Assigning species to redds: exploring uncertainty in a central California coastal creek

Abstract

The California Coastal Monitoring Plan (CMP; Adams et al. 2011) advocates a
two-stage approach (i.e., redd counts and fixed counting/trapping stations) to
estimate adult salmonid abundance in the northern monitoring area (Figure 1).
Redd counts often fail to produce robust estimates of abundance for coho salmon
and steelhead in watersheds at the southern end of the Central California Coast
(CCC) ESU/DPS for two reason:
1. Sandbars form across creek mouths each summer and persist until large
winter storms produce sufficient streamflow to erode the sandbar (Figure
2A,B). Once the sandbar is opened, adult coho salmon and steelhead
often concurrently move into the stream and begin spawning (Table 1,
Figure 2C).
2. Redds are most often encountered after construction without live fish or
carcasses in the immediate vicinity. Hence, definitive species assignments
are often not possible (Figure 3).
To address the uncertainty surrounding redds of unknown origin, two species
assignment methods are frequently applied in support of the CMP:
• A logistic regression model that makes species predictions based on the
timing of redd construction and redd geometry (Gallagher & Gallagher 2005,
hereinafter G&G).
• The k-nearest neighbors (kNN) algorithm which assigns species based on a
majority rule of known nearest neighbors in time and space (Ricker et al.
2014).
The relative performance of these methods, and their applicability to watersheds
at the southern end of the CCC ESU/DPS, remain poorly understood

Full text

The California Coastal Monitoring Plan (CMP; Adams et al. 2011) advocates a
two-stage approach (i.e., redd counts and fixed counting/trapping stations) to
estimate adult salmonid abundance in the northern monitoring area (Figure 1).
Redd counts often fail to produce robust estimates of abundance for coho salmon
and steelhead in watersheds at the southern end of the Central California Coast
(CCC) ESU/DPS for two reason:
1. Sandbars form across creek mouths each summer and persist until large
winter storms produce sufficient streamflow to erode the sandbar (Figure
2A,B). Once the sandbar is opened, adult coho salmon and steelhead
often concurrently move into the stream and begin spawning (Table 1,
Figure 2C).
2. Redds are most often encountered after construction without live fish or
carcasses in the immediate vicinity. Hence, definitive species assignments
are often not possible (Figure 3).
To address the uncertainty surrounding redds of unknown origin, two species
assignment methods are frequently applied in support of the CMP:
•A logistic regression model that makes species predictions based on the
timing of redd construction and redd geometry (Gallagher & Gallagher 2005,
hereinafter G&G).
•The k-nearest neighbors (kNN) algorithm which assigns species based on a
majority rule of known nearest neighbors in time and space (Ricker et al.
2014).
The relative performance of these methods, and their applicability to watersheds
at the southern end of the CCC ESU/DPS, remain poorly understood.
Background
Assigning species to redds: exploring uncertainty in a central California coastal creek
Lindsay E. Hansen1*, Karlee M. Liddy1*, Rosealea M. Bond2, Ann-Marie K. Osterback2, Cynthia H. Kern2,
Alexander E. Hay2, Jeffrey M. Perez2, Joshua M. Meko2, and Joseph D. Kiernan2*
1Watershed Stewards Program of the California Conservation Corps and Americorps, Placed at NOAA Southwest Fisheries Science Center, Santa Cruz
2Institute of Marine Sciences, University of California, Santa Cruz & Fisheries Ecology Division, Southwest Fisheries Science Center, NMFS, NOAA, Santa Cruz, CA.
*Correspondence to: lindsay.hansen@noaa.gov, karlee.liddy@noaa.gov, joseph.kiernan@noaa.gov
Methods
Key Findings
The Question
How well do commonly used redd assignment methods perform in Scott Creek, a typical central California
coastal watershed with high temporal and spatial overlap of spawning coho salmon and steelhead?
Three coho salmon on a redd.
B. Organized steelhead redd. Disorganized coho redd.
Literature Cited:
Adams, P. B., L. B. Boydstun, S. P. Gallagher, M. K. Lacy, T. McDonald, and K. E. Shaffer. 2011. California coastal
salmonid population monitoring: strategy, design, and methods. California Department of Fish and Game, Fish
Bulletin no. 180. 82 pp.
Gallagher, S. P. and C. M. Gallagher. 2005. Discrimination of Chinook salmon, coho salmon, and steelhead redds
and evaluation of the use of redd data for estimating escapement in several unregulated streams in Northern
California. North American Journal of Fisheries Management 25:284-300.
Gallagher, S. P., and M. K. Knechtle. 2005. Coastal northern California salmon spawning survey
protocol. California State Department of Fish and Game, Fort Bragg, California.
Gallagher, S. P., P. K. Hahn, and D. H. Johnson. 2007. Redd Counts. Pages 197-234 in D. H. Johnson, B. M. Shrier,
J. S. O'Neal, J. A. Knutzen, X. Augerot, T. A. O'Neil, and T. N. Pearsons, editors. Salmonid field protocols handbook.
Techniques for assessing status and trends in salmon and trout populations. American Fisheries Society, Bethesda,
Maryland.
Ricker, S., J., Ferreira, S. P. Gallagher, D. McCanne, and S. A. Hayes. 2014. Methods for classifying anadromous
salmonid redds to species. Coastal Salmonid Population Monitoring Technical Team Report. California Department
of Fish and Wildlife, Arcata, California. 24p
The use of redd counts for population monitoring can be particularly challenging in locations where multiple salmonid species are present and
adults exhibit considerable overlap in spawning timing. We found the two species assignment methods produced very different predictions
(Figures 5,6).
Figure 4: Diagram of a salmonid redd (A). Redd geometry was calculated using pot (red lines) and tailspill (blue
lines) measurements. Examples of salmonid redds in the Scott Creek Watershed (B).
Figure 1: Two stage monitoring framework adopted by the CMP to estimate
adult abundance.
Table 1: Variable timing in sandbar opening results in strong temporal
overlap of coho salmon and steelhead in the Scott Creek (Santa Cruz Co.)
watershed.
Table 3: Regional escapement estimates derived from spawner to redd ratios developed using alternative
methods to classify redds of unknown origin. San Mateo and Santa Cruz County streams, spawn year 2015.
Figure 5: The two species assignment methods examined produced substantially different results in most
spawn years.
Figure 6: During spawn year 2015, most of the redds encountered in the Scott Creek watershed were
classified as unknown in the field (A). Unknown redds were subsequently assigned to species using the
G & G (B) and kNN (C) assignment methods.
Challenges with each species assignment method:
●G & G equation is based on timing and redd geometry. In Scott
Creek, the equation often failed due to temporal overlap of
coho salmon and steelhead on the spawning grounds.
We found, the G & G equation consistently misclassified early
returning steelhead as coho salmon. This tendency
substantially overinflated coho salmon abundance at the
regional scale (Tables 2 and 3).
●The kNN algorithm performed well when there were sufficient
numbers of known (100% certainty) observations (i.e., plenty of
neighbors).
Spawn year 2015 was the only year with enough known
observations for the kNN to make accurate predictions. For all
other years, the method consistently classified unknown redds
as steelhead, thereby overinflating steelhead abundance at
the regional scale (Tables 2 and 3).
Recommendations and Next Steps
●Between the two species assignment methods examined, we
recommend using the kNN algorithm over the G & G equation
for populations at the southern end of the CCC ESU/DPS.
●Relaxing the 100% certainty criteria applied in our study (e.g.
incorporating information on live fish and/or carcass
observations in the “area” of the redd) may be useful in some
cases.
●Resources should be devoted to the development of alternative
kNN approaches that utilize predictors beyond time and space
(e.g., redd morphology).
Pot length
Pot width
Tailspill
width 2
Tailspill length
Tailspill
width 1 Two steelhead on a redd.
1Redd data from Goin et al. 2015. Escapement estimates for Central California Coast coho salmon (Oncorhynchus kisutch) and steelhead (Oncorhynchus mykiss) in
coastal San Mateo and Santa Cruz County streams. Scott Creek data are partial as the complete spawning run was not surveyed. Report submitted to the California
Department of Fish and Wildlife in partial fulfillment of FRGP Grant Agreement Number P1230418.
2 Derived using spawner to redd ratio calibrations developed at the Scott Creek life Cycle Monitoring Station (see table 2).
Figure 2. Mouth of Scott Creek (A) closed by sandbar formation (barrier
beach) during the summer and (B) opened following the onset of winter
storms. Panel C shows the temporal overlap of coho salmon and steelhead
on the spawning grounds (spawn year 2015).
A. B. C.
Figure 3. The proportion of redds encountered during redd surveys in Scott
Creek attributed with 100% certainty to coho salmon (red) or steelhead
(blue), versus those of unknown origin (gray), 2014–2018. The number
above each bar is the total number of redds in a given spawn year.
This study was conducted in Scott Creek (Santa Cruz, County), a
regional Life Cycle Monitoring Station which has a weir to enumerate
returning adults. Our steps included:
1) Completed Spawning Ground Surveys (SGS) in all 25 km of
anadromous habitat, every 7–10 days, using standard protocols
(e.g., Gallagher and Knechtle (2005) and Gallagher et al. (2007)).
2) Assessed the performance of the two commonly applied species
assignment methods (discussed above) to redds which had fish
on them (100% certainty).
3) Applied the species assignment methods to unknown redds.
4) Estimated spawner to redd ratios (S:R) using both assignment
methods and applied the resulting ratios to regional redd count
data in San Mateo and Santa Cruz County streams.
A. B. C.
Table 2: Escapement estimate derived at the Scott Creek LCMS and the
corresponding spawner to redd ratios generated using G & G and kNN.
Acknowledgements:
We are extremely grateful to Seth Ricker (CDFW) for useful discussions about kNN and for
his help with the analysis. Additional thanks go to the Scott Creek field crews and volunteers
who helped with weir trapping and SGS. We wish to thank Big Creek Lumber Company and
California Polytechnic State University’s Swanton Pacific Ranch for providing land access.
Financial support was provided, in part, by NOAA Fisheries West Coast Regional Office and
California Department of Fish and Wildlife’s Fisheries Restoration Grant Program.
A.
kNN
Species NS:R S:R
Coho salmon 163 ±12 1.43 1.22 - 1.64 1.30 1.12 - 1.49
Steelhead 86 ± 3 1.13 1.05 - 1.21 1.32 1.23 - 1.41
95 % CI
G & G
Spawner to redd ratios (S:R)
± 1 SE
Estimated
escapement
95 % CI
Coho salmon
Spawn Year
2014 2015 2016 2017 2018
Number of redds
0
20
40
60
80
100
120
140
Steelhead
Spawn Year
2014 2015 2016 2017 2018
Number of redds
0
20
40
60
80
100
120
140
G & G
kNN