Predicting road culvert passability for migratory fishes

Aim: Our goal was to predict road culvert passability, as defined by culvert outlet drop and outlet water velocity, for three fish swimming groups using remotely collected environmental variables that have been shown to influence thepassability of road culverts. Location: Laurentian Great Lakes Basin, north-eastern North America, on the Canada–USA border. Methods: We generated four boosted regression tree models, one for road culvert outlet drop and one each for the three culvert outlet water velocities, and predicted the probability of impassable road culverts on low-order streams (Strahler 1-4) based on the models. Independent variables in the models included the upstream area draining to the culvert, slope at the culvert, stream segment gradient and stream reach gradient. Results: Gradient of the stream segment was the most important predictor inthe outlet drop model, while upstream drainage area was the most important predictor in the three water velocity models. A majority of road culverts onlow-order streams are estimated to be passable even for weaker swimming fishes. Moderate to highly impassable road culverts are distributed across many low-order streams throughout the basin, but particular regions are predicted tohave higher densities than others due to topography. Main conclusions: Predicted passability of road culverts by migratory fish isrelated to natural gradients in topography and stream size. While the probability of any particular culvert being impassable is low, the vast number of culverts in the basin means that, together, they could pose a greater challenge tomigratory fish than dams. Our modelling framework could be used in anyregion where culverts are prevalent in the riverscape. The resulting estimates of passability to fishes can guide surveys towards the most problematic hydrological regions and structures and contribute to broad-scale prioritization of barrier removals to restore ecological connectivity.