Report

An exploration of alternative methods to deal with time-varying selectivity in the stock assessment of YFT in the eastern Pacific Ocean Alexandre Aires-da-Silva and Mark Maunder CAPAM – Selectivity Workshop La Jolla, USA, 11-14 March, 2013 Outline • Background on YFT assessment Stock Synthesis (SS3) model Selectivity issues: time-varying process Retrospective pattern in recent recruitments • Explore SS3 approaches to deal with timevarying selectivity Ignore time-varying selectivity (base case model) Full time-varying selectivity (deviates) Time-varying for terminal years only YFT fishery definitions Baitboat Unassociated Longline 40 40 40 30 30 30 20 20 20 10 10 0 0 11 5 10 12 10 10 0 10 10 6 20 20 20 30 30 30 40 40 150 140 130 120 110 100 90 80 70 40 150 140 130 120 110 100 90 80 Floating Objects 40 30 30 20 20 4, 16 7 10 2, 14 0 140 130 120 80 70 8 0 3, 15 10 150 Dolphin 40 10 70 9 10 1, 13 20 20 30 30 40 40 150 140 130 120 110 100 90 80 70 150 140 130 120 110 100 90 110 100 90 80 70 YFT Stock Synthesis model • • • Quarterly time-step model Fishery definitions: 16 fisheries Data weighting: the CV of the southern LL fishery was fixed (0.2), others estimated (NOA, DEL) Growth modeling: Richards curve, L2 and variance of length-at-age are fixed • • Modeling of catchability and selectivity: Catchability coefficients for 5 CPUE time series are estimated (NOA-N, NOA-S, DEL-N, DEL-I, LL-S) Size-based selectivity curves for 11 of the 16 fisheries are estimated (fit to size composition data) Logistic selectivity for LL-S and DEL-S, and dome-shape for other fisheries YFT size selectivity OBJ time-varying selectivity? F1-OBJ_S F2-OBJ_C F3-OBJ_I F4-OBJ_N OBJ LF residual pattern F1-OBJ_S F2-OBJ_C F3-OBJ_I F4-OBJ_N Retrospective pattern Projections CATCHES SPAWNING BIOMASS Purse seine Longline Numerical and convergence issues • Unstable selectivites (OBJ) Sensitive to initial parameter values and phases Long run times (> 4 hours) Issues inverting hessian matrix (steepness run) Objectives of study • Test approaches available in SS to deal timevarying selectivity Improve selectivity process (time-varying) Minimize retrospective pattern Shortcoming: more parameters, longer run times • Simplify model Less data, collapse fisheries (OBJ) • Some considerations We assume that retrospective pattern is mainly driven by model misfit to recent OBJ LF data caused by misspecified selectivity We recognize that other sources of bias and misspecifcation may exist A single “lumped” OBJ fishery F1-OBJ_S F2-OBJ_C F3-OBJ_I F4-OBJ_N Model 0: Constant selectivity • Selectivity: Estimate “average” constant selectivity • Data: Fit to OBJ length-frequency data for all historic period • Base case model configuration 1.2 1.0 Selectivity 0.8 OBJ-F1 - SAC3 OBJ-F2 - SAC3 0.6 OBJ-F3 - SAC3 OBJ-F4 - SAC3 0.4 OBJ - lumped 0.2 0.0 0 25 50 75 100 125 Length (cm) 150 175 200 Model 0: Constant selectivity Model 1 - Full time-varying selectivity • Selectivity: Quarterly time-varying selectivity • Estimate quarterly deviates on base selex parameters of double normal OBJ selectivity curve • Data: Fit to OBJ LF data for all historic period • SD of quarterly deviates need to be defined: First run: freely estimate devs with high flexibility (SD=1) Second run: Use SD of estimated devs from first run in penalized likelihood approach Model 1 - Full time-varying selectivity OBJ Fishery 1.2 1.0 Selectivity 0.8 0.6 0.4 0.2 0.0 0 25 50 75 100 125 150 175 200 Length (cm) Paramter P1 - peak P2 - top P3 - ascending P4 - descending M1-P2fixed 0.13 fixed at -15 0.55 1.03 M1-P2est 0.14 1.08 0.51 0.41 Model 1 - Full time-varying selectivity Model 1 - Full time-varying selectivity Model 1 - Full time-varying selectivity Constant selectivity model 0 Time-variant model (M1-P2fix) Model 1 - Full time-varying selectivity 400,000 Mod 0 - cons. Selex (BC) 300,000 Mod 1 - Full tvar selex 250,000 200,000 150,000 100,000 50,000 0 2006 2007 2008 2009 2010 0.40 2011 2012 2013 2014 2015 Year 0.35 Spawning biomass ratio Recruitment (x1000 fish) 350,000 0.30 0.25 0.20 0.15 Mod 0 - cons. Selex (BC) 0.10 Mod 1 - Full tvar selex 0.05 0.00 2006 2007 2008 2009 2010 2011 Year 2012 2013 2014 2015 2016 Model 2 – “hybrid” approach • Recent period is the most influential on management quantities (recent recruitments, Fs) • Time-varying selectivity process in recent period only • Estimate quarterly deviates on base selex parameters of double normal OBJ selectivity curve • Fit to OBJ LF data for recent period only 3 terminal years (3-year average used for management quantities) 5 terminal periods (a longer period) • As for early period, fix to “average” constant selectivity from terminal years (base parameters) Model 2 – “hybrid” approach Tvar selex- 3 years Tvar selex - 5 years Model 2 – “hybrid” approach Tvar selex- 3 years Tvar selex - 5 years Model 2 – “hybrid” approach 350,000 Mod 0 - cons. Selex (BC) Mod 1 - Full tvar selex 250,000 Mod 2 - Hybrid 3 yrs 200,000 Mod 2 - Hybrid 5 yrs 150,000 100,000 50,000 0 2006 2007 2008 2009 2010 2011 2012 2013 2014 Year 0.40 0.35 Spawning biomass ratio Recruitment (x1000 fish) 300,000 0.30 0.25 0.20 Mod 0 - cons. Selex (BC) 0.15 Mod 1 - Full tvar selex 0.10 Mod 2 - Hybrid 3 yrs 0.05 Mod 2 - Hybrid 5 yrs 0.00 2006 2007 2008 2009 2010 2011 Year 2012 2013 2014 2015 2016 2015 Conclusions • Allowing for OBJ time-varying selectivity helped to minimize retrospective pattern in recent YFT recruitment estimates • Balance between the amount of selectivity process (# of params.) needed in the model and the OBJ LF data to include in model fit (whole series or few recent years only?) • Allowing for time-varying selectivity (quarterly deviates) in terminal years of the assessment only while fitting to LF data for this period seems a reasonable compromise • An “average” constant selectivity curve is applied to the early period while not fitting to the LF data for that period • A simulation study is needed to more rigorously investigate selectivity issues and associated bias in the YFT assessment QUESTIONS? Total catches 500 000 450 000 LL 400 000 LP 350 000 DEL NOA Catch (t) 300 000 OBJ 250 000 200 000 150 000 100 000 50 000 0 1975 1980 1985 1990 1995 Year 2000 2005 2010 Models Fix selectivity • Assume “average” stationary OBJ selectivity • “Drop” (not fit) all OBJ LF data • Fix to base selectivity parameters estimated in full time-varying runs (models 1) Models Fix selectivity M2-P2fixed M2-P2est Models Fix selectivity M2-P2fixed M2-P2est Recruitment – all models 700,000 SAC3 600,000 M1-P2fix Recrutiment (x 1000 fish) M2-P2fix 500,000 M3-P2fix_3YRS M3-Pfix_5YRS 400,000 M4-Pfix_Tblocks_5YRS 300,000 200,000 100,000 0 1975 1980 1985 1990 1995 Year 2000 2005 2010 2015 SBR – all models 0.8 SAC3 0.7 M1-P2fix M2-P2fix 0.6 0.5 M3-P2fix_3YRS M3-Pfix_5YRS SBR M4-Pfix_Tblocks_5YRS 0.4 0.3 0.2 0.1 0 1970 1980 1990 2000 Year 2010 2020 Model type 1 a) MODELS 0 and 1 Model 0 Fit to OBJ LF Base sel params Devs MANAG QUANT msy Bmsy Smsy Bmsy/Bzero Smsy/Szero Crecent/msy Brecent/Bmsy Srecent/Smsy Fmultiplier SAC3 Yes Estimated No 262,642 356,682 3,334 0.31 0.26 0.79 1.00 1.00 1.15 Yes, all period Estimated No 262,852 348,836 3,208 0.31 0.25 0.78 1.04 1.07 1.20 MODEL 1 CONFIGURATION M1-P2fixed M1-P2est Yes, all period Yes, all period Estimated Estimated Yes, all qrts Yes, all qrts 255,597 353,123 3,304 0.31 0.25 0.81 0.87 0.90 1.07 260,027 348,560 3,203 0.30 0.25 0.79 0.91 0.91 1.05 Model type 3 c) MODELS 3 Fit to OBJ LF Base sel params Devs MANAG QUANT msy Bmsy Smsy Bmsy/Bzero Smsy/Szero Crecent/msy Brecent/Bmsy Srecent/Smsy Fmultiplier MODEL 3 CONFIGURATION M3-P2fixed-3yrs M3-P2fixed-5yrs Yes, last 3 yrs Yes, last 5 yrs Estimated Estimated Yes, last 3 yrs Yes, last 5 yrs 261,728 350,789 3,278 0.32 0.26 0.79 0.99 0.99 1.14 257,126 351,377 3,273 0.31 0.25 0.8 0.84 0.86 1.03