ABSTRACT
Benthic diatoms have been widely used to assess the ecological status of freshwater ecosystems, especially in the context of recent international water framework directive policies (e.g. the WFD). Despite diatom-based indices are known to respond fastly to water quality degradation, they are not designed to precisely identify the nature of pressures co-occurring in the environment. Based on large scale monitoring data, we aimed at building models able to estimate the risk of stream impairment by many types of anthropogenic pressures from taxonomy-based and trait-based characteristics of diatom assemblages. Random forest models were built to individually evaluate the impairment risk of diatom assemblages for six chemical and five hydromorphological or land-use related pressure categories. Eight models provided good impairment risk assessment (Area Under the Curve≥0.70). Under multi-pressure scenarios, models built for chemical pressures exhibited a better accuracy than hydromorphological or land-use related ones. Models were able to detect both ecological restoration and degradation, based on long-term surveys. These models have been implemented in a R user-friendly routine, to help stream managers to early identify degrading processes and prioritize management actions.