We conducted a field study during 2020–21 to describe habitat use patterns of juvenile Chinook salmon (
For more information on the USGS—the Federal source for science about the Earth, its natural and living resources, natural hazards, and the environment—visit
For an overview of USGS information products, including maps, imagery, and publications, visit
Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.
Although this information product, for the most part, is in the public domain, it also may contain copyrighted materials as noted in the text. Permission to reproduce copyrighted items must be secured from the copyright owner.
We thank the U.S. Army Corps of Engineers for funding this research and extend a special thanks to Rich Piaskowski, Jake MacDonald, and Rachel Laird for their interest and involvement. Several people provided insights and knowledge about the Willamette River including Greg Taylor with U.S. Army Corps of Engineers, Luke Whitman from the Oregon Department of Fish and Wildlife, Brian Bangs with the U.S. Fish and Wildlife Service, and Stan Gregory with Oregon State University. Additionally, we thank U.S. Geological Survey colleagues Brad Liedtke, Laurel Stratton Garvin, Collin Smith, and Jim Peterson for assistance with study design planning, development, implementation, and analysis. Data are not currently available from the funding organization, the U.S. Army Corps of Engineers. Contact the U.S. Army Corps of Engineers for further information.
U.S. customary units to International System of Units
Multiply | By | To obtain |
Length | ||
---|---|---|
inch (in.) | 25.4 | millimeter (mm) |
foot (ft) | 0.3048 | meter (m) |
mile (mi) | 1.609 | kilometer (km) |
Area | ||
acre | 4,047 | square meter (m2) |
square foot (ft2) | 0.09290 | square meter (m2) |
square mile (mi2) | 2.590 | square kilometer (km2) |
Volume | ||
cubic foot (ft3) | 0.02832 | cubic meter (m3) |
Flow rate | ||
foot per second (ft/s) | 0.3048 | meter per second (m/s) |
cubic foot per second (ft3/s) | 0.02832 | cubic meter per second (m3/s) |
International System of Units to U.S. customary units
Multiply | By | To obtain |
Length | ||
---|---|---|
millimeter (mm) | 0.03937 | inch (in.) |
meter (m) | 3.281 | foot (ft) |
kilometer (km) | 0.6214 | mile (mi) |
Area | ||
square meter (m2) | 0.0002471 | acre |
square meter (m2) | 10.76 | square foot (ft2) |
square kilometer (km2) | 0.3861 | square mile (mi2) |
Volume | ||
cubic meter (m3) | 35.31 | cubic foot (ft3) |
Flow rate | ||
meter per second (m/s) | 3.281 | foot per second (ft/s) |
cubic meter per second (m3/s) | 35.31 | cubic foot per second (ft3/s) |
Temperature in degrees Celsius (°C) may be converted to degrees Fahrenheit (°F) as follows: °F = (1.8 × °C) + 32.
Temperature in degrees Fahrenheit (°F) may be converted to degrees Celsius (°C) as follows: °C = (°F – 32) / 1.8.
Akaike Information Criterion
area under the curve
resource selection function
river kilometer
Science of the Willamette Instream Flow Team
U.S. Geological Survey
Flow management is important for the U.S. Army Corps of Engineers which owns and operates the Willamette Project encompassing 13 dams located on large tributaries to the mainstem Willamette River in western Oregon. Resource managers consider multiple factors when making flow management decisions including considerations protecting and enhancing spring Chinook salmon (hereinafter referred to as Chinook salmon;
The U.S. Geological Survey (USGS) has provided substantial scientific support to managers tasked with flow management decision-making in the Willamette River Basin (
Sampling was conducted in the mainstem Willamette River and lower reaches of the North Santiam and McKenzie Rivers. Sampling occurred on the mainstem Willamette River between the mouth of the McKenzie (river kilometer [rkm] 281) and Santiam (rkm 167) Rivers. Data collection on the North Santiam River occurred in the reach between the Jefferson Bridge Float Launch (rkm 6) and the Stayton Bridge County Boat Ramp near Stayton, Oregon (rkm 27). On the McKenzie River, sampling occurred in the reach between the Hendricks Bridge County Park Boat Ramp (rkm 33) and Taylor Landing (rkm 41;
Map showing reaches (pink shading) where habitat sampling occurred during April–July 2020 and 2021 on the mainstem Willamette, Santiam, and McKenzie Rivers.
Figure 1. Map showing reaches where habitat sampling occurred during April–July 2020 and 2021 on the mainstem Willamette, Santiam, and McKenzie Rivers
Fish use and habitat data were collected using two approaches. A stratified sampling design was used to ensure that the distribution of available habitat was adequately represented in our sampling and to provide data suitable for validating the performance of the
To implement the stratified sampling design, we used outputs from the hydraulic model of
Representative hydraulic model prediction (
Figure 2. Diagrams showing representative hydraulic model prediction and detailed sampling water depth and velocity groups used to proportionally distribute cells for a sampling location at a specific streamflow
Table 1. Depth and velocity groups used to stratify habitat suitability categories
[Proportion and cell numbers are provided for a sampling location at a specified streamflow to illustrate the distribution of cells by group using the hydraulic model prediction. Water depth in meters, water velocity in meters per second.
Group | Water depth | Water velocity | Habitat suitability | rkm 252.6–254.2 |
|
Proportion by |
Number of cells | ||||
A | 0.0–0.5 | 0.00–0.25 | Yes | 0.300 | 6 |
B | 0.0–0.5 | >0.25–0.50 | Yes | 0.240 | 5 |
C | >0.5–1.0 | 0.00–0.25 | Yes | 0.306 | 6 |
D | >0.5–1.0 | >0.25–0.50 | Yes | 0.154 | 3 |
E | 0.0–0.5 | >0.50–0.75 | No | 0.176 | 4 |
F | 0.0–0.5 | >0.75–1.00 | No | 0.093 | 2 |
G | >0.5–1.0 | >0.50–0.75 | No | 0.358 | 7 |
H | >1.0–1.5 | 0.00–0.25 | No | 0.258 | 5 |
I | >1.0–1.5 | >0.25–0.50 | No | 0.115 | 2 |
A targeted sampling design was used to collect habitat data at cells where juvenile Chinook salmon were observed. For this design, the snorkeler(s) moved slowly downstream while continuously scanning for juvenile Chinook salmon. Shorelines were observed at random where conditions allowed for snorkeler safety, and area was observed to the extent limited by visibility. To increase the potential for observing juvenile Chinook salmon, snorkelers attempted to observe various conditions (for example, depth, velocity, cover) present on the selected section of river. Once juvenile Chinook salmon were observed, downstream movement was discontinued, an approximately 2 m2 cell was visually established, and the fish were observed for approximately (~) 60 seconds to determine number of fish in the cell. Once the number of fish at the cell was recorded, the sampling crew collected the full suite of habitat data for the cell and then resumed a downstream search for more juvenile Chinook salmon—repeating this process for the remainder of the sampling day.
We compared river flow and water temperature conditions during our study to similar periods during 2015–19 using daily mean streamflow and river temperature data from USGS streamgage 14166000 Willamette River at Harrisburg, Oregon, from April 01 to July 31. To characterize seasonal sampling streamflow and river temperature conditions in a river segment, we used the nearest upstream gage to the sampling locations. Daily mean streamflow and river temperature data were obtained from existing USGS stream gages accessed on the National Water Information System website (
We collected information about fish presence and habitat attributes in each cell that was sampled using both the stratified and targeted sampling designs. Fish presence data included the identification and enumeration of all fish species observed in a cell. Juvenile salmonids were categorized into two size classes based on visual estimation by each snorkeler: less than or equal to (≤) 60 mm and greater than (>) 60 mm. To describe habitat conditions, we collected data on several habitat attributes in each cell. The latitude and longitude (measured at the center of the cell) of each cell location was recorded using a global positioning system (Trimble TDC600 handheld with a Trimble Catalyst DA1 antenna). Water depth (in meters) was also measured at the center of the cell using a 1.5 m topset wading rod. Water velocity was measured using a Sontek Flowtracker handheld acoustic doppler velocimeter. Each water velocity measurement consisted of a 40-second (s) average for two velocity components (the x- and y-planes) from which we calculated a magnitude for the total velocity vector, referred to hereinafter as velocity. Cells that were 0.46 m or shallower were sampled at a single depth (approximately 60 percent of the total depth); cells that were deeper than 0.46 m were sampled at 20 percent and 80 percent of the total depth. Multiple velocity measurements for a cell (depths greater than 0.46 m) were averaged to create a single depth averaged velocity for those cells. The Sontek Flowtracker also recorded water temperature (in degrees Celsius [°C]) in each cell. Substrate was characterized as the diameter in millimeters of the dominant substrate material present within the cell. Bed slope (in degrees) was measured perpendicular to the shoreline using a manual slope inclinometer. Distance-to-shore and distance-to-cover measurements were obtained using a laser rangefinder (Bushnell Scout 1000 ARC, Bushnell Outdoor Products)—distance-to-cover measurements were only recorded if cover was located within 10 m of the cell boundary (we assumed that fish could not effectively use cover located more than 10 m away). Cover type was visually estimated and categorically assigned to one of 7 variables: small woody debris (100 mm diameter or less), large woody debris (greater than 100 mm diameter), aquatic vegetation, terrestrial vegetation, boulder, undercut bank, or not available based on the nearest cover available for the cell. All data collection and sampling occurred during daytime hours.
For a selected stratified sampling location, sampling occurred over a 2-day period using methods adapted from
To assess the performance of the habitat suitability criteria used in the hydraulic habitat model, we compared fish presence at cells predicted as habitat or non-habitat using the SWIFT habitat criteria. All sampled cells were assigned as habitat or non-habitat based on water depth and velocity measured in the cell using the SWIFT Median habitat criteria for juvenile (pre-smolt) Chinook salmon—water velocity from 0 to 0.38 meters/second (m/s) and water depth from 0.05 to 1.07 m. The cells were then visualized using water depth and velocity bivariate plots by month to illustrate patterns of fish habitat use.
Data collected using the stratified sampling design and the targeted sampling design were merged and analyzed as a single dataset for analysis. Data from multiple sources (GPS receiver, acoustic doppler velocimeter, and field datasheets) were merged, examined for discrepancies, and reconciled to create a final dataset. Data compilation, proofing, visualization, and analysis were performed using R statistical software (
Habitat cells were assigned to one of three general categories, based on their location in the river, to allow for comparison of fish use between categories. The three geomorphic unit categories were: main channel, which was defined as the river segment containing the primary streamflow; side channel, which were segments of branching streamflow that maintained connection on both the upstream and downstream end of the segment (and could contain multiple braided sections); and alcoves, where areas were disconnected from upstream flow with connection to the main channel or a side channel on the downstream end (
Example of general habitat designation: main channel (pink dots), side channel (blue dots), and alcove (green dots).
Figure 3. Image showing example of general habitat designation: main channel, side channel, and alcove
We used logistic regression models to develop resource selection functions (RSF) that estimated the probability of fish presence in a given habitat cell for juvenile Chinook salmon in the Willamette River Basin. Logistic regression was performed using generalized linear models and logit link functions with the form of:
is the probability of observing Chinook salmon;
are fitted parameters;
is a variable/interaction of interest;
is an additional variable/interaction of interest; and
is the
To parameterize our models, we used depth and velocity with their associated quadratic and interaction terms. Quadratic terms were included to represent the expected biological response displaying an optimal value with asymmetrical tails. The interaction of depth and velocity was included under the assumption the response of fish to a given velocity may depend on depth. We established an a priori candidate set of models (including a null model) to compare relative model performance. The logistic regression models were fit to a binary variable indicating whether Chinook salmon were seen (1) or not seen (0). Finally, a separate set of models were fitted for each month (April, June, July) when data collection occurred to account for observed increases in fish size due to growth of a given year-class of juveniles.
Developed logistic regression models were evaluated for fit and compared for performance to select the best-fitting model. Model fit was evaluated by estimating c-hat as a measure of overdispersion and performing a Hosmer-Lemeshow Goodness of Fit test using the R package vcdExtra (
To illustrate the response of the model to depth and velocity, we created combinations of water depth and velocity as inputs to the resource selection function. For each sampling month we sequentially increased either water depth (0–1.5 m, by 0.01 m) or velocity (0–1.5 m/s, by 0.01 m/s), while setting the alternate parameter constant at the mean value observed during that period. Secondarily, we created a set of possible combinations for water depths (0–1.5 m, by 0.01 m) and velocity (0–1.5 m/s, by 0.01 m/s) for application of the resource selection functions.
To illustrate the difference between the habitat assessment methodologies of the SWIFT habitat suitability criteria and the resource selection function produced in this study, we used receiver operating characteristic curves to select thresholds to produce Narrow, Median, and Broad categories. For example, one could select 0.5 as the threshold for the probability of presence above which a cell would be classified as habitat and below which it would not be considered as suitable habitat. As the probability threshold increases, the error in false classification of habitat is reduced but not all true habitat classification is incorporated. Thus, a higher probability threshold results in a narrow range of values classified as habitat with reduced errors in false classification. A lower probability threshold incorporates a wider range of values classified as habitat but increases the potential for classifying habitat falsely. The Median category probability thresholds were determined as the point on the curve with the greatest distance from the random chance line. Narrow and Broad category thresholds were calculated as the median of either side of the remaining curve bisected by the Median category threshold. We used the Median threshold probabilities to illustrate how the probability threshold selection corresponds to the true positive and true negative classification. These thresholds represent the minimum probability from which to obtain depth and velocity criteria from the resource selection function. The range of the criteria is determined by the minimum and maximum values of depth and velocities for all probabilities in the resource selection function greater than the threshold value.
Streamflow and water temperature conditions on the mainstem Willamette River during the study were similar to those observed during previous years in April, June, and July (
Mainstem Willamette River streamflow and water temperature for April (
Figure 4. Graphs showing mainstem Willamette River streamflow and water temperature for April, June, and July at Harrisburg, Oregon, 2015–21
Table 2. Sampling dates, river, mean streamflow, mean river temperature and number of habitat cells by sampling design
[Mean streamflow units are in cubic feet per second, mean temperature is in degrees Celsius]
Sampling date | River | Mean streamflow | Mean temperature | Stratified cells | Targeted cells |
June 02–03, 2020 | Willamette | 8,200 | 15.8 | 39 | 0 |
June 15–17, 2020 | Willamette | 8,270 | 13.7 | 87 | 0 |
June 29–July 01, 2020 | Willamette | 5,255 | 15.7 | 88 | 0 |
July 13–15, 2020 | Willamette | 4,770 | 16.9 | 88 | 0 |
July 21–23, 2020 | Willamette | 4,950 | 19.1 | 0 | 53 |
April 18–20, 2021 | North Santiam | 2,300 | 11.5 | 30 | 23 |
April 21–22, 2021 | McKenzie | 1,120 | 10.4 | 30 | 29 |
April 28–29, 2021 | Willamette | 5,565 | 11.8 | 0 | 28 |
June 01, 2021 | Willamette | 5,740 | 15.7 | 0 | 23 |
June 02, 2021 | McKenzie | 3,320 | 13.3 | 0 | 19 |
June 03, 2021 | North Santiam | 1,370 | 18.1 | 0 | 30 |
June 07–10, 2021 | Willamette | 5,095 | 14.4 | 0 | 67 |
A total of 634 cells were sampled during 2020–21 with 362 cells sampled using the stratified sampling design and 272 cells sampled using the targeted sampling design (
Sampling locations for reaches on the mainstem Willamette River from Eugene to McCartney (
Figure 5. Maps showing sampling locations for reaches on the mainstem Willamette River from Eugene to McCartney, McCartney to Peoria, Peoria to Corvallis, and Corvallis to the Santiam River confluence, 2020 and 2021
Sampling locations on the North Santiam River (
Figure 6. Maps showing sampling locations on the North Santiam River and McKenzie River in 2021
Histograms showing distribution of data collected for individual habitat variables. Distance-to-cover data does not include 243 cells where cover was not available within 10 meters of the habitat cell.
Figure 7. Histograms showing the distribution of data collected for individual habitat variables
We found that SWIFT habitat criteria provided a reasonable representation of juvenile Chinook salmon habitat use in the Willamette River during April while there were substantial differences between habitat criteria and observed fish use during June and July (
Monthly plots showing water depth and water velocity measured at each habitat cell during the study. Yellow circles are habitat cells where juvenile Chinook salmon (
Figure 8. Monthly plots showing water depth and water velocity measured at each habitat cell during the study
Table 3. Number of habitat cells where juvenile Chinook salmon were observed, and not observed in relation to the Median habitat suitability criteria used by the hydraulic habitat model
Median criteria | Chinook observed | |
No | Yes | |
April | ||
---|---|---|
Suitable | 34 | 83 |
Unsuitable | 16 | 7 |
June | ||
Suitable | 134 | 67 |
Unsuitable | 74 | 34 |
July | ||
Suitable | 70 | 15 |
Unsuitable | 60 | 40 |
Juvenile Chinook salmon were frequently observed in habitat cells located on the main channel and side channels during our sampling but were rarely observed in habitat cells located in alcoves. Juvenile Chinook salmon were observed in 39 percent (246 cells) of the habitat cells which were surveyed (
Table 4. Number of habitat cells by habitat category, month, and juvenile Chinook salmon observation
Habitat category | Habitat cells with juvenile |
|||||
April | June | July | ||||
No | Yes | No | Yes | No | Yes | |
Main channel | 31 | 42 | 75 | 69 | 110 | 46 |
Side channel | 19 | 47 | 104 | 31 | 18 | 9 |
Model selection criteria resulted in models of similar complexity, containing multiple common parameters, and increasing model accuracy from April to July, as measured by AUC. The selected models for April, June, and July each included parameters for depth, velocity, and quadratic terms for depth and velocity. Additionally, the June and July models also included an interaction between depth and velocity. For April, a total of five candidate models had delta AICc values less than 2.0 (
Table 5. Candidate models characteristics: modeling structure, parameters, AICc, delta AICc, and AUC values and model accuracy for selected models
[
Modeling structure | Parameters | AICc | Delta AICc | AUC | Model accuracy |
April | |||||
---|---|---|---|---|---|
V | 2 | 176.06 | 0.00 | -- | -- |
V+D+D2 | 4 | 176.63 | 0.57 | -- | -- |
V+D+V2+D2 | 5 | 177.00 | 0.94 | 0.635 | 0.781 |
V+D+V2 | 4 | 177.72 | 1.65 | -- | -- |
V+D | 3 | 177.80 | 1.74 | -- | -- |
V+D+V2+D2+VD | 6 | 178.43 | 2.37 | -- | -- |
null | 1 | 184.52 | 8.46 | -- | -- |
D | 2 | 186.21 | 10.14 | -- | -- |
June | |||||
V+D+V2+D2 | 5 | 318.72 | 0.00 | -- | -- |
V+D+V2+D2+VD | 6 | 318.95 | 0.23 | 0.726 | 0.790 |
V+D+V2 | 4 | 330.03 | 11.31 | -- | -- |
V+D+D2 | 4 | 338.24 | 19.52 | -- | -- |
V | 2 | 350.02 | 31.31 | -- | -- |
null | 1 | 351.32 | 32.60 | -- | -- |
V+D | 3 | 351.48 | 32.76 | -- | -- |
D | 2 | 353.26 | 34.54 | -- | -- |
July | |||||
V+D+V2+D2+VD | 6 | 201.05 | 0.00 | 0.837 | 0.854 |
V+D+V2+D2 | 5 | 203.08 | 2.03 | -- | |
V+D+D2 | 4 | 206.08 | 5.03 | -- | -- |
V+D+V2 | 4 | 215.29 | 14.24 | -- | -- |
V | 2 | 220.00 | 18.95 | -- | -- |
V+D | 3 | 221.42 | 20.37 | -- | -- |
null | 1 | 259.59 | 58.54 | -- | -- |
D | 2 | 261.35 | 60.30 | -- | -- |
We plotted resource selection functions separately for depth and velocity to illustrate the predicted response to changes in these variables (
Model predicted probabilities of observing juvenile Chinook salmon (
Figure 9. Model predicted probabilities of observing juvenile Chinook salmon for velocity at mean depth and depth at mean velocity with confidence intervals by month
Because of the higher order terms in the model (for example, interactions), we also plot the resource selection functions for water velocity and water depth to illustrate how the probability of presence depends jointly on both variables (
Estimated presence probability from the resource selection function (RSF) for juvenile Chinook (
Figure 10. Estimated presence probability from the resource selection function for juvenile Chinook salmon based on water velocity and water depth
Thresholds were established for the resource selection function to compare with the SWIFT habitat suitability criteria categories (Narrow, Median, and Broad) using receiver operating characteristic curves. The probability thresholds above which cells were classified as habitat were 0.73, 0.59, and 0.42 for Narrow, Median, and Broad categories, respectively, for April. At the Median category threshold for April the true positive rate was 0.88 and true negative rate 0.44 (
Receiver operating characteristic (ROC) curve (bold solid line) with Narrow (left vertical dashed line), Median (vertical solid line), and Broad (right vertical dashed line) category thresholds for selected models. Area under the curve (AUC) is an indication of model performance. For comparison perfect chance (the diagonal dotted line) representing a random flip of a coin has an AUC = 0.5.
Figure 11. Graphs showing receiver operating characteristic curve with Narrow, Median, and Broad category thresholds for selected models
Habitat use limits from the habitat suitability criteria for fry sized salmonids was compared to the April resource selection function values (
A greater difference was observed comparing the pre-smolt values from the habitat suitability criteria with the resource selection function limits for July (
Table 6. Comparison of habitat suitability criteria and resource selection function classification categories used to estimate habitat suitability for juvenile Chinook salmon
[
Habitat metric | HSC | RSF | ||||
Broad | Median | Narrow | Broad | Median | Narrow | |
Fry | April | |||||
---|---|---|---|---|---|---|
Velocity (m/s) | 0–0.46 | 0–0.38 | 0–0.15 | 0–0.57 | 0–0.45 | 0–0.27 |
Depth (m) | 0.05–1.52 | 0.05–1.07 | 0.05–0.61 | 0.02–1.18 | 0.16–1.04 | 0.37–0.83 |
Pre-smolt | July | |||||
Velocity (m/s) | 0–0.91 | 0–0.50 | 0–0.38 | 0.08–1.50 | 0.27–1.50 | 0.47–1.50 |
Depth (m) | ≥0.05 | 0.05–1.07 | 0.05–0.69 | 0.27–1.50 | 0.40–1.45 | 0.54–1.31 |
This study used empirical observations of habitat use by juvenile Chinook salmon in the Willamette River Basin to develop resource selection functions useful for hydraulic habitat modeling efforts in the future. Previous studies provided data on habitat use by juvenile Chinook salmon in the Willamette River Basin (
Our sampling included sites located in the main river channel, side channels, and in alcoves, and we found that juvenile Chinook salmon were routinely present in main channel and side channel habitats but were seldom observed in alcoves. Off-channel habitat has been shown to be important for juvenile Chinook salmon (
We found that habitat use by juvenile Chinook salmon underwent a seasonal shift that was likely driven by the increasing size of fish over time. The resource selection functions developed from data collected during our study showed that optimum water velocity for juvenile Chinook salmon in April was near 0.1 m/s and increased to nearly 0.9 m/s in July, while optimum depth increased from approximately 0.6 m to 0.7 m during the same period. The resource selection functions also showed that the probability of juvenile Chinook salmon occupying a given habit decreased substantially as velocity and depth departed from these optimum values. These findings are supported by other studies that have shown that juvenile Chinook salmon move farther offshore and into deeper water as they grow during spring and early summer (
These resource selection functions can be used to improve hydraulic habitat models, particularly in comparison to coarser approaches such as the SWIFT habitat suitability criteria. For example, the hydraulic habitat model classified habitat as suitable, using the SWIFT criteria, if habitat cells had water depth as shallow as 0.05 m. We found that juvenile Chinook salmon were rarely observed in water depth less than 0.25 m, which means that the SWIFT criteria resulted in an overestimation of habitat availability relative to our observations. Additionally, all habitat cells with water velocity and depth attributes that met the SWIFT criteria were designated as suitable habitat, so this approach did not allow for proportional predictions of habitat use based on various water depth and velocity combinations. These observations illustrate the value of using in-basin data to develop resource selection functions with proportional probabilities to optimize predictive hydraulic habitat models.
For information about the research in this report, contact
Director, Western Fisheries Research Center
U.S. Geological Survey
6505 NE 65th Street
Seattle, Washington 98115-5016
Manuscript approved on January 9, 2023
Publishing support provided by the U.S. Geological Survey
Science Publishing Network, Tacoma Publishing Service Center