Home

"user manual"

1. 25 A An example The Northern Cod case study 27 B General description of all entries of the model 45 C Some ninis In qmodelTItlllg snc trot Ree talit uae hee oe cute bent edu aa ut de tenu ato 51 1 Introduction 1 1 General Overview of the Program Coleraine Coleraine is a user friendly general age structured model for fisheries stock assessment It combines a familiar Excel environment with a general and powerful AD Model Builder Otter Research Ltd 1999 application Coleraine has a statistical age and sex structured model with a very general structure It allows for several fisheries to be modeled at once and can be simultaneously fitted to many different sources of information like catch at age and or size data from the fishing fleet and surveys and several indices of abundance commercial fishery and survey The estimation is performed using maximum likelihood theory in a first step and a Bayesian approach in a second Prior information on model parameters may be incorporated given the Bayesian framework of this statistical approach Uncertainty around the estimates of the derived parameters of interest can be assessed directly from the bayesian posteriors Once the model is fitted Coleraine allows the user to do policy evaluation by assessing the consequences of different harvest strate
2. 126 1 15 15 D 01 l 127 Error variance L 128 1 15 15 0 01 0 Figure A 1 4 Parameter information required by the estimation model The different columns represent 1 phase number 2 lower bound of the parameter 3 upper bound of the parameter 4 type of prior distribution 5 mean 6 standard deviation 7 initial guess 99 Priors 129 Error variance A 130 i 15 15 i i 01 131 Logg CPUE 132 1 15 15 0 1 55 133 Lag q CPUE error 134 1 5 5 n T 0 6 l 135 Log q Surve pe o 3 0 5 0 0500 dne l 137 Survey L Full es 14 0 Qd 06 fi 138 Survey L full delta wo 1 m4 0 m9 06 l 141 Survey variance L 142 5 15 15 0 0 06 1 143 Survey variance Fi 144 1 15 15 0 6 10 145 L variance 1 146 1 15 15 0 6 1 33 147 L variance n 142 15 15 n T 0 6 816 143 Dummy variable keep Far error troubleshooting 150 1 15 15 0 6 Figure A 1 4 continued Parameter information required by the estimation model Coleraine Users Manual 31 B E pr F 1527 153 Likelihoods 0 not used 1 norm 2 lognorm 154 155 CPUE likelihood Type 561 2 157 Commercial catch at age likelihood type 168 12 153 Commercial catch at length likelihood 1 160 161 Survey likelihood type e2 2 163 Survey Indes type 1 weight 2 numbers w 165 S
3. 715 50 Hilborn et al DATA Total catch in weight by year One row for each commercial fishing 1 method It needs to match with the total number of year of the estimation model 21 Totalnumberofobserved CPUE datapoints 0060600000 CPUE matrix First column Index number second column method 3 number third column year 4 digits fourth column point estimate fifth column coefficient of variation 57 m index matrix First column Index number second column sf year 4 digits third column point estimate fourth column coefficient of variation 6 HN umber of observed commercial catch at age datasets Commercial catch at age data The sample size will weight the likelihood First column method number second column year 4 digits third column sample size others Starts at age 1 of females and goes to pool age In the next entry to the right it starts with the 1 year old males and goes to pooled age These values are proportions that for both sexes combined add up to 1 one 8 o Totalnumber of observed survey catch at age datasets Survey catch at age data First column index second column year third column sample size others Starts at age 1 of females and goes to pooled age In the next entry to the right it starts with the 1 year old males and goes to pool age These values are proportions that for both sexes combined add up to one Total
4. Not used in the model 23 E NEN Uu This dummy parameter is not involved in the calculations and must be y 22 nV excluded from any estimation phase It is useful for doing deterministic projections no estimation involved and to be able to start in any iteration mceval from the maximum likelihood estimates LIKELIHOODS The different entries for the likelihoods are as follows Not used but likelihoods and predictions are evaluated in last run Not used Normal Lognormal Robust normal Robust lognormal Robust lognormal for proportions 1 Likelihood type for the CPUE data from 1 to number of CPUE Coleraine Users Manual 49 5 indces 870N HUB Likelihood type for the commercial catch at age datasets from 1 to n methods H Likelihood type for the commercial catch at length datasets from 1 to n methods D Likelihood type for the Survey indices from 1 to n Survey indices 6551 The survey indices are in weight 1 or numbers 2 6 survey vulnerability is based on age 1 or length 2 Likelihood type for the survey catch at length not separated by Sexes Likelihood type for the survey catch at length separated sexes 1 to n Survey indices Likelihood type for the survey catch at age separated by sexes 1 to n Survey indices FIXED PARAMETERS Rows 1 10 are sex specific parameters if using a model separated by sexes First column female
5. Bayesian posteriors for the derived parameters of interest are obtained by using the Markov Chain Monte Carlo MCMC method Otter Research Ltd 1999 To run 20 Hilborn et al MCMC go to MCMC gt Run MCMC Figure Section 3 4 The program will start by fitting the model to the data using maximum likelihood To continue with the bayesian part the variance covariance matrix must be available An example on how to use the MCMC and Mceval routines is shown in Appendix 1 4 5 Policy Evaluation After the estimation is completed different policies can be evaluated The necessary parameters for running the projection under different assumptions are incorporated in the Projection Parameter section during the input of the data Once MCMC is executed the colera20 psv should be in your directory It contains important information on the uncertainty of the current year biomass which is used in policy evaluation One nice feature of Co era20 exe is that the user can modify at any time the projection conditions without having to rerun MCMC To run Mceval go to MCMC gt Run mc eval You can do this immediately after running MCMC or at any time in the future After running Mceval some new text files will be generated in your working directory project out Contains the posteriors of the virgin vulnerable biomass virgin Spawning biomass projected time series of spawning biomass and vulnerable biomass for each chosen harvest strategy level vb
6. CPUE Master j T K FALT eres AEomsel c 71111 71111 T Figure A 1 11 Graphs of predicted and observed survey index and CPUE Selectivity of the survey and commercial gears are presented in the Sheets SurvSel and ComsSel respectively One of the most useful features of Coleraine is the possibility of doing multiple successive runs and organizing and summarizing the output information The Excel sheet called Tracker stores vital information for the different runs Figure A 1 12 such as file path name of input files likelihoods parameter values and phases Tracker gets updated during the redraw process of the output viewer 5 Updating the Information of Re runs Once the new entries of the parameters and likelihoods have been specified for another run then steps 3 and 4 can be repeated Each new run can have a different name so that you can return to that file at any time The only variation in this repeated sequence of events is the use of the View Re run of Open file command This will just update the graphs and not create new output files 40 Hilborn et al MAIN MEMU Create or Modify Input File Run Model k View Cutout View Mew Output File MCMC k View Re run of Open File 0 200 zz 40 mE p Cm mw 2 dw a ni between charts Starting Position 1st view Other Output File After updating the output file 3 or 4 times the program might crash because of m
7. Ne 57 s 1 e 1 vi uw PU Uncertainty in the initial age structure is incorporated by using log normal error 2 n 57 Nj N e where N 0 0 for a 22 A 1 The plus group has an independent error component with its own variance where P stands for plus group and for initial 1 2 3 Stock recruitment Recruitment follows a Beverton Holt stock recruitment relationship with log normal error structure S Ro m C t t R 2 a BS where is the recruitment residual for year t pe N 0 o and S is the spawning biomass in year t computed as S w o N where maturity ogive is the fraction of females that have reached maturity at age a a and wi is female weight at age and time Recruitment at equilibrium in the absence of fishing equals Mace 1994 whereSpR gt wf De MED where SpR is the spawning biomass per recruit a function of the surviving proportion weight at age and maturity ogive of females The model was re parameterized with a 4 Hilborn et al steepness parameter z the proportion of the virgin recruitment that is realized at a spawning biomass level of 20 of the virgin spawning biomass Francis 1992 Thus both parameters can be formulated as a function ofz R and SpR pcs l 4z R zmi p 4zR S R SpR Low Steepness Recruits Spawning Biomass This graph shows the Beverton Holt
8. caleraz Bb 2 e caleraz r z 38 North cod is bestLF Find Files that match these search criteria File name Case Text or property Find Mew Files of type fal Files Last modified any time New Search 42 Found 42 Hilborn et al During the execution of this program you will see the following DOS window MS colera O0 ed value 36519 mi mun L ua fom x em C c5 dI e acceptance T DL oL Gam 22 i al Paz KA H ffs Tu 1 lmn mE Dun a HET min mo Paz KA B Im naim 24 jn r rar onim m Gl T St al zt aL eti 24 jn rm iw URL m rz m Gl a Gl u E i gt E m Ll kuL ul k i9 m Lal ul m VA fi m to E RE d Mj t JT n m Lal rate rl Gl Ree eee Uu mp Wi RI LAL Tl rM rM ra Era mErE mra ru iw D ei 59 gg n gp enje ie n 2 ii m E wo IJ Uz m EPT ance rate This window displays the simulation number and the acceptance rate This statistic is a diagnostic element for convergence and should tend towards values between 0 15 and 0 4 When the program is done with the simulations it will generate a colera20 psv file This file contains values of all the model parameters collected from the Markovian chain The
9. 0 0003263 0 0072 266 1 1363 R 0 nd40651 0 01615 67 1 1964 n 0 0 0074687 0 05023 2685 1 1965 50 0 0 0002402 0 01463 269 1 1366 50 0 0 00219377 0 03772 270 1 1967 R 0 0 0018262 002525 er1 1 1368 R 0 0 0004563 0 00374 fre 1 1363 50 0 0 000118 0 00837 273 1 1970 50 0 0 0153283 0 04494 274 1 1971 R 0 0 000086 0 02355 275 1 1972 h 0 0 0006015 O 01718 226 1 1973 h 0 01257 o 1 13 4 50 0 0 0017937 niz28 era 1 1975 50 0 0 0023364 0 02207 279 1 1376 R 0 0 0001015 0 09375 260 1 1977 R 0 0 0007743 0 05174 261 1 1378 50 0 0 0 01427 292 1 1979 50 0 0 0 01179 265 1 1980 50 0 0 0003316 0 02556 264 1 1321 R 0 02543 265 1 1982 50 0 01373 286 1 1383 50 0 0 0001458 0 02131 297 1 1384 50 0 0 0000246 0064 265 1 1985 50 0 00421 269 1 1385 R B4ZE 06 000533 290 1 1387 50 0 0 0002738 0 071549 291 1 1388 50 0 00001513 0 01622 232 1 1333 50 0 0 000054 0 01145 293 1 1330 R 0 0 0003727 0 04944 234 1 1331 Fil n d2484 0 02208 G 0 0427 0 07753 0 07128 0 05113 0 17708 0 18002 0 14353 0 07658 0 1492 0 15646 0 20351 0 13969 0 05015 0 07844 2282 0 47002 15335 0 12643 0 12177 0 08347 0 25325 0 11223 0 12169 0 10373 0 0377 0 06131 0 08853 0 11311 0 26064 0 22468 Figure 1 7 continued Data section for the Northern cod case study 34 Hilborn et al E E Li E F u H 295 Msurvey CA 251 a 287 DUMT DATA Survey catch at age data sur
10. Forward Projections under Different Harvest Policies Organizing the Projection Outputs su About the data The data used in this exercise corresponds vvith a cod population Gadus morhua of the North Atlantic The exercise consists of a time series of landings in hundreds of metric tons starting in 1962 and ending in 1991 The catch at age data cover this entire time range and are represented by individuals from age 2 to 20 Other sources of available information to which the model will be fit to are a CPUE index 1978 to 1988 coming from the fishing fleet and relative index of abundance 1981 to 1991 originating from a series of Research Trawl Surveys Some biological information such as weight at age and natural mortality is available or assumed known for this stock 1 Creating the Input Sheet The first step is to unzip the file Coleraine zip in a directory on your hard drive We recommend using a folder with a short path sequence e g C Coleraine Cod which will prevent some potential problems with the interface You should have by now four Excel files in your directory maincode InputTemplate newgraph and tracker and one executable colera20 exe Open maincode xls and enable the macro option Go to the new created menu button MAIN MENU and select the Create New Input File option Figure A 1 1 MAIN MENO Create or Modify Input File Create Mew Input File Run Model Open Existing Excel Format Input File View
11. Output Reconstituke Excel input File From text File MEME Check Format Before Running Model Check Parameter Values Before Running Model 28 Hilborn et al Figure A 1 1 Menu option for creating an Input sheet This command creates a new Excel sheet that contains spaces for dimensioning parameters These entries will structure the rest of the Input sheet in a future step Start typing in the entries as specified by Figure A 1 2 E E E E A E E 1 INPUT FORM FOR GENERA 26 4 End Year f ages 50 Humber gf commercial catch at length data set g ni Ail er che ew fec sections BERE B 3 28 Mlasimum age Number of weight at age data sets 4 Debugger 1 0 off 23 201 S3 30 Age at which pooling occurs 54 Number of surveys all indeces B 31 Hun Options 32 7 First age for catch at age da Mumber of survey catch at age data sets 8 33 Era NIN 8 1 Use Weight at age data 34 First Length For catch at length 58 Mumber of survey catch at length data sets l 35 11 2 Use pin file instead of dat Fi 36 Length increment For C oL dat EU Number of survey no sex catch at length data 12 32 13 3 Initial gear 0 read in 1 qea 36 Mumber of length classes for 62 Survey vulnerability type for selectivity curves 14 33 15 4 Gear used in projections z 40 Length at which data are poole B4 Age at full recruitment a starting number for de
12. Slareoe Le m t TEN 14 2 1 Hardware 5 14 2 2 Tteequilelmebls 14 2 3 Outline of Working 4 14 24 n aa ii 14 2 5 File Management aa ada T UM 15 3 Getting Your Data into Coleraine 888 88 88 83 8 8 16 97 General SUES a ll Ra a aila 16 Sues Detal Valugs 3 c AR aaa lin 16 oe 16 32A Ne Maini MENUT ere arabada 16 3 5 Step By Step Procedure for Entering Data 17 4 Running Colera 7 E E E 19 19 42 2 Parameter ESUmallobi su 19 4 3 Phases in the Estimalonuuuumuau al d lar aya 19 4 4 Obtain Bayesian Posteriors iltud eei eie qp rettet ab Bay aad 19 4 5 Poliey Evaluar xssasciie i y l gab t 20 DP TOCCSSING OUMU p T 21 MO I M Pc 21 5 2 Text Output Files Produced by Colera20 exe 21 5 3 Using the Viewer Part 1 Create or Open a Results File 21 5 4 Using the Viewer Part Il Making Graphs 22 5 5 Storing important information of different runs in one file Tracker xls 23 silet
13. data and any penalties based on priors many people argue that this should not be called Maximum Likelihood but instead call it mode of the posterior estimation or MPE However if you think of priors as additional forms of likelihood then this is still an MLE fitting and it is really all semantics anyway unless you have some very strong and informative priors Thus will call it MLE even if it is really MPE Rule 1 Graph the data and the fit The first task in fitting an age structured model is to get the model to fit the data You will have two major ways of judging the goodness of fit the numerical values of the negative log likelihoods as found in the master worksheet of the viewer and Tracker and in the graphs It is difficult to overemphasize the importance of actually looking at the data to make sure it is fitting You can t look at a negative log likelihood and say whether the fit is good they can only be used relatively you must plot the data and the fits and see if you are coming close to the points Once you get close then you can start looking at the negative log likelihoods for fine tuning In general it is often difficult to see any differences in the graphs when the likelihoods change by only a few units Rule 2 Start with good beginning parameter values The key issue here is that you usually need to start with very good parameter starting values to ensure that the non linear search algorithms work well Gener
14. in E NorthCod l zi EE Cancel File name 1 Save as pe Text Files Coleraine Users Manual 37 After this is completed a DOS window will appear showing valuable information of phases number of estimated parameters total value of the objective function and gradients associated with each parameter XP colera 0 end of data sect ion Initial statistic Function value 6 War 1 iteration B function evaluat ig imum gradient pl sv rag Wa lue Gradient iar kt a ie I oo Eerans 5 H p Tl p a Lu 41 2853 ot l l Interme mln LI IT War Lab 114 i M function Eu oampaonent mag E ent illar Li ee LL m ui E rmi m Rut z F 4 ea Fi n i 1 a n E 2 m um mn n oF L 4 jJ Bm i e qu mq Pu Pu ation 4 ent component mag m tar 4 Creating an Output Viewer The output of the estimation model is a text file and as such it does not have a very user friendly graphic environment Coleraine was design to automate the creation of graphic outputs The creation of this viewer is done using the following commands MAIN MEMU Create or Modify Input File Run Model View Output MEMC emp ae ea A view R e run of Open File view Other Output File Save in NorthCod File name I pics anl cole
15. likelihood entries This allows us to turn on and off the different likelihood components in this case CPUE Survey Index and Commercial Catch at Age by using the specified values 0 likelinood component is ignored 1 likelihood component is ignored but the predictions and implied likelihood value is computed 2 lognormal error structure 12 robust multinormal for proportions Cells 164 and 166 determine the nature of the Survey Index and selectivity ogive respectively 30 Hilborn et al A B oO E F 5 33 Priors 100 107 RO Recruitment in virgin condition 1 10000 600 103 h steepness of spawner recruit curve 104 1 0 01 D 0 6 01 105 1 natural martalit 106 0 01 3 1 1 Oz 107 Log init dew prior dewiates For initial age structure uniform ar normal anl 105 108 loq rec dey prior funiform or normal onl fof CU SS 111 Initial 8 1 aldz in ur 11A0 untished 1 a o 2 o 005 il 113 Intial u exploitation rate For initial age structure O untished mS CTS 115 Plus scale 116 1 2 E il 117 S fullest Far length 10 80 093 001 5 113 amp full delta Far males as different from females 120 3 3 E l 121 Log variance of left side of selectivity curve by length Far bath zesez 122 123 Log variance of righthand side of double normal selectivity curve For bath si 124 125 Error 5 Full
16. model This arrangement allows the user to change the initial guesses of the input parameters and redraw the graphs in an iterative process 3 Running the Estimation Program At this point the estimation phase should be ready to start Line 5 contains a switch for including a debugger into the estimation program during the input process of the data If it is set to 1 than the user will need to type n and press return whenever 36 Hilborn et al the cursor prompts for some entry This basic mechanism allows us to some extent to trace the problem at runtime To run the program use the following command MAIN MENO Create or Modify Input File H a _ Run This File Run Other File Run Model view Output k MEME 0 n 1362 1367 1372 1977 1392 1387 1362 13567 1372 1372 1382 1387 Observed Survey Index Predicted Survey Index M Y 2 AA AB AC AD AE AF 4 1 2 CPUE gear 1 Index 1 Survey gear 1 Index 2 3 i 10 2500 E 2 2000 2 1500 8 B 9 1000 W 500 11 12 2 13 14 15 16 1 44 b H InputForm 4 Output NameList 4 Defaults Graphs Figure A 1 8 continued Deterministic model outputs before running the estimation model This process will invoke a dialog box where you can specify the name of the text file that contains all the information of this run e g Case1 Please name this run name must be 8 characters do not add file extension Save
17. natural step after the simulation is to generate the prediction given the desire harvest policy The program will ask the user to continue with the projection phase through the following dialog box Microsoft Excel lt asks the user again to input the text file coleraz0 b01 ia COLERAZO mem E caleraz b z bal coleraz p l Find Files that match these search criteria File name Casel 46 filets Found ee MR ee U Lu Files of type fal Files 7 Organizing the Projection Outputs The projection out file contains the values for the virgin vulnerable biomass virgin spawning biomass projected biomass and spawning biomass in each year for each of the chosen harvest levels 100 600 ER R p 1 1 Virgin vulnerable Biomass Virgin Spawning Biomass Biom 1332 Biom 1393 1440 25 1440 25 440 25 1440 25 1440 25 440 28 1440 25 4440 25 440 25 1440 25 4440 25 1440 25 1440 25 1440 28 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 544 153 620 44 1 515 553 412 565 303 07 7 206 2689 153 477 620 437 515 5643 412 561 303 0 73 206 205 153 475 620 436 515 547 Coleraine Users Manual 43 Spawnerz Catch 195 2024 37 567 915 202 179 185 806 153371 150 838 1076 23 39 7147 Tasata 62 5052 n3 r23r 57 823 1232 35 181 553 Figure A 1 13 Output file f
18. o The bounds and initial guess are in natural logarithmic scale Initial age structure residuals They go from age 2 to nages 1 They are normally distributed with mean 0 and variance o The bounds and initial guess are in natural logarithmic scale Fraction of R in the initial study year This parameter is defined in the Exploitation rate in the initial state This parameter is defined in the range 0 1 0 Residual term in the plus group at the initial state They are normally distributed with mean 0 and variance o The bounds and initial guess are in natural logarithmic scale Age of full selectivity for females of gear type of the fishing fleet full gil Age difference between sexes in the age of full selectivity of fishing fleet gear type Variance of the left half Gaussian distribution of the selectivity of gear The bounds and initial guess have to be in natural logarithmic scale Variance of the right half Gaussian distribution of the selectivity of gear 1 The bounds and initial guess have to be in natural logarithmic scale 13 Residuals of age of full selectivity by gear i at time f The number of S r parameters per gear to be estimated will depend upon the number of 48 Hilborn et al years catch at age data are available They are normally distributed with 7 mean and variance 14 s Residuals of variance of the left half Gaussian distribution of t
19. parameter estimates might be a function of the initial parameter values in the non linear search This is generally true for situations with a very flat joint likelihood surface and where it is likely that the minimizer gets stacked at different local minima Therefore we strongly recommend that you try different sets of starting values 4 3 Phases in the Estimation Trying to estimate all the model parameters simultaneously in a non linear model situation may not be advisable It is convenient to keep some of the parameters fixed during the initial part of the minimization process and carry out the minimization over a subset of parameters The other parameters are included in the minimization process in a number of phases until all of the parameters have been included Research Ltd The first entry first column of each prior in the Prior Section controls the phase in which the parameter will enter the minimization process A negative number implies that the parameter will not be estimated and the value of a positive number determines the phase in which a particular parameter enters the minimization process A suggested phase order would be the following read Appendix C Phase 1 q to fit the index data Phase 2 R to adjust the trajectory Phase 3 Selectivity parameters Phase 4 Recruitment residuals Phase 5 Changes in selectivity Phase 6 Deviation in initial conditions 4 4 Obtain Bayesian Posteriors
20. relationship formulated as a function of the steepness and virgin recruitment This parameterization is very convenient because z is clearly defined between 0 2 1 Note If the total catch landings and weight at age data are in the same units then R is going to be in numbers In other situations R should be multiplied by the ratio of the units of the catch to the weight at age in order to scale it to the correct units this is not a necessary task but is important for understanding the meaning of the units of the virgin recruitment 1 2 4 Growth Fish growth is modeled according to a von Bertalanfy model with mean size at age given by L D exp amp a t9 We assume that the distribution of size at age is log normal with standard deviation sd which is a linear function of mean size at age Coleraine Users Manual 5 S S S S L O Zh O sd E mn m n lico This is basically a linear interpolation between the standard deviation of the mean length at the first L and last E age The distribution of log L at age length age relationship by sex is symbolized by pllog L 122 and has mean wand standard deviation o respectively equal to Hi log L 2 E O s L The length proportions at age can be approximated as 2 ollog JA I l l ny S s2 S pliogz He Ja l 1 2 S 2 where A is the width of the interval in log scale Th
21. second males Proportionality parameter of the allometric length weight relationship Exponent parameter of the allometric length weight relationship Asymptotic length of the von Bertalanffy growth model Von Bertalanffy k parameter Von Bertalanffy t parameter Brody parameter Mean length of individuals of the first age Mean length of individuals of the last age Standard deviation of the Gaussian distribution that describes the variability around the mean length of individuals of the first age Standard deviation of the Gaussian distribution that describes the variability around the mean length of individuals of the last age Maturity at age of females in the population ogive Female fraction of the total that recruits every year Vulnerability at age by sex at the beginning of the fishery N ages x N ages upper triangular aging error matrix Number of weight at age dataset Annual weight at age First row year second row sex others weight at age starting from age 1 to age n If you do not have estimates of parameters 1 5 you will need to enter observed weight at age data from the first year to the last year 1 The same applies if you are using a model in numbers not in biomass units but you will have to set all weight at age values to 1 b A Sy L S i S OO S S k 9 1 5 L S t NO A 7 Q l amp N O AIOI
22. 16 410 50 4 17 42 Mumber of commercial fishing 66 Save this cell Create the Rest of the 18 Dimenzioning Paramete 43 1 13 44 Mumber of sexes Anata ci B8 20 1 Mumberof CPUE indeces td 450 1 69 zr 46 Total number of CPUE data pq 70 22 2 Mumber of survey indeces 47 23 48 Number of commercial catch 172 24 3 Start ear prusi he dolya 48 25 1962 60 Number of commercial catch 474 Figure A 1 2 Filling in the dimensioning entries of the nput Form for the Northern Cod dataset Given that no length related data are available at this time entries in rows 35 39 and 41 represent only dummy data In order to avoid run time errors you should enter a larger number in row 41 than in row 35 Once you are done click the Create the Rest of the Form button The program will create all the necessary spaces for data entry 2 Inputting the Data This section comprises the following subsections 2 1 Gear Type 2 2 Projection Parameters 2 3 Priors 2 4 Likelihoods 2 5 Fixed Parameters and 2 6 Data These will be reviewed in sequential order Coleraine allows us to specify the names of the different gear types and indices Figure A 1 3 Spaces for these entries are located in rows 72 76 These names will label several graphs in the output viewer and are very useful to keep track of gear type names in multiple fleet situations Figure A 1 3 also shows the Projection Parameters Thes
23. S84 2 PSSE 9 PSSE 8 8 8 8 1 4 5 Prior Prior information on main parameters of the model can be incorporated by using three different density functions uniform normal and lognormal If the parameter is being estimated active and has a pre specified prior the natural logarithm of this density will be added to the global objective function 1 4 6 Global objective function Parameter estimates are obtained by minimizing the overall objective function f log L penalties prior 1 5 Policy evaluation Coleraine can also be used to evaluate alternative management policies by simulating future stock trajectories n years into the future A Bayesian approach is Coleraine Users Manual 13 implemented to carry out this task The Monte Carlo Markov Chain MCMC technique is used to generate samples from the joint posterior probability distribution and the marginal distribution of any model and derived parameter can be readily approximated from it We use the AD Model Builder s Otter Research Ltd 1999 implementation of MCMC which is based on the Hasting Metropolis algorithm Gelman et al 1995 Two types of harvest strategies are implemented in Coleraine namely constant catch and constant harvest rate strategies Details on the entries can be found in section 6 1 The projections are carried out using one selectivity ogive s This selectivity pattern is computed as an average selectivity at age for the different fishing f
24. SAFS UW 0116 Revised May 2003 COLERAINE A Generalized Age Structured Stock Assessment Model USER S MANUAL VERSION 2 0 R HILBORN M MAUNDER A PARMA B ERNST J PAYNE P STARR University of Washington SCHOOL OF AQUATIC amp FISHERY SCIENCES CONTENTS b OL 1 1 1 General Overview of the Program Coleraine 1 1 2 General Overview of the Estimation Model 1 1222 ls ADUNG ANCE GYVNAIMICS DV SOX l b b lav ana taste 1 dud 2 TIAL CODO S c da 2 NS Molto e 72 ERES 3 UA 07 RR E ERO POOR TEN 4 1 2 5 Weight at age relationshibp 8 88 6 EZG 00 OLU m a n b d 6 END ERE TT 9 1 3 1 Predicted abundance indices 9 1 3 2 Predicted age and size composition 9 1 2 ODIECUVEe ELER a a o yas 10 1 4 1 Robust normal likelihood for proportions 11 Tp 7101710 0021410 2 6191105500 11 TAS Total JIKOlllIQOQ cie a acd d nadine 11 12 POTUM O a DU ER 12 1 4 6 Global objective function 12 to ONCY CV AIAN O Ipsec TT 12 2 cetin
25. action recruiting of each sex 1in a 1 ze model 0 5 0 5 in a two sex model z l Figure A 1 6 Biological parameters that are entered as auxiliary information to the model A B E LI E F 3 207 vinit 22 80 0 wu 203 Age which way does it qa 204 205 206 207 206 r2 c oo oo ooo ooo eee oe Coo oe oo ooo oe eee oe Se zun D oo Go oe oo ee eee eo ooo So SS Se Se SS Se Se Se 8 8 SS ee SF 8 co D oo oe oo ee ee oe oo oo A 673 Ss SS 3738 ya ya 3738 ya e ya ya ya ya y 223 224 Ma of weight at age data sets 251 226 Weight at age on annual basis year ses aliadas 227 1353 1 1 1 1 1 Figure A 1 6 continued Biological parameters that are entered as auxiliary information to the model Coleraine Users Manual 33 B B 230 Data san foe Catch by method and year LI 233 602 03 02 645 524l B11 234 Total dara points 235 11 236 CPUE Indes Method Year value Cv 1 1 1 1 1 1 1 1 1 1 248 Mumber of Survey data points 291 11 250 Survey zurnmar Indes Year value Cv 210 Figure A 1 7 Data section for the Northern Cod case study B B 262 Number af commercial catch at age data sets 263 20 264 Catch at age data method year sample size 1 2 3 265 1 1962 R 0
26. ally the important parameters are the average level of recruitment R being the key parameter the q s for the Surveys and CPUE and the selectivities of survey and commercial gear You must get these within 50 or so of the actual MLE best fit to get good performance The VisualBasic software in Coleraine lets you view deterministic trajectories and their fits Use this feature before starting non linear fitting The key to getting parameters about right is to have an average exploitation rate in mind We recommend 20 As a starting estimate assume that the fishery has taken about 20 of the population An estimate of exploitation rate lets us start with the parameter R Begin by understanding the relationship between average recruitment R usually and vulnerable population size and spawning stock size This is the so called spawning biomass per recruit SBPR or for vulnerable stock it is vulnerable biomass per recruit If the weight of individual fish is in kilograms then the usual SBPR will be numbers between 1 and 10 It naturally depends upon natural mortality growth and age at maturity but 1 to 10 encompasses most of the SBPR s we have observed Thus if SBPR is 5 and R is 1 000 then we would have 5 000 units of SBPR in the unfished state Usually catches are measured in tons so this mean that 1 recruit with weights in kilograms is really a thousand recruits when compared to catches in tons Thus if catches are on the o
27. artment of Commerce National Marine Fishery Service Southwest Fishery Center La Jolla California 32 p Mace P 1994 Relationships between common biological reference points used as threshold and targets of fisheries management strategies Can J Fish Aquat Sci 51 110 122 Otter Research Ltd 1999 AD Model Builder documentation on line http otter rsch com admodel htm Seber G A F and C J Wild 1989 Nonlinear Regression John Wiley amp Sons 26 Hilborn et al Acknowledgements We would like to thank many people that have contributed to improve the model and or this manual Among others we have Juan Valero Carolina Minte Vera Ivonne Ortiz Vera Agostini John Field Andre Punt Arni Magnuson Trevor Branch Viviane Haist lan Stewart Shelton Harley and Marcus Duke The development of this software and manual was supported by the NZ Fishing Industry Board later NZ Seafood Industry Council and the NZ Foundation for Research Science and technology Coleraine Users Manual 27 APPENDIX A The Northern Cod case study Introduction The motivation for doing this exercise is not to present the best available explanation of the stock dynamics for this particular population but to familiarize the reader with the use of this software This section comprises seven parts Creating the Input Sheet Inputting the Data Running the Estimation Program Creating an Output Viewer Updating Re run Information on Outputs
28. at age for the surveys is affected by selectivity at size and the length age relationship according to 22 L 7 Jia SS a S 77 m 2 S g ps Sit Note If you do not have model based weight at age estimates you will need to input observed data for each year The vulnerable biomass computation for the last year 1 is based by default on model based estimates of weight at age so if you do not have those be sure to add one more line of observed weight at age data If your model is in numbers catch biomass etc you should enter Nyer 1 vectors of weight at age data filled out with values 1 1 2 6 Selectivity Selectivity is a process that can be modeled based on age or size This model supports an age based selectivity for the fishing fleet and a size or age based selectivity for the surveys In this model the only sex specific variation in the selectivity function arises from the difference between ages of full recruitment 1 2 6 1 Selectivity as a function of age The selectivity function implemented in the model is a double half Gaussian function of age Coleraine Users Manual 7 2 a x for 450 Sg Bi Sat E S ON UN fall exp f for a RV n Su t DA full S full where j is a dummy variable with value 7 for females and for males and is the sex specific difference in age of full recruitment for each gear The next graph shows some of the shapes that t
29. c Enter the rest of the data into the form d Check for blanks and other common problems 3 2 Default Values Entering the required priors is tedious and can be confusing so our program pastes in default values to the prior section Our intention is to help you remember what type of values should go in each cell however each value should be carefully checked before running the model On the other the results of the estimation might be sensitive to the initial parameter values therefore a menu option allows you to quickly do a deterministic run of the model to ensure that the estimation model starts at reasonable values 3 3 Two Warnings Do not attempt to enter dummy data yourself We have designed the program to do it for you where data are missing You will notice that the input form has several sheets They should not be deleted or renamed The sheet called NameList defines the sizes of each data range in the input sheet and the sheet called Defaults shows the defaults used Output is used in creating a text output file and is cleared each time the main program runs Graphs is used by the parameter checking routine Note that many cells contain formulas and many are named Generally speaking loss or change of names on ranges or sheets will cause the Excel macros to crash If you want to change default values try changing the numbers in Defaults but always check the name and formula boxes to makes sure you a
30. e entries do not play a role in the estimation phase and will be needed in the policy evaluation section The values must be entered at this stage though to get a correct sequence of entries in the input file more information on the specified values will be given later The Priors section is by far the most interactive area of the entire input form It contains entries for all the relevant information of potentially estimable parameters The decision to estimate some parameters or not will depend on the available information and our hypothesis about the model structure In this case we are estimating the virgin Coleraine Users Manual 29 recruitment row 102 recruitment deviations from the deterministic Beverton Holt Ba EP ESTE et ITE E ed 69 Names for Gear and Indez types use column 1 70 CPUE Indices 2 CommTrawl FS Survey Indices 24 Sure Trav 75 Commercial Gears l 76 CommTrawl 77 T8 Projection Parameters 80 81 Strategy Type 12 constant catch 2 harvest rate End year of projections 2 2001 85 Start Strategy lower bound of catch For projections Lee 2o oF End Strategy upper bound of catch For projections eg s t E Step Strategy interval ta use Far catch projections 81 Strategy U Proportion proportion of catch taken by metho Z 1 EE Assessment cu eror in population size estimate Lg Kum Assessment rho autocorrelation in asse
31. e search criteria File name NorthCod Files of type Fs ff 48 File s Found The next step is to chose the predictions in which you are interested check off the boxes in the dialog box The program will create two Excel files and you need to save them as text files with the names abel ind and data out Coleraine Users Manual 45 APPENDIX B General description of all the necessary input information for running the program DEBUGGER 0 On Turn the debugger on off It is part of the executable code and helps the 1 Off user to find errors at runtime RUN OPTIONS 1 default 2 user defined Upper bound for the exploitation rate value 0 1 E 0 Read in Gear that was operating at the beginning of the fishery If value 0 7 than it uses zi otherwise it is one of the specified selectivities In the projections only one gear is used If multiple gears were used in the estimation then a weighted average of the current year selectivity aad ogives is used 0 the user specifies the weights assigned to each gear 1 use same gear proportions as last year of data weights are the exploitation rates DIMENSIONING PARAMETERS __ bo rect than the nuber oda gem can be greater than the number of fishing gears Number of Survey indices number of different types of indices not the actual number of data 3 T Startyear ofthe estimation program 8 4 End year of the es
32. efault data can and should be overwritten Go to Main Menu gt Create or Modify Input File gt Check Format Before Running Model This calls a macro that will search for blanks and color them yellow It will also check the format of years which must have 4 digits they are colored red It can be run repeatedly with no ill effects g Then go to Main Menu gt Create or Modify Input File gt Check Parameter Values Before Running Model This calls a macro that runs population projections using the given initial parameter values i e the values in column 7 of the prior section It graphs the results in a sheet called graphs You can go back and change the default values and run this macro again until you have reasonable starting parameter values h At this point you are ready to run the estimation program If you are going to run the file immediately there is no need to save the file or to close it just go to Main Menu gt Run Model gt Run This File The invoked macro does three things 1 creates a slightly modified text only version of the file you just made 2 saves both the Excel and text versions you will be prompted for names and 3 calls the kernel AD Model Builder program passing it the text file The Excel format input file you created will stay open so that you can go back to it when you have seen the output But if you want to save it for later use the regular Excel Save menu and save it as an E
33. emory problems in your system If that happens close the output file and re open it following the updating sequence again tracker File Path rayiM A 2 EH RUH15 Input File case txt B Likelihoods 8 0 53603 8 Comm CA 1053 61 Comm CL SUEY 0 22056 Survey CA Survey CL Surv CL 45 Penatties 15 0387 1 Survey CL 2 Parameters luq varLesi Figure A 1 12 Summarization of one run in the Excel sheet Tracker 6 Forward Projections under Different Harvest Policies Coleraine allows the user to explore and evaluate the consequences of future management actions This is done in a Bayesian framework by using a Markov Chain Monte Carlo MCMC method Having specified priors and likelihoods in the model the estimation model uses numerical techniques for obtaining the posterior probability distribution for model and output parameters To specify the simulation conditions we need to go back to lines 82 96 Figure A 1 3 These entries allow the user to set the extension of the forward projection and the type of harvest strategy involved in the analysis In any case we need to set the starting ending and step values of the chosen harvest strategy This will allow us to evaluate the consequences associated with each strategy value Coleraine will estimate posterior probabilities for the time series of Coleraine Users Manual 41 projected spawning stock s
34. equation s N S g i t pool S prin ou MQ at S g ATS 5 1 where represents a matrix of age misclassification and M pools the age frequencies for ages a Apoa into a plus group 10 Hilborn et al Real age 1 2 3 4 5 6 7 8 9 10 Incorrect age O AN Oo a i WN The above figure shows how to set up the misclassification matrix If no information on age misclassification is available an identity matrix i e diagonal of 1 has to be used Similarly size compositions are predicted as 5 8 5 ATS Sit 270 a 155 1 2 252 Fa Nas S l a when selectivity is a function of fish size or as 2 7 S g 5 N gt 2 2 n a t s a when selectivity is a function of mean length at age The range of values of the predicted proportions at age are determined by the dimensioning parameters specified in the input form same for length and do not need to match with the ones specified for the dynamics 1 4 Objective Function Different sources of information contribute to the overall objective function This can be summarized as follows e Survey index Relative or absolute index of abundance for each index type CPUE Fishery related index of abundance by commercial fishing gear index Coleraine Users Manual 11 e Catch at length Survey by gear length time sex undetermined sex Commercial fleet by sex gear length time e Catch at age Survey by se
35. ers shouldn t stay in one range for a long time and then walk back The excursions from the average values should be brief You can plot pairs of parameters against each other this two dimensional graph should look like a shotgun blast and not have clumps of points in the space There are many formal statistical tests for convergence Gelman et al 1995 There is also a software package CODA that performs many of these statistical tests Check mean and variance of blocks of data Check for autocorrelation in samples ideally not autocorrelated Sensitivities to priors Average recruitment s qs Sample sizes for multinomial likelihoods Profile current stock size by fixing q and look at likelinood components Look at different sources of data one at a time
36. extinct from the known removals and also shows an exploitation rate is the 20 range then it is time to run the non linear convergence using the run this model option Start with the simplest assumptions It is important to do your first fitting with the simplest possible model Begin by assuming unfished equilibrium Estimate selectivities fixing right hand side Do not estimate recruitment residuals Do not estimate initial conditions Do not estimate M Do not estimate steepness The next step is to free up recruitment residuals Then free up initial conditions Then explore M and steepness Phases First however you have to specify the phases for each parameter This sequence of phases may be very important or it may not matter to much In general put the most important parameters in the early phases and the less important ones in the last phases For data series that show a decline in abundance as fishing increases the so Coleraine Users Manual 53 called one way trip the key parameters will usually be R and thus initial population size and the q s As a general rule we suggest Phase 1 q to fit the index data Phase 2 R to adjust the trajectory Phase 3 Selectivity parameters Phase 4 Recruitment residuals Phase 5 Changes in selectivity Phase 6 Deviation in initial conditions However in some analyses that have considerable catch at age data with clear strong cohorts you cannot fit the ca
37. gies harvest rates or catch levels on certain statistics of interest e g predicted vulnerable biomass which are reported as Bayesian posteriors Other salient features of this model are as follows Temporal changes in the selectivity of the fishing fleet Temporal changes in the catchability of the fishing fleet Survey selectivity led as age or size based The model simulte sly fitted to length and age data Robust normal likelihood function for proportions Automated process for saving condensed information on different runs 1 2 General Overview of the Estimation Model A general description of the different components of the estimation model Colera20 exe is presented in the following sections of this manual The following notation is used throughout Section 1 Subscripts a Age Length t Time Superscripts g Gear Fishery or Survey S Sex 1 2 1 Abundance dynamics by sex Abundance at age and sex is propagated according to the following difference equation 2 Hilborn et al ui aN em u for a 1 2 A where M is the instantaneous rate of natural mortality age A is a plus group and u is an age specific exploitation rate for all gears combined which is obtained by summing over all gear types s Sg 2 07 8 The exploitation rate for each gear is a product of its age specific selectivity s and the exploitation rate of fully selected fish at a specific ti
38. hart is activated If the menu suddenly disappears simply click on any non object such as a cell on the sheet b To start working with a new dataset choose Main Menu gt Create or Modify Input File gt Create New Input File This opens a template c Fill in the appropriate dimensioning data See Appendix 2 for clarification of the meanings of each value Every dimension is critical so make sure you understand the distinctions made d When done push the button labeled Create the Rest of the Form 18 Hilborn et al This will create the rest of the form using the dimensions you entered Beware that pushing this button again would erase any information on the page below the button and redraw the form Novv enter your data VVe recommend pasting in data from other spreadsheets although cells must contain numbers and not formulae Use Paste Special Values It is critical to avoid renaming cells by mistake if you do not use paste special all the time this might happen Note before pasting All outlined areas must be filled with numbers See Appendix 2 for clarification of the data types and formats required Some arrays are filled with dummy data Do not change them Some arrays may already have zeros in some positions e g if your catch at age data begin at age 6 ages 1 through 5 will appear as columns of zeros in the catch at age data section This is the required format Priors and other d
39. he selectivity of gear at time f The number of parameters per gear to be estimated will depend upon the number of years catch at age data are available They are normally distributed with mean O and variance 2 OF a LY be estimated will depend upon the number of years where we do have catch at age data They are normally distributed with mean 0 and y g Residuals of the variance of the right half Gaussian distribution of the selectivity of gear at time f The number of parameters per gear to variance o 1 1 5 6 g Catchability coefficient of CPUE series i The bounds and initial guesses have to be in natural logarithmic scale 1 17 PUE Temporal residuals of the catchability coefficient of the CPUE series They are normally distributed with mean 0 and variance o 9 0 i S 5 A 5 1 Age difference between sexes of the age of full selectivity of the survey ur gear type j e t Catchability coefficient of survey index The bounds and initial d guesses have to be in natural logarithmic scale S Variance of the left half Gaussian distribution of the selectivity of survey L gear type j at time t The bounds and initial guesses have to be in natural logarithmic scale S Variance of the right half Gaussian distribution of the selectivity of survey gear type at time f The bounds and initial guess have to be in natural logarithmic scale Not used in the model
40. he same method as SurvSel for deleting series C CPUE catch per unit effort Graphs the observed CPUE against CPUE index trajectories Master a direct copy of the results file Results dat but the viewer program has located and named each data range so the graph sheets can find the data to graph GraphMaster a template as explained in the 5 4B 5 5 Storing important information of different runs in one file Tracker xls A useful feature of Coleraine is the way vital information of different runs gets summarized Whenever the estimation model has been run and the information has been transferred to the viewer then the file Tracker xls gets updated with relevant information such as likelihoods parameter values phases file path information This file is automatically open after you open maincode xls The next figure shows the different parts of Tracker using the Northern Cod example If a likelihood component is not used during the estimation it will be colored gray Parameters that are estimated will be in color red phase 1 yellow phase 2 green phase 3 etc 24 Hilborn et al 3 Microsoft Excel tracker File Edit view Insert Format Tools Data Window Help MAIN MENU ig xi CABernst CABerstriialerainetasteerzionor Code SUNT C HUN Path of txt file Input File caso l tst 2 T Different runs Likelihoods Input txt file associated to 8 CPUE 0 512812 8 Com
41. his three parameter model can adopt The thick line represents an asymptotic right hand side curve very high right hand variance as opposed to the thinner line which has a declining right hand limb smaller right hand variance Selectivity S Age Survey selectivities are assumed to be constant over time vvhile commercial selectivities are allovved to change over time Residuals are estimated for the periods when we do have catch at age data Sa tes ee h N 0 pul ay WHERE oS 0 Sea yay amp N 0 ot J tl 7 1 1 jV t 2 where is the right or left side variance 8 Hilborn et al Trends in selectivity have been associated with changes in spatial allocation of fishing effort Jacobson et al 1997 and the variation considered in this approach is independent of sex The following figure shows a declining pattern in the right side of the selectivity curve over time It also shows a decrease in age of full selectivity between the first and the last time period Selectivity Age 1 2 6 2 Selectivity as a function of size Only the selectivity of the survey is allowed to be size based A double Gaussian function of size with time invariant parameters is used The selectivity at age is computed by integrating the selectivity at size over the size proportions at age Thus Sat L ep log L u c dlogL The integral above can be approximated by discretiz
42. imensioned input file for the estimation program or graphing output data Therefore the user interacts directly only with regular Excel files and Excel operates normally most of the time The only exception is that it pauses while macros are running and a new menu is added to the command bar that allows the user to call macros more easily This menu is active as long as the Excel file Maincode xls containing the macros is open 2 4 Necessary Files Five files are critical to running this program 1 Colera20 exe A compiled version of the ADMB estimation program Your Path statement in Autoexec bat must include the directory where this program resides It may be easiest for you just to put it in the working directory 2 Maincode xls Excel file that contains all the macros it must be open for the macros to function It can be anywhere Excel can find it and it should not be modified 3 InputTemplate xls Needed to create new input sheets It is invoked from maincode xis Coleraine Users Manual 15 4 Tracker xls Excel file that stores vital information e g path information file names parameter values likelihoods for each run 5 Newgraph xls Used to create new output files It is automatically called from maincode xls 2 5 File Management The use of an Excel interface for data entry and graphing adds an extra step to file management since Colera20 exe requires text format input and produces text format output I
43. in the constrained optimization c Prior type column D Three types of priors are available in this model uniform 0 normal 1 and lognormal 2 d Mean column E Mean of the normal or lognormal distributions If the prior is uniform a dummy value must be entered Coleraine Users Manual 47 e CV column F Coefficient of variation of the probability distribution If the prior is uniform a dummy value must be entered f Default value column G Initial guess for the parameter This data will be used in the maximum likelihood estimation of the parameters In many cases the final estimates depend on the initial guesses so several combinations of initial values should be explored Prior Mean Standard Lower Upper type deviation bound bound o 38 Initial Phase 83 Priors 100 guess FANEN Recruitment in virgi condition 162 21 0 10000 Y Y B00 103 h steepness of spawner recruit cure 104 1 0 01 5 OF 0 6 0 105 M1 natural mertalit 106 1 0 01 03 01 0 1 02 Figure B 2 1 Location of each of the 7 entries for each parameter LA Virgin recruitment from the Beverton Holt recruitment model oteepness parameter from the Beverton Holt recruitment model This parameter is defined in the range 0 2 1 0 BEI DIEI Natural Natural mortality by sex 7XK by sex B e7 Initial age structure residuals They go from age 2 to nages 1 They are normally distributed with mean 0 and variance
44. ing the size distribution into n size classes denoted as as IE sg Sg fs Sat Sa 1 1 where s is the size selectivity function evaluated at L the length at the mid point of interval For converting the size based selectivity into an age based selectivity we weight the selectivity at size by the size proportion at the respective age If we do not rescale the new selectivities at age very likely no age has been fully selected This would not affect the estimation procedure but would be reflected in the catchability coefficient Coleraine Users Manual 9 1 3 Data 1 3 1 Predicted abundance indices Commercial CPUE and survey indices here denoted as are assumed to be directly proportional to the vulnerable biomass in the middle of the year g_ g 0 5M g ATS g T q x 2 NeW 5 where q is the gear specific catchability The temporal index for the catchability coefficients is incorporated only for the commercial CPUE catchability coefficients of the surveys are not allovved to have a temporal variation A random vvalk model is used to model the temporal changes thus CPUE logla logla 6 2 c The parameter n is used to control the amount of year where yen rn N 0 qE to year variation allowed ing 1 3 2 Predicted age and size composition The predicted age composition in proportions of the catch at time t by sex and gear is represented by the following
45. io pst Posteriors of the time series of estimated vulnerable biomass for the first fishing gear mbio pst Posteriors of the time series of estimated spawning biomass recs pst Posteriors of the time series of estimated recruiments explrate pst Posteriors of the time series of estimated exploitation rates for the first fishing gear params pst Posterior probability of the virgin spawning biomass virgin recruitment recruitment steepness natural mortality by sex catchability coefficient per fleet and catchability coefficient of each survey index Coleraine Users Manual 21 5 Processing Output 5 1 Overview The estimation program produces about 20 files each time the program is run Only a few of the files generated are relevant to the user and none need be opened since our viewer macros copy the main results directly into Excel The benefit of using the Viewer is that it takes the estimation output file which is usually large 800 lines or more sparsely labeled and dense and makes it easy to find and display information from it The process of viewing output is entirely independent as far as the macros go of the process of creating input files and running the model Essentially all the viewer does is to copy the Results dat file into an Excel worksheet and then add 11 more worksheets to the Excel file including a graph making template to make it easy to graph the data and manipulate the graphs in blocks Memory pr
46. is relation can be visualized in the following graph Age Log Length The proportions of length at age are used in many sections of the model depending on the nature of available data They are used to compute the predicted size compositions to convert a length based selectivity into a selectivity at age and to compute the mean weight at age when the selectivity function of the survey is a function of length 6 Hilborn et al 1 2 5 VVeight at age relationship VVeight at age is a vital piece of information in the assessment because it is involved in the vulnerable biomass calculations It can be directly incorporated into the model as observed data design based estimators or by using a model based approach parameters of the von Bertalanffy growth curve and the weight length power function By default the program uses the observed data The rest of the temporal weight at age information arises from the following calculations a If selectivity is a function of age mean weight at age is predicted from the following equation bi o i21 2 Sas sb Wii b L e where the exponential is a correction for the variance of the log normal distribution of size at age If the survey selectivity is based on age than the weight at age for the commercial fleet is the same as the one for the surveys However selectivity can be modeled as a function of fish size only for surveys in which case the mean weight
47. ize and vulnerable biomass These results are output to a text file called project out To run the projections we use the Run MCMC command MAIN MEMO Create or Modify Input File Run Model View Output Run mc eval MEME This action will prompt you to enter the number of MCMC simulations This number will depend on the number of iterations needed to satisfactorily describe the posterior probability of the output parameters In some cases this could be a long process taking millions of iterations Microsoft Excel Input the number of MCMC simulations Cancel fi 0000 program will prompt you to enter the text file name of the specific run you are interested in If you introduced changes to the Excel input sheet you will need to run the estimation program before running MCMC in order to pass these modifications to the text file Select the file to run Ei E pics EJ coleraz0 bar 20 Open j Add resu 3 caleraz caleraz std 38 NarthCod bn variar c admodel cov culeraz eva eigv rpt 38 North Codout vhin ps admodel dep F T oleraz isa expirakbe pst e params pst Advanced El admodel hes coleraz0 log 38 InputTemplate is project 3 casel res caleraz pO01 B likeliha as project Casel 20 02 38 maincode e recs pst e cmpdiff tmp 3 caleraz mbio pst is results a colera20 b 1 caleraz r 1 38 Inewgraph lig Screenshot
48. l parameter values may be something else and we want to find the probabilities associated with other possible parameter values i e explore the uncertainty in the real parameters Thus we use the MLE as a jumping off point for MCMC which does a random walk over the posterior space Typically we let the MCMC do about 1 000 000 samples from the posterior space and keep every 1000 th of these samples Assuring MCMC convergence We recommend that you start by doing 1 000 000 MCMC draws and saving 1 000 of these We know we will have started in the middle of the posterior density because Ad Model Builder begins MCMC at the MLE parameter values Your first step should be to plot the total negative log likelihood as it evolves in the MCMC There are two things you are seeking First is the absolute value of the value It should wonder around the MLE value the lowest you will see and about 2 times the number of parameters above the MLE Thus if the MLE is at 30 and you have 10 parameters the value of the likelihood really posterior density should range between 20 and 10 2 10 units higher Second you should look at the evolution of the likelihood It should look like a random walk not like a directed walk The average value in the final 500 saved points should be about the same as the first 500 saved points You should also check the evolution of other parameters Again they should look like random walks not one way walks The paramet
49. leets weighted according to the gear specific exploitation rates in the final year covered by the assessment gt 5816 Sat u s proj 2 1 S a The user can also specify other weights to compute the average selectivity Section 6 1 There is a constraint on the total exploitation rate to be less than 0 99 during the entire projection Two sources of uncertainty are incorporated in the projections 1 uncertainty in current population size and 2 process uncertainty The uncertainty in the current population size is determined during the MCMC simulation and is a function of all the uncertainty associated to the main parameters of the model Process uncertainty on the other hand is simulated during the execution of mceval by allowing for i Variability in recruitment which is modeled as S 5 0 2 A t ole t R 2 l os li Implementation error due to errors in future assessments only for harvest rate strategy We assume that future estimates of exploitable biomass B would be log normally distributed around the true value so that D 0 5M S proj S ATS n p x k 5 and represents the log of the assessment error in year t The ass are assumed to ass be normally distributed with variance and serial autocorrelation o such that asset P ass a ed N 0 ass c 1 m o 14 Hilborn et al 2 Getting Started 2 1 Hardvvare Requirements This program sho
50. m CA 15185 1151394 that particular run 10 Comm CL Em 11 Survey 887854 0 530792 12 seti R Different Likelihoods and Urey Sum CL MoSex 0 penalties in the model 15 Penalties 14 0733 14 0623 16 Survey CL 2 F arameters PU risen Rinit uinitlInizena plusscaleUnisex Sfullest log_varLest loq varFiest errSFull errvarL Loi 2 III qCF UEerr i 33 log qsurveq 2 zurveqshull luq surveyvarl 1 1 lag sureeuyearFi 10 10 log InitialOew log RecOey survey Stull 1 1 0 0 D T D Parameters of the model in color if they are estimated p Coleraine Users Manual 25 Bibliography Francis R I C C 1992 Use of fish analysis to assess fishery management strategies a case study using orange roughy Hoplostethus atlanticus on the Chatham Rise New Zealand Can J Fish Aquat Sci 49 922 930 Fournier D A J Hampton and J R Sibert 1998 MULTIFAN CL a length based age structured model for fisheries stock assessment with applications to South Pacific albacore Thunnus alalunga Can J Fish Aquat Sci 55 1 12 Gelman A J B Carlin H S Stern and D B Rubin 1995 Bayesian data analysis Chapman and Hall New York Jacobson L D and five co authors 1997 Empirical fishery selectivity estimates for Dover sole sablefish and thornyheads in the deep water Dover fishery U S Dep
51. me s s u esq a t 8 t Formulations below are identical whether g refers to a fishery component or to a survey except that the mortality induced by the surveys is negligible and can be ignored The alternative approaches used for the selectivity function are explained in a later section Assuming that total commercial catches in biomass for each gear C are known without error and that fishing takes place in a short time interval in the middle of the year the annual exploitation rate by gear is given by si 8 t 0 5 M5 8 9 ATS 5 dye Sat Nut M aod 5 u which is equal to the ratio of total catch to vulnerable biomass in the middle of the year 1 2 2 Initial conditions The initial condition assumptions built into the model allow for the estimation of three parameters R virgin recruitment fraction of R in the first year and wu exploitation rate for the first year The initial vulnerability at age pattern by sex has to be incorporated by the user in the Fixed Parameter Section item 13 Also the fraction of N and more generally N j year that recruits to each sex is represented by a user defined constant 2 Thus the initial population age structure is represented by Coleraine Users Manual 3 Ni oCR where C A C 1 A i a l N N e PD Q v w for a z2 4 1 i l The plus group for the initial year is given by a A l v u L al aM A 1 m Na
52. n brief the process of working with a dataset runs as follows Excel Maincode xls InputTemplate xls Tracker xls Newgraph xls DATA INPUT GRAPHICAL OUPUT Create input file Create output file Viewer Creates text file Colera20 exe Text input file ADMB application Therefore each run of the model has four data files associated with it two input files one in Excel format and one in text format and two output files again Excel and text formats Note Excel format input and output files may remain open while the model runs and the Excel file named Maincode xls which contains the macros controlling the user interface must remain open for the user interface to work 16 Hilborn et al 3 Getting Your Data into Coleraine 3 1 General Issues Colera20 exe reads in all data from a file before running and is very discriminating about data structure A single missing value will shift all the following arrays to the left and if you are lucky cause the program to crash Therefore most arrays must be filled with dummy data even if they are not used in the calculations We have designed the following procedure to minimize problems a Enter into a template the small set of numbers that define the dimensions of the rest of the data arrays e g number of years of data maximum age and length of fish number of sexes b Use the filled in template to create an input form in the proper dimension
53. n the Graphmaster sheet and then pressing the yellow button called Remake all Graph Masters This will cause the other sheets to be re written and you will lose any changes you have made to individual graphs C The worksheets one by one The different worksheets pasted into the viewer are as follows General graphs of vulnerable biomass vs catch exploitation rate spawning biomass vs recruits and spawning biomass vs recruits Note that the graphs with two series each have two Y axes The following sheets all graph predicted against observed data SurvNoSexCL catch at length survey data in which gender is undetermined by survey index and year Coleraine Users Manual 23 SurvC L survey catch at length data by survey index year and gender ComC L commercial catch at length data by method year and gender SurvC A survey catch at age data by survey index year and gender ComC A commercial catch at age by method year and gender Surveys surveys by survey index and gender It compares predictions with observed data SurvSel survey selectivity There is a graph for each gender and method and a different series for each year The survey years are shown in row 12 extending to the right of column 9 To make the graph more readable one can delete series by replacing years in row 12 with any text which will cause that year to be omitted from the graph CommsSel commercial selectivity by method year and gender Uses t
54. nd the yearly point estimates of the CPUE The Survey entries follow the same pattern starting with index number in column A year in column B point estimate in C and coefficient of variation in D Catch at age data in proportions are entered in lines 265 294 and have the following structure Column A Fishing method Column B Sampling year Column C Sample size number of observations in a specific year Column D Proportions at age in the catch at age matrix For a two sex model the female proportions at age entries are entered first and the male proportions immediately after them The entire row of proportion has to add up to 1 one For this dataset the rest of the entries are dummy data but they need to be entered to keep the program from crashing 32 Hilborn et al B B E D E F 3 H 174 175 Fized Parameters 1 125 127 Bi zcalar af length weight relationship 178 16 05 173 bii esponent af length weight relationship 186 32 181 L infinity af the vanBertanlanFfy growth equation 182 103 183 k af the vanBertanlanffy growth equation 184 037 186 t of the vonBertanlantfy growth equation 186 137 157 Brody parameter Ee 02 183 Mean length of age 1 Fish Eo 32 131 Length at oldest age w zl 133 5 4 af length at age of 1 year old Fish 1341 64 135 54 af length at age of oldest fish 136 isl 137 Maturity agive 138 133 Fr
55. nsions of a data array would distort the viewer badly 5 4 Using the Viewer Part Il Making Graphs Once a Viewer file is open you are in a regular Excel file with the only exception being that the buttons appearing on the sheets are linked to macros in the MainCode xls workbook which must also be open for them to work The graphs can be customized individually as you would do with any Excel graph but the viewer macros also make it easy to manipulate them in blocks A Making changes to one sheet Anything highlighted in blue can be changed by the user The changes will go into effect when you press the Make Graphs button again In rows 1 8 there are a few parameters that control all the graphs on the sheet height width etc In the block of data starting in row 12 the options affect each graph individually Typing Y in the blue cells below columns marked Female or Male will activate that graph anything else will cause the graph to be skipped It is possible to delete all the graphs on a sheet using the Delete Graphs button but it is not normally necessary since existing graphs on a sheet are removed each time you press the Make Graphs button B Making changes to all of the sheets at once The sheet called Graphmaster is essentially a template for all other sheets Options such as graph height or width pool length or age or names of survey indices or commercial fishing methods can be changed for all sheets by changing them o
56. number of observed commercial catch at length data separated by sex Commercial catch at length data separated by sex First column method second column year third column sample size others Starts at the first defined length interval of females and goes to pooled length In the next entry to the right it starts with the first defined length interval for males and goes to pool age These values are proportions that for both sexes combined add up to 1 one Total number of observed survey catch at length data separated by Sex Survey catch at length data separated by sex First column index second column year third column sample size others Starts at the first defined length interval of females and goes to pooled length In the next entry to the right it starts with the first defined length interval for males and goes to pool age These values are proportions that for both sexes combined add up to 1 one Total number of observed survey catch at length data not separated by Sex os Survey catch at length data not separated by sex First column 15 index second column year third column sample size others catch at length data Coleraine Users Manual 51 APPENDIX C Steps in model fitting for age structured models using Coleraine Obtaining MLE estimates The first step in model fitting is to get the best possible fit to the data Since Coleraine uses both the likelihood of the fit to the
57. o get reasonable starting values for all the model parameters conditioned on the available data This exercise can be easily done by using the following command MAIN MENU ate or Modify Input File Create Mew Input File Run Model Open Existing Excel Format Input File View Output Reconstitute Excel input File From text File Check Format Before Running Model Check Parameter Values Before Running Model b Coleraine Users Manual 35 This procedure generates a group of graphs containing model predictions of vulnerable biomass spawners and recruits selectivity pattern CPUE and Survey Index Figure A 1 8 F G H 1 2 Vulnerable Biomass gear 1 Exploitation rate gear 1 3 7000 03 R 6000 5000 9 4000 3000 10 01 11 2000 12 1000 13 n 0 e 1362 1367 1372 1977 1982 1997 1362 1367 1372 1377 1982 1987 Vulnerable Biomass Catch Exploitation rate 4 4 b MS InputForm Z Output NameList Defaults x Graphs 4 Figure A 1 8 Deterministic model outputs before running the estimation model 1 2 Spawning Stock and Recruits Selecthaty gear 1 9 Combined 2 6000 5 m i 4000 8 Mh 8 3000 in 2000 l 11 12 1000 13 D E 1362 1372 1382 E M it Spawrners Recruits Catch at age Selectivity at age 4 4 b bl InputForm Output 4 Mamelist Defaults Graphs 4 Figure A 1 8 continued Deterministic model outputs before running the estimation
58. oblems Graphs take up an enormous amount of memory and for some undetermined reason Excel does not reallocate RAM after a graph is deleted Therefore after you have created and deleted a number of graphs e g 50 on a laptop Excel may run out of memory even if no graphs are currently visible This typically happens when you are trying to make more graphs The solution is to end the macro just choose End and close the file The memory is released and you can then reopen the file this problem seems to be fixed for Excel 2000 Closing Excel itself seems unnecessary 5 2 Text Output Files Produced by Colera20 exe The relevant output files are in text format and among them only one is used by the viewer Results dat Results dat contains an exhaustive listing of the maximum likelihood estimates for the model and derived parameters a re listing of some of the data and fixed parameters and a listing of most of the prediction made This includes numbers at age fecundity vulnerable biomass survey trajectories and so on It is always placed in the same directory as the text input file If there is an existing Results dat file in that directory it will be overwritten Other output files generated during the parameter estimation process include Colerain par contains the maximum likelihood estimates of the free parameters Colerain cor shows standard deviation and correlations between the estimated parameters and Colerain std s
59. ormation is used for building a default maturity and initial selectivity ogive PROJECTION PARAMETERS After fitting the model to the data these parameters are used for evaluating policy options 1 Strategy type used in the forward projections Constant catch 1 and Harvest rate 2 Year to which the projections should be caried out Tower limit for constant catch or harvest rate strategy 00 T pperlimitfor constant catch or harvest rate strategy mdementinthestatey 00 S BB Proportions used to vveight each selectivity curve in the current year to average them out needs to add up to 1 one BB Coefficient of variation in the population size estimates of the future assessments only for constant harvest rate strategy Degree of autocorrelation in the assessment errors only for constant harvest rate strategy PRIORS FOR THE MODEL Seven entries are required for each row in the Priors section a Phase column A The phase determines the order in which the specified parameter will be estimated during the nonlinear search A value of 1 means that this parameter will be estimated in the first phase 2 in the second and so on Any negative number implies that the parameter is fixed at the value specified in the starting conditions and will not be estimated by the model b Bounds These correspond with the lower Column B and upper Column C bounds imposed by the user on the parameter values
60. ot get convergence when you free up additional parameters or change the value of a fixed parameter Here you can often detect problems by looking at the tracker output If you free up an additional parameter for instance freeing up natural mortality or the right hand side selectivity and the likelihood does not get better smaller then something is wrong Anytime you make more parameters free the model must fit as well or better When looking at Tracker you have to look at the total likelihood some data sources may fit worse having been sacrificed to fit other sources better Note that if you change the sample sizes in length or age frequency data or take elements out of the likelinoods used then the total likelihood will be changed and you can no longer directly compare likelihoods Another way to check on convergence and model behavior is to do a likelihood profile by manually running the model with different values of a key parameter such as R or M For instance you might run the model with M fixed at 1 15 2 25 3 and 35 Then also run the model with M estimated The likelihood with M estimated should be the lowest 54 Hilborn et al one and a plot of likelihood vs M should be smooth with a clear minimum at the value estimated when the parameter is free Bayesian integration Finding the MLE is just the beginning it is one fit to the data that happens to be best by the definition of maximum likelihood but the rea
61. raZ b l ah colerazd pO InputTemplate m B Add resu bai coleraz b z E caleraz El likeliha i admodel cov caleraz bar B caleraz r 1 55 maincade as admodel dep 5 caleraz 20 02 bal mhin pst admodel hes bal caleraz eva caleraz 39 Inewgraph ias LA casel res coleraz e caleraz std 38 North cod Casei bal caleraz lag bn eigv rpt FF NorthCod ing cmpuifF Emp m 20 01 explrate pst 59 NorthCod 5 k NerthCedout Save as type Files 38 Hilborn et al This procedure creates an Excel file that contains several sheets Some of them have controls to specify the setup of the graphs The user can also specify which graphs should be created The worksheet General has several graphs which are shown in Figure A 1 9 It includes vulnerable biomass recruitment spawners and harvest rate trends The outputs for the age structured data are reported in the sheet ComC A Figure A 1 10 This sheet has many entries to control the graph Other outputs of interest for this dataset are shown in Figure A 1 11 Fits of predicted CPUE and relative index of abundance of the surveys to the observed values setup and selectivity ogives are displayed in different sheets A E E Li E 14 15 Yulnerable Biomass us Catch 16 CommtTrawl 17 18 13 20 z1 2 23 A 1362 1367 1372 1377 1382 1987 1332 es Ulner able Bioma
62. rder 52 Hilborn et al of 1 000 tons then a R of 1 000 would imply an exploitation rate on an unfished stock or 20 As a starting estimate choose an R so that the exploitation rate would average about 20 You should always have these numbers in mind SBPR or vulnerable biomass per recruit average levels of catch and an estimate of the likely exploitation rate If you have measured weight of individuals in grams then each unit of recruitment will be 1 000 000 fish assuming catches are still in tons We can use our target exploitation rate of 2096 to help us define the starting estimates of qs for CPUE and surveys as well Assume that the average CPUE is 5 and the average catch is 1 000 tons If exploitation rate is 20 then the average biomass is 5 000 and thus q is 5 5000 or 1 10000 Getting starting estimates for the selectivity values is a little tougher Initially it is usually simplest to assume that males and females have the same selectivity and that there is not a descending right hand limb The age of full selectivity can be set either to the age at which catch at age is maximum or the age that corresponds to the mode of the length frequency data We have found that setting the left side standard deviation of the selectivity curve to 2 log value 69 works well as a starting condition i e it is not knife edged but it isn t too broad Once you have starting values that give a trajectory that does not go
63. re not changing names or formulas Both Defaults and NameList are protected but may be unprotected using the Excel menu Tools gt Protection gt Unprotect sheet 3 4 The Main Menu The following figure describes all the different components of the Main Menu that are added to the principal Excel toolbar Each component will be explained in later sections of this manual Coleraine Users Manual 17 Create Mew Input File Open Existing Excel Farmat Input File Reconstituke Excel input File From text File Check Format Before Running Model MAIN MEMU 7771 Check Parameter values Before Running Model b Run Model 6 Update graphs View Output Redraw graphs MEMEC Run Other File h Text File Worksheet xls File View New Output File T7 View Re run of Open File View Other Output File Run MEME Run me eval MEMC ko CODA The three main procedures in the Main Menu are related to the creation or modification of files running of the estimation program and viewing the model outputs 3 5 Step By Step Procedure for Entering Data Adhering to the following procedure will minimize problems a Open MainCode xls You will notice a new menu item is added to the toolbar at the top of your file called MAIN MENU It will disappear when you close Excel Note the menu becomes inactive when an embedded object such as a c
64. rom the policy evaluation mceval phase Figure A 1 13 show the set of evaluation policies rows 2 7 8 13 etc for each MCMC simulation that was stored Coleraine automatically specifies the spaces between saved values in order to get output files with 1000 simulations In other words if we have 6 catch levels we should have 6000 rows of data in our projection out text file This data can be sorted and analyzed in many different environments One of these is CODA Convergence Diagnosis and Output Analysis Software an S Plus function written to analyze Gibbs sampling outputs It has many built in statistical methods for monitoring convergence of Markov chains For downloading the software and documentation the user can visit the following website http www mrc bsu cam ac uk bugs classic coda04 readme shtml CODA has a very specific input data structure which is not compatible with projection out file Coleraine enables us to create two necessary input files label ind and data out by using the following command MAIN MENU Create or Modify Input File Run Model View Output Run MEMEC Run me ewval MEME bo You need to specify the Excel input file that you are using in this run 44 Hilborn et al Zz S4 CLE TS 2077777 se emp 5 ar Turn bal COLERAZO mcm explrake pst e params 20 01 InputTemplate project i 4 Find Files that match thes
65. ss Catch 26 cu m n ka ad c l Pat qa cn m co h cum Recruitment M3 O m CH Cn O O 222 D D O 5 1000 2000 3000 4000 5000 T Stock 40 41 Deterministic F Linear F Observed F 4 4k H General 1 5urvicimL Ezploitation Hate CommtTrawl LE 0 4 0 2 i 1362 1367 1372 13 1382 1387 1332 Spawning Biomass vs No of Recruits CommtTrawl 5000 600 4000 3000 400 2000 200 1000 1362 1367 1372 1377 1352 1987 1332 Spawning Biomass Fiecruits Comi iml 4 r Figure A 1 9 Graphs of some major model outputs B B L CommTrawl 1962 linizez Proportion 4 4 y M Z Comc l Survc A comc a 4 Figure A 1 10 Graphs of commercial observed and predicted catch at age data Coleraine Users Manual 39 A E B 12 Sure Trawl HA F A B C CummTr anl 22 600 23 1362 1367 1372 1377 1322 1367 1332 26 Years 1362 1367 1922 1977 1382 1987 1332 Tears P2 Dr Pm rm P Ka Pe Pe Pe z m cm e a Ka o m Len 4 419 Pll Comem a h Surveys Cor sla 4 Comel CPUE Graph Surulrawl linizez Commtrawl Unizez Proportion IX dai Proportion 0 0 Age 1362 1963 1154 1355 1366 1367 1358 1353 1970 1371 A ComSel
66. ssment error sel l Figure A 1 3 Names for gear types and profection parameters recruitment model row 110 and catchability coefficients for the CPUE and relative index of abundance of the Survey rows 132 and 136 The remaining parameters are fixed and therefore assumed known with values dependent on their initial guesses Given there was no structured information from the survey data the survey selectivity parameters were not estimated For this run the selectivity parameters of the fishing fleet were not estimated However this can be done with the long time series of catch at age data The model currently assumes that the population started at an unfished equilibrium state This assumption can be readily relaxed by freeing up the initial exploitation rate and the fraction of the virgin recruitment in 1962 rows 112 and 114 Departures from the deterministic exponential decay in the initial age structure are also allowed by incorporating additional residuals in the estimation row 108 Given that this is a one sex model no available sex structured data selectivity parameters for rows 120 and 140 are dummy data It is important to remember that all the residuals catchability coefficient and variance parameters for the selectivity ogives are entered in log scale The Prior and the Likelihood sections can be modified between different runs to explore the consequences of different assumptions Figure A 1 5 shows the cells for the
67. st log normal likelihood function has the following representation 2 E log Li Yos exl 0 5 z t I8 t In all the cases the variances are entries specified by the user not estimated within the model and residuals the difference between the observed and expected values logarithm of the observed and predicted values for the previous 1 4 3 Total likelihood 12 Hilborn et al The total log likelihood is the result of the sum of the individual log likelihood components g g Surveyr g Surveyrg Commrg Commrg log l log L iS gt log x log Lz 2 log 2 b log log Engin Survey Surve Comm Comm ny Mage Mage NCPUE Nength Mength 1 4 4 Penalties Several penalties might be affecting the overall objective function depending on different model assumptions In general the penalties correspond with prior assumptions made about some of the stochastic processes involved specifically recruitment variability and variability in the initial age structure 2 E PSS 0 5 4 t RO 7 PSS 0 5 x 1 yO time series trends in catchability by gear 2 E PSS 0 5 L q and time series trends in the parameters of the age selectivity functions for the different commercial fisheries p g 2 E E PSS su 0 5 SP PSS 0 5 E and PSS 0 55 z t t 1 sil LV RV Hence the overall penality vvould be the sum of the individual components penalties PSS PSS 5 PSS 4 PS
68. tandard deviation of the estimated parameters 5 3 Using the Viewer Part 1 Create or Open a Results File A Viewing the results of a file you just ran Choosing Main Menu gt View Output gt View New Output File will run a macro that opens a Results dat file and creates an Excel workbook with 11 sheets in it If the model just ran the program should still have the name of the input file in its memory In this case it will take Results dat from the same directory since any older Results dat would have been overwritten Otherwise it will ask you to define the Input File used 22 Hilborn et al B Viewing a different Excel output file The option Main Menu gt View Output gt View Other Output File will open an existing 12 sheet Excel workbook that was previously created using the View New Output File option C Viewing the results of a re run open file We wanted to make it possible to quickly modify parameter values run the model and look at the output without going through the process of creating a new Excel output workbook and without losing modifications that might have been made to the graphs The menu option called Main Menu gt View Output gt View ReRun of Open File is designed to be used when a model has been re run and a 12 sheet Excel viewer file is currently open It simply pastes a new Results dat file into the sheet called Master It is important to realize that any change to a parameter that would change the dime
69. tch at age data without getting the recruitment residuals right and you need to move the recruitment residuals up to the first or second phase One of the limitations in the current November 2001 implementation of Coleraine is you cannot specify starting estimates for year class strengths so if you have dominating catch at age data you need to get the year class strengths more or less right by putting them in phase 1 or phase 2 before trying to estimate the selectivities and possibly even the q s Assuring convergence How do you know if the model is actually finding the best possible fit Step 1 is first to look at the fit plotted against the data In general if you are fitting the data well the model has almost certainly converged Step 2 is to restart the estimation from different initial points usually different R q and selectivity values If the fits converge on the same point it is a pretty good sign that the model has converged If you both get a good fit to the data and multiple starting points converge to the same place you can be fairly sure you have reached a true minimum of negative log likelihood A third way is to use the AD Model Builder feature of estimating the Hessian matrix derivative of the negative log likelinood with respect to each parameter If this matrix is not positive definite then the model has not converged Note checks during fitting When doing sensitivity runs you always need to be aware that you might n
70. timation program 8 Number of ages used in the numbers atage matrix 6 Age at which the catch at age data are pooled SS 7 rirstage with effective catch at age values in the datasets 8 First length with effective caich at length values in the datasets Length increment in the catch at length dataset same for Commercial and Survey data 0 Number of length intervals in the catch atiengih dataset zan interval at which catohvat length data are pooled a Number of different commercial fishing methods 13 J Number of sexes considered in this analysis 1 pooled 44 Total number of available CPUE data points Tora commercial indices EE NN lt years gears I v lt years gears 17 Total number of weight at age datasets If you are not including the Gear 2 46 Hilborn et al growth and allometric parameters of the length weight relationship you should enter Start year End year rows of data 18 Total number of survey index data points all types combined 19 Total number of observed survey catch at age datasets years survey gears 15 Total number of observed survey catch at length datasets lt years survey gears Ls 15 not determined BP Survey vulnerability type 1 age based 2 length based 22 Age of full recruitment to the fishery This inf
71. uld run on almost any Pentium machine vvith at least 16 Mb of RAM More memory will improve the performance of the program The program requires about 1 6 Mb of hard disk space VVe recommend the use of Excel 2000 because of some vvell documented problems vvith the memory handling that existed vvith previous versions vvhen a large number of graphs vvere created 2 2 Softvvare Requirements VVindovvs 95 98 2000 NT and Excel 97 2000 are the only softvvare required Our program s user interface consists of macros written in VBA for Excel 97 2000 a scaled down version of Visual Basic included with Microsoft Excel 97 2000 In some cases the user may have to load this part of the program into Excel by using the Add Ins menu 2 3 Outline of Working Environment This package consists of two parts 1 an estimation program Colera20 exe written with an application called AD Model Builder Otter Research Ltd 1999 that does all the model parameter estimation and policy evaluation and 2 a user friendly Excel 97 2000 interface for entering data and viewing the model outputs The estimation program was described in Section 1 2 and comes to you as a compiled and unalterable file The Excel user interface is controlled by macros written in Visual Basic for Applications VBA Our objective was to allow the user to work with Excel s full capabilities while automating some complicated and time consuming procedures such as setting up a correctly d
72. urvey vulnerability type 12 age 2z length 166 1 167 Survey na zes C L likelihood type 158 169 Survey catch at length likelihood type ia ol 171 Survey catch at age likelihood type l 123 Figure A 1 5 Switches controlling the use of different likelinood components The Fixed Parameter section allows the user to input auxiliary information on biological parameters such as weight at age parameters von Bertalanfy growth parameters sd 1 age last age maturity ogive sex ratio pre 1962 selectivity ogive ageing errors and observed weight at age over time Given the lack of length data for this dataset many of these parameters are dummy variables Figure A 1 6 The last section of the input sheet corresponds with the data entries In row 233 we incorporate the time series of landings We have only one row of entries because this is a uni fleet fishery The number of landing observations should match the number of specified years for this analysis Lines 235 and 249 contain the number of CPUE and relative index observations For the CPUE column A specifies the index type and column B the fishing method The fishing method is related to the selectivity used to compute the vulnerable biomass The index type on the other hand is related to the catchability coefficient This means that one fishing method can have more than one index type e g temporal variation Column E specifies the coefficient of variation arou
73. vey year sample size 31 32 a3 238 1 1364 1 1 1 1 1 1 239 1 1365 1 1 1 1 1 1 300 Number of commercial catch at length data sets i l 202 DURE DATA Commercial catch at length data method year sample size 11 12 1 303 1 1364 1 304 1 1365 1 305 Number af survey Cael sm 3B DEPT DATA Survey catch at length data method year Sample size 11 12 13 309 1 1364 1 309 1 1365 1 210 umber of survey na zes COL data sets 4 212 DUMT DATA Survey na zes CooL data method year sample size 11 12 13 313 1 1364 1 314 1 1365 1 315 EndOFF orm 316 Figure A 1 7 continued Data section for the Northern cod case study At this point all the data should be in the Input sheet but before running the estimation model we recommend checking for missing data and inconsistencies in the spreadsheet Coleraine has a built in command that allows the user to do so Check Format Before Running Model Potential problems are highlighted in red or yellow After running this command check for colored cells in the entire spreadsheet MAIN MEMO Create Mew Input File m Run Model k Open Existing Excel Format Input File b view Output k Reconstituke Excel input File From text File MEME k Check Format Before Running Model k Check Parameter Values Before Running Model The next step is to do some deterministic projections before invoking the estimation program The motivation behind this is t
74. x gear age time Commercial fleet by sex gear age time The objective function includes likelihood components for the different data types and penalties on the variability of the stochastic parameters as specified by their bayesian prior distributions 1 4 1 Robust normal likelihood for proportions We use the robust likelihood formulation proposed by Fournier et al 1998 for the age sex and size sex catch compositions The observed frequency data is incorporated to the likelihood function as proportions at age and sex 7 Or at length 2 The robust normal model was selected instead of the more traditional multinomial error model because it is more robust to outliers Fournier et al 1990 Nyears Aplus 858 Sg InZ 0 5 2 2 log Pa 1 Pa 4 1 4 t 1 S az A initial Nyears Aplus M 2 p 2 log exp 0 01 Lin l where A and zr are respectively the number of classes and the inverse of the assumed sample sizes Aplus Aj 4 and Nyears are the age of the plus group the initial age observed in the samples and the number of available age composition samples respectively A similar formulation is used for the size sex compositions and is applicable for survey or commercial data 1 4 2 Abundance indices Different likelihood functions can be used for the commercial and survey indices These are normal log normal robust normal and robust log normal distributions The robu
75. xcel workbook xls Later you can check or modify it using Main Menu gt Create or Modify Input File gt Open Existing Input File Main Menu gt Create or Modify Input File gt Reconstitute Excel input file from text file or run it directly from Main Menu gt Run Model gt Run Other File gt Workbook xls file Coleraine Users Manual 19 4 Running Coleraine 4 1 General Issues Once you have entered all your data you may want to fit the model to the available information The instructions for running an open or another file were given in the section 3 5 g h of this manual When you choose any of these options the VB macros internally call the estimation executable program Colera20 exe for you Always remember that the estimation results are a function of the fixed parameter value input data likelihood type and priors For some entries Maincode xls gives you default values This was done to help you input the correct type of data but we strongly advise you to ensure all values are related to your own stock or match your assumptions about the dynamics 4 2 Parameter Estimation The statistical model used in the parameter estimation process uses a formal maximum likelihood approach Given the complexity and non linear structure of the model non linear estimation procedures using a Quasi Newton minimization algorithm are used to fit the model to the available data It has been recognized Seber and Wild 1989 that the final

"user manual"

Contents

Download Pdf Manuals

Related Search

Related Contents

&quot;user manual&quot;

Contents

Download Pdf Manuals

Related Search

Related Contents

"user manual"