Fish navigation of large dams emerges from their modulation of flow field experience

Abstract

Navigating obstacles is innate to fish in rivers, but fragmentation of the world's rivers by more than 50,000 large dams threatens many of the fish migrations these waterways support. One limitation to mitigating the impacts of dams on fish is that we have a poor understanding of why some fish enter routes engineered for their safe travel around the dam but others pass through more dangerous routes. To understand fish movement through hydropower dam environments, we combine a computational fluid dynamics model of the flow field at a dam and a behavioral model in which simulated fish adjust swim orientation and speed to modulate their experience to water acceleration and pressure (depth). We fit the model to data on the passage of juvenile Pacific salmonids (Oncorhynchus spp.) at seven dams in the Columbia/Snake River system. Our findings from reproducing observed fish movement and passage patterns across 47 flow field conditions sampled over 14 y emphasize the role of experience and perception in the decision making of animals that can inform opportunities and limitations in living resources management and engineering design.

Keywords: ecohydraulics; fish movement behavior; fish passage; hydraulic pattern; individual-based model.

Fig. 1.

Illustration of fish behavior responses B{1,2,3} to hydraulic pattern. (A) Direction of oriented swimming for each behavior (dashed black arrows) using two water flow velocity vectors (solid black arrows), where the longer vector represents faster water. Mean orientations of B{1,3} are with and opposite to the water flow direction, respectively, but B{2} orientation points toward faster flow, which is often in a direction different from water flow (streamline). Fish swimming is added to water movement (passive transport). (B) In heterogeneous flow, such as around a cube, distinguishing the contributions of fish swimming and passive transport is not straightforward. (C) For example, white-to-blue arrows illustrate the resultant fish movement (swimming + transport) in response to acceleration magnitude (A_M, in meters per square second) for behaviors B{1,2,3}. (B–D) Note slow (blue) and fast (red) water speed [velocity magnitude (V_M) in meters per second] contour lines. In C, solid black arrows depict the general water flow direction. Behavior B{2} can result in localized holding (milling) when faster water (red V_M contour line) is upstream of the fish. Upstream movement or milling resulting from B{2} can resemble upstream movement from B{3} even though the two behaviors are different. B{3} is generally more prolonged in the direction opposite to the flow vector. In contrast, B{1} orients swimming with the flow vector in the absence of B{2,3}. Mean patterns of V_M and A_M arise predictably in rivers from form resistance (e.g., rock, woody debris) and skin friction (e.g., water/boundary interface) (40, 41). (D) Turbulence is represented as TKE (square meters per square second). A horizontal slice at the midpoint of a 1-m cube placed at the bottom-center of an 8-m wide by 4-m deep channel is depicted in B–D. Flow was rendered using Reynolds averaged Navier–Stokes (RANS) CFD with an upstream boundary inflow of 32 m³⋅s⁻¹ for an average water velocity of 1 m⋅s⁻¹, which was selected to visualize hydraulic pattern easily. The CFD model was developed by staff at IIHR–Hydroscience and Engineering, University of Iowa.

Fig. 2.

Root-mean-square error (RMSE) values of the alternative models evaluated in our exploration of factors describing observed patterns of juvenile salmonids passing through routes (bypass, turbines, spillway) at seven large dams. Models increase in complexity from left to right, where pass randomly is the simplest model and general is the most complex model. (Far Right) General model is also fit to fictitious data as a test for whether the number of parameters, as opposed to the movement hypothesis, is primarily responsible for the model’s ability to fit data. The RMSE values (SI Appendix, Table S3) reflect the mean differences in modeled vs. observed proportions through the routes. For each model, RMSE values for the seven dams are grouped together as blue bars (Lower Granite Dam is the left-most blue bar and The Dalles Dam is the right-most blue bar). The equally weighted mean RMSE across all 47 datasets is shown as red bars. The SD is based on 10 random number seeds (SI Appendix, Model Evaluation). Routes with zero passage, such as closed routes having the same zero observed and modeled passage, are not factored into RMSE values.

Fig. 3.

Observed vs. modeled passage proportions through bypass, turbines, and spillway routes for passive particle (Left, no fish behavior) and simulated fish (Right, general model) using identical simulation attributes. Linear regression of observations vs. general model passage proportions (46, 47) for bypass (n = 41), turbine (n = 46), and spillway (n = 38) routes are, respectively, slope (0.76, 0.95, 0.94), intercept (5.98, 5.00, 0.82), and r² (0.70, 0.82, 0.89). Routes with zero passage and scenario C2 are not included (SI Appendix, Model Evaluation). Plots of simpler variants of the model are shown in SI Appendix, Fig. S2.

Fig. 4.

Patterns of water speed (V_M, meters per second; contour lines) and acceleration (A_M, meters per square second; contour fill) (A) used in the general model generate a single fish path (D) similar to the patterns of observed fish movements (C) (48). Neutrally buoyant, nonswimming particles follow water flow paths or streamlines (B) after release at depth similar to that of observed fish (C), 3.7 m below water surface. Fish movement differs substantially from mean flow (B), and the difference between flow paths (B) and fish (C and D) illustrates the contribution of fish behavior in dam passage. Swimming effort is not trivial, because V_M > 0.18 m⋅s⁻¹ in front of the dam, as shown in A, can exceed the cruise speed of a 90-mm long fish. A shallow, ∼1-m deep floating boom can elevate A_M sufficient to trigger B{2}, resulting in movement parallel to the boom (location 1). Elevated A_M surrounding the surface bypass collector (SBC) can trigger prolonged upstream movement, B{3}, to the boom (location 2) and exploratory milling (location 3) through recursive cycles between B{3} and B{1,2}. Acclimatization over time diminishes the response to the A_M contour, resulting in eventual passage. The CFD model was developed by staff at IIHR–Hydroscience and Engineering, University of Iowa, and acoustic tag telemetry data (48) were provided by the US Geological Survey, Columbia River Research Laboratory (SI Appendix, Fig. S6).