Saline streams occur in the most arid regions of the Mediterranean, primarily in the Eastern Iberian Peninsula, North Africa, and Sicily (Italy). These streams exhibit a high mineralization degree due to the presence of high soluble salts in the basin (lithology and subterranean deposits). Their biological communities are composed of euryhaline and specialist species that can withstand osmotic and thermal stress, which constrains the presence of typical freshwater taxa. Among aquatic macroinvertebrates, Coleoptera, Diptera, and Hemiptera orders account for a high amount of species diversity.

Conductivity is recognized as the main variable that shapes species presence and richness. Such streams are currently threatened by human practices, mainly by irrigation agriculture. Freshwater and nutrient inputs enhance the reduction in salt concentrations, which modifies community structure and produces a decline in specialist species.

Aims

1. To develop a predictive model for reference macroinvertebrate communities in saline Mediterranean streams at different taxonomic levels
2. To evaluate if this model is able to detect human disturbances

Methods

A database with the biological data from extensive surveys across Spain, Morocco, and Sicily was constructed. Community composition was characterized for each sampling date: individuals were counted and sorted at family level, except for Coleoptera and Hemiptera, which were counted and sorted at species level. Electrical conductivity was measured in each sampling date that was used as a salinity proxy. Climatic, lithologic, geomorphologic, land use, and hydrological variables were also obtained from GIS layers for each stream basin.

Statistical methods

Different approaches have been used to model communities. Traditional approaches first define assemblage types and then model their relationship with the environmental variables (e.g., RIVPACS Moss et al. 2009). Novel approaches, such as Multiresponse Artificial Neural Networks, display some advantages over traditional methods which make them more suitable to model community structure. They are non-parametric methods that can model non-linear relationships and even many response variables at the same time (i.e., many dependent variables). In such models, environmental variables are used as independent variables, and the probability of taxa presence are the predicted variables (e.g., Olden 2003, Joy and Death 2004, Olden et al 2006). Variables to be modeled are macroinvertebrate families which occur in saline streams (around 30 different taxa) and the Coleoptera and Hemiptera assemblages at genus and species level (around 40 species).