Using streamflow observations to estimate the impact of hydrological regimes and anthropogenic water use on European stream macroinvertebrate occurrences

Abstract

Understanding the drivers of stream macroinvertebrate distribution patterns—the most diverse animal group in freshwater ecosystems—is a major goal in freshwater biogeography. Climate and topography have been shown to explain species' distributions at continental scales, but the contribution of natural and anthropogenically altered streamflow is often omitted in large‐scale analyses due to the lack of appropriate data. We test how macroinvertebrate occurrences can be linked to streamflow observations and evaluate the relative importance of streamflow regimes and water use for macroinvertebrate occurrences from 19 orders across Europe. We first paired species sampling locations with hydrological gauging stations considering 5 combinations of the geographic distance and difference in flow accumulation (upstream contributing area). We then used Generalized Linear Models to assess the influence of the streamflow regime, simulated water use, and climate and topography on the occurrence of macroinvertebrates. The pairing method that assigned species records to the closest gauging station in terms of both distance and flow accumulation performed best. Most of the species studied occurred preferentially in river habitats with low mean annual streamflow and streamflow variability, high winter streamflow, and low levels of water withdrawals for irrigation or manufacturing. We conclude that flow accumulation is a useful proxy to evaluate the proximity of species records to gauging stations, omitting species records that do not belong to a given stream reach. The strong contribution of streamflow and water use indicators on macroinvertebrate occurrences underline their importance for yielding robust occurrence estimates.

Projected effects of Climate‐change‐induced flow alterations on stream macroinvertebrate abundances

“…River biota depend on a range of environmental variables, including natural habitat conditions as well as stressors. While the effects of a variety of environmental variables and stressors such as land-use, climate, and substrate conditions on riverine species are well understood (Miserendino et al, 2011;Schröder et al, 2013), the relationship between riverine species' abundances and river flow is less often explored (Kuemmerlen et al, 2014(Kuemmerlen et al, , 2015Pyne & Poff, 2017), although it has been widely stated that flow (i.e., discharge) is one of the key habitat variables in river ecosystems (Arthington, Bunn, Poff, & Naiman, 2006;Dewson, James, & Death, 2007;Domisch et al, 2017;Poff et al, 1997).…”

A high-resolution streamflow and hydrological metrics dataset for ecological modeling using a regression model

“…The gauging sites were paired with species sampling sites in QGIS 22 within a buffer of 3 km. To ensure the sample sites were placed on the original stream, the original flow accumulation value had to be within 10% of the flow accumulation of the newly allocated grid cell 37 . A total of 327 sites remained that had associated observed and simulated discharge, as well as species presence data.…”

Inside or Outside: Quantifying Extrapolation Across River Networks

“…Cross-validation procedures are often applied to approximate independent assessments where truly independent data are not available (Arlot & Celisse, 2010), with data repeatedly split (or blocked) into training and independent evaluation subsets for model testing (Harrell et al, 1996). Cross validation using random splits has frequently been applied, including assessing performance of multiple linear regression and random forest (RF) models of water losses (Jenkins et al, 2018), generalized linear models of macroinvertebrate distribution (Domisch et al, 2017), boosted regression trees of plant species richness (Bailey et al, 2017), and hybrid RF-generalized linear models of species richness (Li et al, 2017). These cross-validation procedures assume training, and evaluation subsets are independent, but this assumption may be violated if pseudo-replication or spatial autocorrelation persists between subsets (Roberts et al, 2017).…”