Mark Gard

Comparison of spawning habitat predictions of PHABSIM and River2D models

Abstract This study compared the predictions of two instream flow habitat models, the Physical Habitat Simulation System (PHABSIM) and River2D, with regards to spawning habitat for chinook salmon, Oncorhynchus tschawytscha, and steelhead trout, Oncorhynchus mykiss. Spawning habitat was simulated with both models for eight sites in the Sacramento River, five sites in the American River and one site in the Merced River, California, using habitat suitability criteria developed from data collected on redds in each of these rivers. For four out of five cases, both models correctly predicted that the combined suitability, calculated as the product of the depth, velocity and substrate suitabilities, of occupied locations was significantly greater than the combined suitability of unoccupied locations. There was little difference in the flow‐habitat relationships for each site and set of habitat suitability criteria predicted by the two models. The use of River2D, rather than PHABSIM, is still warranted given its ability to model complex flow conditions which cannot be simulated with PHABSIM.

Response to Williams (2010) on Gard (2009): Comparison of spawning habitat predictions of PHABSIM and River2D models

Williams (2010) raises a number of interesting points with regard to the testingof instreamflowmodels.Before responding to the critique and suggestions of Williams (2010), I note several inaccuracies in Williams (2010). River2D does not divide the study area into cells. River2D calculates depths and velocities at nodes of the triangulated irregular network, and then interpolates between the surrounding nodes to calculate depths and velocities at any given location. There are two types of two-dimensional (2D) hydraulic models – finite element and finite difference (Steffler and Blackburn 2002). River2D is a finite-element model. It would be accurate to say that finite difference 2D models divide the study area into cells. Gard (2009) used the one-dimensional model throughout the sites, since the entire area of all of the sites did not have transverse flow or other complex flow patterns. There definitely are other portions of the Sacramento, American, and Merced Rivers where PHABSIM could not be used due to transverse flow or other complex flow patterns, but they were not in the 13 study sites addressed by Gard (2009). The weighted useable area presented inGard (2009) is not normalized by channel length. However, Williams (2010) is correct that usually weighted usable area (WUA) is normalized by channel length. The comparison of measured and simulated velocities in Gard (2009) is not for redd locations, but rather for the sites overall – this is for validation of the hydraulic model, before any consideration of biology. In the interest of brevity, my responses to Williams (2010) will focus on the American River as an example. We are exploring a number of biological verification tests, in addition to the Mann–Whitney tests used in Gard (2009). A key aspect of whatever test is used is having a quantitative measure to conclude whether or not the combined hydraulic and biological models have been verified. Box plots, as shown for the American River data (Figure 1), do not appear to be an effective means to present the magnitude of differences and precision of estimates. Rather, I suggest that histograms of combined suitability (Figure 2) are a better means to display effects and precision. Although the histograms provide a good visual comparison, they still do not provide a quantitative measure to conclude whether or not the combined hydraulic and biological models have been verified. Gard (2009) used statistical tests of combined suitability and WUA curves because, right or wrong, statistical tests are the norm for publications in the scientific literature (Johnson 1999). Given the hypothesis to be tested, the combined suitability index (CSI) of occupied locations are greater than the CSI of unoccupied locations; a Mann–Whitney U-test is a logical choice, given that CSI is not normally distributed (Figure 2). The rationale for using a Kolmogorov–Smirnov test is that it is a logical choice to test the hypothesis: the flow-habitat relationships predicted by PHABSIM and River2D have different shapes. I agree with Williams (2010) that it is difficult to determine what the appropriate degrees of freedom is for the Kolmogorov– Smirnov test, as used in Gard (2009). It would be good to have a quantitative objective method to determine how many of the 55 comparisons of PHABSIM and River2D flow-habitat relationships have practical differences, such as resulting in different flow management decisions, but I am not aware of any such method. Scatter plots of measured versus predicted depths, velocities, and substrate sizes at redd locations (Figures 3–5) indicate that most of the errors in redd predictions can be attributed to the hydraulic models. Although River2D performed somewhat better (correlations of 0.23 and 0.46 between, respectively, measured and predicted velocities and depth, and 26% of redds with the same simulated and measured substrate size) than PHABSIM (correlations of 20.008 and 0.36 between, respectively, measured and predicted velocities and depth, and 19% of redds with the same simulated and measured substrate size), both hydraulic models had substantial errors. It should be noted that the River2D simulations in Gard (2009) were the first River2D models we had developed, and our more recent

Applications of New Technologies to Instream Flow Studies in Large Rivers

Abstract An acoustic Doppler current profiler, underwater video system, hand-held laser range finder and global positioning receiver were used to collect data for instream flow studies on the Sacramento and lower American rivers in California. The use of the equipment decreased the time required to collect spawning criteria data for Chinook salmon Oncorhynchus tshawytscha in deep water in a given area by a factor of 3.4 and doubled the number of transects that could be modeled with the same budget. With the application of quality control criteria, discharges could be measured with an average accuracy of 2.7% versus gauge data with an accuracy of 5%. The total time required to collect data for two-dimensional habitat sites varied with the length and complexity of the sites, and was equivalent to the total time required for physical habitat simulation (PHABSIM) data collection for shorter sites, and less for longer sites.

Demonstration Flow Assessment and 2-D Modeling: Perspectives Based on Instream Flow Studies and Evaluation of Restoration Projects

Abstract Railsback and Kadvany (2008) make a compelling case for the use of demonstration flow assessments (DFA) for instream flow assessments as a more robust method than traditional one-dimensional habitat simulation techniques, such as the Physical Habitat Simulation System. But, based on experience with DFAs used for instream flow assessments and evaluation of stream restoration projects, the methods presented by Railsback and Kadvany (2008) may not give reproducible results, and DFAs have significant drawbacks relative to two-dimensional (2-D) hydraulic and habitat modeling. Application of the DFA methodology presented in Railsback and Kadvany (2008) to assess a stream restoration project on the Trinity River, California, did not give reproducible results, with substantial disparity between replicate surveys in the total quantity and spatial distribution of habitat. As an empirical two-dimensional habitat modeling method, DFAs have several drawbacks. Compared to 2-D models, DFAs require interpolation...