Progress on Objective 3-5: Monitoring Designs, Trade Offs, and Recommendations Dan Rawding (WDFW) & Jeff Rodgers (ODFW) February 23, 2012
ISTM Objectives 1. Identify & prioritize decisions, questions, and objectives 2. Review existing programs and designs 3. Identify monitoring designs, sampling frames, protocols, and analytical tools 4. Use trade-off analyses to develop recommendations for monitoring 5. Recommend implementation and reporting mechanisms
Other Objectives Objective 3 Identify monitoring designs Sample frames Protocols (AFS) Analytical tools (WDFW DAGS) Objective 4 Trade offs Objective 5 Recommendations & Reporting
Objective 3: Identify monitoring designs, sampling frames, protocols, and analytical tools As part of Objective 2 we identified monitoring designs Juvenile designs are indices (parr abundance), distribution, and outmigrants Few well defined methods Concentrate on adult designs
Adult Designs Weirs and mark-recapture designs are well documented, but have limited application in many LCR situations Small populations Protracted migration/spawning times High turbid river Higher cost Spatial/Temporal sampling designs Area-Under-the-Curve Redd based estimates
Adult Designs 1. Define the spatial extent of the sampling universe (frame) 2. Evaluation of different sampling designs based on LCR data (census of redds or AUC estimate by 50-100m reaches)
Sample Frame Lower extent of spawning is well documented due to easy access and many surveys in lower reaches Upper extent is based on professional opinion (WDFW distribution model) or simple gradient model (NOAA) Transparent science based approach to determine upper extent & spawning distribution In collaboration with Steve VanderPleog, Bryce Glaser and Andy Weiss.
Field Methods During the spawning season determine upper most extent of distribution for Chinook, steelhead, & chum using census or random draws Provide training on redd & species identification & enumeration and use GPS to record all locations and start and end point of surveys Define reaches to sample including a GPS coordinate for end of upper extent surveys or survey > 400m above last observed redd
ARC-GIS Acquire 10m Digital Elevation Model (DEM) and precipitation data (PRISM) Use ARC-GIS, and DEM Analysis software (Bld_Grid, and Netrace) to delineate a stream network, and develop unique ID for each 100m stream reach (260,506 reaches) Determine mean annual flow, basin size, mean reach gradient, max reach gradient, elevation, and precipitation for each reach
Classification of Upper Extent
Statistical Analysis Logistic regression log(p/(1-p)) = Bo + B1*cfs + B2*grad + B3*elev Construct profile of 4 * 100m reaches above upper most redd and 4 * 100m below upper most redd 56 suitable steelhead profiles Assumptions Similar # of points with presence & absence Normality and limited correlation between coefficients Percent Correct Classification, Sensitivity, Specificity Goodness of fit (AUC & R 2) AIC for model selection Cross validation
The Gradient Problem River 20 m wide Stream 3 m wide each square represents a 10m cell so barriers are not discernable in small watersheds
Summary Statistics count 56 Exact 37.5% Under 32.1% Over 30.4% mean 211 median 0 Measure Value sensitivity 0.69 specificity 0.82 pcc 0.76 kappa 0.52 auc 0.85 R-Squared 0.44 pchi 0.82 Summary Statistics and Goodness of Fit for Logistic Regression Frequenc 24 18 12 Historgram of Error Distances for LCR Modeled Streams 6 0-1000 predicted -600-200 200 600 1000 1400 1800 Error Distance(m) 2200 2600 Confusion Matrix observed 1 0 1 135 37 0 62 179
Habitat Protection Using Sensitivity of 0.95
Other species Repeated for other species (coho, Chinook, and chum) For coho salmon juvenile use electroshocking before downstream emigration Chum abundance dropped from 500,000 to less than 25,000 (5% of historic levels. So, little data available for modeling.
Distribution Model Results Same attributes (flow and gradient) provided best model for Chinook but flow, gradient, and elevation were significant variables for coho. Appears Chinook spawning distribution is variable based on flow and possible abundance. Less annual variability in steelhead distribution. GIS attributes from same model (NETRACE) available for Oregon and so are the WA equations
Sample Frame Questions!
Sampling Designs Two papers by Courbois and Jacobs explored sampling designs for Sp/Su Chinook and bull trout based on redd surveys. However, surveys did not occur over the entire spawning period, so it is unlikely their data was census. Work in progress w/ Martin Liermann (NOAA)
LCR Census Data 2005-11 winter steelhead redd census in Mill, Abernathy, and Germany creeks. Hatchery release on Abernathy Cr & weir operation at RM 4. 2003-4 & 07-10 tule fall Chinook salmon redd census in Coweeman River 2004-06 tule fall Chinook salmon redd census in EF Lewis River 2010 coho salmon redd census in Mill, Abernathy, and Germany creeks
Objectives Compare bias, precision, & interval coverage for SRS, SYS, GRTS sampling Evaluate stratification of same designs Evaluate ratio estimator based on sampling index reaches & near peak spawning time sample entire frame. Peak counts are expanded by the ratio of seasons redds visible in index based on peak survey date Evaluate sample frame on bias, precision, & interval coverage
2008 MAG Steelhead Redd Census Clumpiness of data (map) SRS Low Sample fraction>>high uncertainty Sample fraction needs to average 50% to have average CV < 15% Does not include uncertainty with females per redd and sex ratio
Design performance for 30% of reaches using reach length = 500m as measured by 95% confidence interval coverage. Percentage is based on 10,000 iterations Ratio Stratified SRS Years (2005 11)
Root Mean Squared Error vs Sample fraction Mean Squared Error = Bias 2 + Variance Simple Random Averaged across 7 years Stratified Stratified + ratio
Bias vs Sample fraction Simple Random Stratified Stratified + ratio
Coverage vs Sample fraction Stratified Simple Random Stratified + ratio
2011 M-A-G Redd counts by 500m section Data follow Negative Binomial distribution (p-value 0.44) Klumpiness causes problems for application of central limit theorem (normal approximation) to estimate variance # of reaches 90 60 30 0 0 2 4 6 8 10 12 14 Redd Count 16 18 20
Preliminary Steelhead Summary SYS and GRTS designs performed best as measured by RMSE, coverage, and bias. Hatchery operations may increase clumpiness & thus stratification based on historical reach usage improves RMSE, coverage, and bias. Ratio estimator shows promise Variances based on central limit theorem provided lower coverage because data are not normally distributed Coverage improves with zero inflated, negative binomial, or other distributions
Next Steps Evaluated nearest neighbor variance estimator used in GRTS. Complete analysis for other species and populations. Evaluate different sample frames Used first GRTS master sample draw for coho salmon, which was not dense enough to meet rotating panel. LCR GRTS master sample tool being improved and WDFW will use for non-census redd, AUC, and juvenile sampling designs.
Objective 3 Monitoring Design Evaluation Sample Frame Analysis Tools (DAGs) Write it up! Questions?
Objective 4: Trade Offs Abundance Estimates Accuracy Method Sampling Design Cost Census Complete Mark-Recapture Peak/Supplemental Area-Under-the-Curve Peak Count Expansion Random Redds
Objective 4: Trade Offs Cost effectiveness Maintain and improve multispecies approaches to monitoring ODFW & WDFW continue to partner to address common needs (recruitment, critical uncertainties especially for coho and winter steelhead monitoring) Develop cost/distance algorithm for adult sampling designs Trade offs for coho biological sampling
Objective 5: Recommendations Data Management & Reporting Cedric &Danny
Recommendations Recruitment (ODFW & WDFW) CWT groups for integrated hatchery, wild broodstock, or wild smolts for coho & Chinook Continue commercial genetic sampling of wild steelhead released and chum retained to estimate fishery impacts by ESU, strata, or population Expand creel surveys for tributary steelhead and chum fisheries along with CRC to estimate incidental fishery impacts.
Recommendations Index of Parr Abundance & Distribution Increase parr sampling in (ODFW) and adopt WDFW adopt OR parr sampling program Spawner Abundance, Timing, Distribution Address critical uncertainties, bias, and precision by developing specific females per redd, redd duration, observer efficiency, and spawning life estimates especially for coho and winter steelhead (ODFW & WDFW)
Recommendations Age sampling at Barrier Dam (WDFW) Ultrasound or other methods to more accurately determines sex of immature spring Chinook and summer steelhead
Approaches to be developed based on trade offs Diversity (age, origin, sex) Winter steelhead are a challenge since they are repeat spawners Small salmon populations make it challenging to achieve recommended sample sizes Spawning distribution data is unavailable for mark-recapture & weir programs Adult & juvenile abundance, distribution, and diversity estimates are unavailable for turbid systems
Acknowledgements Jen Bayer and PNAMP staff Biologists from ODFW and WDFW Bernadette Graham Hudson (LCFRB) Russell Scranton (BPA)