they explained independent and complementary changes in community dissimilarity.

To estimate if standardized deviations in precipitation (i.e. SPI) modified the importance of spatial and environmental filters, we performed a linear regression model including the correlation value of each Mantel test (Mantel test statistic) as response variable and SPI (linear and quadratic terms), distance type (environmental or spatial) and their pairwise interaction (SPI x distance type) as predictor variables. Also, to test if the importance of spatial and environmental filters varied with dispersal mode and precipitation, we ran a linear mixed-effect regression model including the correlation value of each Mantel test as response variable and SPI (linear and quadratic terms), distance type (environmental or spatial), dispersal mode, and their pairwise interactions (SPI x distance type, SPI x dispersal mode, distance type x dispersal mode) and three-way interaction (SPI x distance type x dispersal mode) as predictor variables. This model included year as a random factor. Before building the models, we applied a logit-transformation to environmental and spatial correlation values (response variables) to reduce variable distribution skewness and improve linearity. For each model we adopted a multi-model inference approach to quantify the regression coefficients, statistical support and importance of the predictor variables, using the function dredge from the MuMIN R package. This function ranks all the models generated from all the possible combinations of predictors included in the global model through the Akaike’s Information Criterion for small sample sizes (AICc). All the generated models were ecologically plausible. FWe retained all the models differing in less than 7 AICc units (∆AICc≤7) from the model ranking first (hereafter, top model). We also estimated the weighted-averaged response of top models (weighted by their Akaike weight) and plotted the fitted values to facilitate model output interpretation. All models were validated by visually checking their residuals for normality and homoscedasticity. For some models, we found a temporal dependence in the model residuals. In these cases, we added an autoregressive integrated moving average (ARIMA) term to the models to account for the lack of temporal independence in the residuals. All analyses were conducted using the R version 3.5.2.

Usage Notes

R code and functions to reproduce the results

Additional R functions

R script to perform Mantel Test to explore the importance  of environmental and spatial distances in explanining community dissimilarity 

R script to perform Multi-model inference to explore the importance  of environmental and spatial distances in explanining community dissimilarity 


Ministerio de Economía, Industria y Competitividad, Gobierno de España, Award: CTM2017-89295-P

European Regional Development Fund, Award: CTM2017-89295-P

Ministerio de Economía, Industria y Competitividad, Gobierno de España, Award: FJCI-2015-25785

European Regional Development Fund, Award: FJCI-2015-25785

Fundação para a Ciência e a Tecnologia, Award: POCI-01-0145-FEDER-007569

Fundação para a Ciência e a Tecnologia, Award: PTDC/CTA-AMB/31245/2017