Home

ASPIC 5.0 User`s Manual - R

1. 40 70 and starting value 50 Constrain the generalized estimate closer to the logistic than in Example 1 GENFIT EFT SSE 40 70 50 5 0 Example 3 Use a fitting procedure that could be accomplished with ASPIC 3 x Fit the logistic model conditioned on yield When fitting the lo gistic model only the first three values on line 3 are required LOGISTIC YLD SSE LINE 4 VERBOSITY amp OUTPUT FILE CONTROL This is a single integer value that controls the amount of output printed to the screen during ex ecution the verbosity and whether the optional Sum prn and rdat files 4 7 are generated 15 To control the amount of screen output set the base value within the range 0 4 A setting of 0 gives al most no little screen output 4 is intended for de bugging and gives too much for practical use The recommended level is 2 moderate screen output To generate SUM and PRN files recommended add 10 to any of the above values Thus the rec ommended default would become 12 To generate the RDAT file add 100 to any of the above values The recommended default verbosity with all files generated would be 112 A value of 102 is also valid to generate the RDAT but not the other optional files The files are controlled in this way instead of by a separate option in the input file to maintain com patibility with older input files LINE 5 BOOTSTRAP CONTROL Line 5 holds one or two integer val
2. LAV Least absolute values robust objective function Before setting values 4 5 and 6 on line 3 define as the decimal fraction defining model shape p Bysy K thus 0 lt lt 1 Then define nint 100q where nint is the nearest integer function and thus 0 lt lt 100 For example in the logistic model Busy K 2 so 0 5 and 50 For the Fox model exp 1 0 3679 and 37 t The fourth value on line 3 is an integer the lowest to consider in GENGRID and GENFIT shape modes A reasonable default might be 25 If present this value is ignored in LOGISTIC and FOX shape modes t The fifth value on line 3 is an integer the high est to consider in GENGRID and GENFIT shape modes a reasonable default might be 75 If present this value is ignored in LOGISTIC and FOX shape modes t The sixth value on line 3 is an integer whose interpretation depends on the shape mode chosen Shape mode Meaning of fifth value GENGRID Step size for grid of shape param eters examined GENFIT Starting value for shape parame ter LOGISTIC Ignored FOX Ignored In GENFIT shape mode a reasonable default is often to use the logistic i e to use 50 t The seventh value on line 3 is a real number amp that sets bounds to constrain the generalized fit near the logistic For example entering 8 0 here means that MSY for the generalized fit must lie be tween 1 8x and 8x the MSY est
3. This line contains a single real number between zero and one Set this value based on your belief about the stock s condition at the start of the data set In the absence of other information a reason able default is 0 5 See also 6 4 on page 18 LINE 15 STARTING GUESS FOR MSY In the absence of other information half the largest yield can be used as a starting guess This should be entered as a real number See also 6 4 on page 18 LINE 16 STARTING GUESS FOR K In the absence of other information a reasonable guess is 2 to 20 times the largest recorded yield This should be entered as a real number See also 6 4 on page 18 LINE 17 STARTING GUESS ES FOR q The program reads as many real numbers from this line as there are data series specified on line 12 The meaning of q depends on the data type that it refers to When it refers to an effort yield data series code CE in Table 1 q is the catchability coefficient When it refers to a biomass index data series codes IO I1 or I2 Table 1 q is the constant relating the index data to the internal ASPIC estimates of biomass e g if q 2 0 the index data are divided by 2 0 before being compared to the estimated biomass When it 17 refers to a biomass estimate series codes BO B1 or B2 Table 1 the user s value of q is ignored but a number must be present as a placeholder For technical reasons optimization is more difficult when q is large Thus a
4. As stated above the recommended procedure is to leave searching off unless it is needed If turned on suggested parameters are 1 50000 Monte Carlo searching and particularly repeated searching in creases execution time considerably LINE 7 CONVERGENCE CRITERION FOR OPTIMIZER This convergence criterion is a real number denoted After each adjustment of the simplex the ob jective function is computed for each vertex of the simplex Convergence is defined to occur when the following condition is met 2 L1 Lol Ly Lo where L is the highest objective function value in the simplex and Lo is the lowest The recommended value is 1 x 1078 which is written as 1d 8 in the input file Using a different value is not recommended LINE 8 RESTART CONTROL Randomized restarts are used by the ASPIC opti mizer to avoid local minima The two values real integer on line 8 control this mechanism tS The first value on line 8 is the tolerance 2 for ending restarts When objective function values from k restarts in a row agree to within this tol erance the solution is accepted The recommended value is 2 3x10 8 which is written as 3d 8 in the input file Changing this value is not recommended ts The second value on line 8 sets k the minimum number of restarts required The recommended de fault for this integer value is 6 Larger numbers can be used if needed to obtain a solution not overly sensitive to
5. Fitting the generalized model is done with a numeri cal solution of the catch equation and is thus slower than fitting the logistic model which has an analyt ical solution Execution will be especially slow in any the following cases 1 Poor agreement between model and data 2 Analysis of more than one data series C ksd 3 Bootstrapping especially in combination with 1 or 2 4 Extensive Monte Carlo trials in fitting Program speed can be improved by reducing the number of time steps per year in the input file see 6 3 on page 16 For the closest approximation to a continuous time model set this value to a large number e g 80 For a fairly close approximation use a number in the range 12 24 recommended For fastest operation set this to 2 steps per year 2 2 2 DEVELOPMENTAL FEATURES The following are believed to work correctly but have not been tested extensively Generalized estimation in all forms with more than one data series Objective functions other than SSE Projections of generalized bootstraps with ASPICP e The values of the AIC Akaike Information Cri terion printed when fitting the generalized model e The F statistic printed for comparing the lo gistic and generalized models may be incor rect when more than one data series is ana lyzed More importantly simulations suggest that such tests are of little value Prager 2002 It is wise to repeat estimation with several differen
6. Output is written to the screen The graphics program AGRAPH incorporates a stan dard Windows GUI with output available to any Windows printer or to a graphics file WMF or EPS It can be executed from the command line by drag and drop or by starting the program from its short cut 3 3 3 NMFS TOOLBOX The U S National Marine Fisheries Service NMFS has developed a toolbox of computer programs for stock assessment The toolbox currently in cludes a graphical editor specifically for ASPIC in put files and graphics that work with ASPIC output It also includes many other stock assessment and projection tools For further information contact Dr Paul Rago at the NMFS laboratory in Woods Hole Massachusetts 4 Overview of ASPIC 4 1 Data requirements Data needed by ASPIC are a series of observations on yield catch in biomass and one or more cor responding series of relative abundance Data on fishing effort rate can be used instead of relative abundance and when used are assumed to repre sent effective standardized effort ASPIC assumes that the supplied abundance index is an unbiased index of the stock s abundance in biomass If data on fishing effort are provided ASPIC assumes that effort divided by yield forms an unbiased index of the stock s abundance In this User s Guide the terms catch and yield are used interchangeably to mean total removals in biomass Similarly CPUE is used to mea
7. ber These are consecutive integers and must be identical from block data series to block Numbers greater than 9999 will not print correctly c2 Second number a real number whose meaning depends on the series type For type CE it is the fishing effort rate f for the year For type CC it is the average rel ative abundance usually based on CPUE For types BO B1 or B2 it is a stock biomass estimate For types I0 I1 or I2 it is a relative abundance value c3 Third number a real number required for CE or CC series only giving the total yield catch in biomass from the fishery for that year For other types of series Bn or In the third number is not needed c4 Final number an additional real number used only when the objective function is set to WIDSSE In that case it is an annual statistical weight which in each year is multiplied by the corresponding weight on line 13 to provide final statistical weight ing for the observation 18 Although yield effort data series are designated type CE effort is entered before yield on these lines Similarly in series type CC the relative abundance appears before the yield As noted in 6 3 it is recommended although not absolutely necessary to scale the catch and index data so that all q lt 0 01 This does not apply to Bn data series for which by definition q 1 6 4 Starting guesses and bounds The user may benefit from changing starti
8. questions Development of ASPIC and related research has been supported by the Southeast Fisheries Science Center of the U S National Marine Fisheries Service and by the author s private efforts This software is distributed to interested scientists free of charge No individual or group is authorized to charge for it or distribute it as part of any commercial product This manual is based on National Marine Fisheries Service Miami Laboratory Document MIA 92 93 55 which was later revised into Beaufort Laboratory Document BL 2004 01 dated February 2004 Typographical conventions In this manual user commands file names and items in input files are displayed in a monospaced font Some important sections are marked by a symbol in the margin as here attention to such material is especially important to obtaining good results from ASPIC Material new in this version of the program is marked by a different marginal symbol as here Michael H Prager Beaufort North Carolina January 2004 and May 2010 Contents Preface Typographical conventions Contents List of Tables 1 Introduction 2 New in version 5 0 2 1 Major changes 2 2 Using new features 2 2 1 Execution speed 2 2 2 Developmental features 3 Installation and Interaction 3 1 Compatibility 0 0 4 444 40 ee 8 32 Installation s recce a eee es Soo teria CE se 9 ane E a Gad aw ia 3 3 1 ASPIC an
9. read tableC aspic sum header TRUE 4 8 Starting ASPIC First prepare an input file in the correct format 6 on page 13 It s easiest to copy one of the sample input files provided to use as a template Then start the program giving the input file name on the command line For example the command aspic sword inp or just aspic sword will cause the program to read an input file named sword inp and produce corresponding estimates and output files If only the command aspic is given the program looks for the default input file ASPIC INP If the SUM file as been enabled summary output from each run in the directory will be written to it The default name is aspic sum To use a different name for the SUM file give the name on the com mand line For example the command aspic sword mysum will read the file sword inp and create or write to the summary file mysum sum along with the usual ASPIC output file s Most errors detected while ASPIC is reading the in put file will cause the program to print a descriptive message and stop If the message is not clear com paring the input file to the samples provided may reveal format errors 2Most operations are also possible by dragging and dropping icons The ASPIC Quick Reference available from the ASPIC shortcuts folder installed on the user s Windows Desktop in cludes more information on such use 12 5 Overview of Auxiliary Programs 5 1 O
10. siderably increase the time required to find a solu tion they can be helpful in avoiding local minima If a solution is difficult to find it can be helpful to enable the Monte Carlo searches When fitting difficult data sets it can be useful to make several runs with different random number seeds Agreement among a number of runs sug gests that the solution is stable Occasionally ASPIC fails to converge to a minimum at all This often indicates that the data do not fit the model very well which can sometimes be veri fied by examining the results with AGRAPH When there is no fit the input file should be checked for errors e g reversed catch and effort values Rarely changing the maximum value of F allowed line 10 of the ASPIC input file can improve con vergence if the problem occurs in EFF optimiza tion mode If the objective function appears from the screen output to have been near convergence simply trying a second ASPIC run that uses the first run s results as starting guesses can sometimes pro vide a good solution If the model includes several data sets fisheries it can be useful to eliminate one or more of them at least temporarily to see if con vergence can be achieved If none of these suggestions is successful estimates can often be made with the following strategy Set one parameter usually B K to a fixed value by set ting the corresponding estimation flag line 18 of the ASPIC input file to z
11. The title usually contains blanks and should be delimited by quotation marks The ASPICP output file will also include the title of the original ASPIC run LINE 2 NAME OF BIO FILE A character string specifying the name of a BIO file from an ASPIC bootstrap run Results from ASPICP will be based on the data in that file LINE 3 ANY CHARACTER STRING In earlier versions this line contained the output PRJ file name As of version 3 15 of ASPICP the output file always has the same root file name as the input CTL file and ends with file extension PRJ LINE 4 CONFIDENCE INTERVALS In previous versions a real number was required here as a placeholder but was not used in the com putations That is still acceptable and using any real number will cause defaults to be used for the options described immediately below As of version 3 16 of ASPICP additional user op tions can be given on this line in the form of one character string and one integer t The first value on line 4 is the type of confi dence interval to use This should be either BC for bias corrected confidence intervals Efron and Gong 1983 or PC for simple percentile confidence inter vals The latter may be useful when the BC intervals appear irregular t The second value on line 4 is an integer either 1 to smooth the resulting confidence intervals or 0 to use unsmoothed intervals The default when the user options are not given i e when a pla
12. difficulties The information in this section is central to obtain ing correct results Please read it thoroughly The optimization method used in ASPIC Nelder and Mead 1965 is quite robust but in unmodified form frequently stops at local minima these represent sub optimal solutions This has been addressed in ASPIC with a restarting algorithm that requires the same solution to be found several times in a row before it is accepted In the author s experience the resulting optimizer is reasonably effective at avoid ing local minima Nonetheless ASPIC like other programs that at tempt complex nonlinear optimization occasion ally finds local rather than global minima Two features of the program beyond the restarting al gorithm already mentioned are available to detect and remedy this problem First solutions obtained at local minima are often not reasonable and this C will often cause one of the parameters to be esti mated at either its minimum or maximum bound In such a case a warning message is printed both on screen and in the output file A second feature that can help avoid local minima is an optional Monte Carlo phase of estimation When enabled this tries to improve the initial fit by ran domly searching for a better one in the neighbor hood of the initial fit If multiple searches are en abled a shorter Monte Carlo search takes place peri odically during fitting Although such searches con
13. fisheries or biomass estimate series or biomass in dex series Data series may be of several types Ta ble 1 but at least one series must be type CE effort and yield or type CC CPUE and yield When more than one series is analyzed common estimates of B K MSY and K are made along with an estimate of qi for each series The interpretation of q de pends on the type of data series to which it pertains A statistical weight w for each fishery is specified by the user in the input file In summing the ob jective function each squared residual from fishery i is multiplied by w If the series have equal er ror variances using weights of unity for each series provides a maximum likelihood solution under the lognormal error structure assumed by ASPIC In FIT program mode the program normalizes the user s wi so that they sum to unity In IRF mode the program adjusts the weights interatively to pro vide nearly equal estimated variances Weights are also adjusted so they sum to unity The computer time needed to obtain estimates gen erally increases as more data series are added The increase is due both to addition of data and in creased difficulty of optimization 4 5 Objective function and penalty term Parameters are estimated under the assumption that the errors in yield or effort are multiplicative with constant standard deviation Thus the residu als are accumulated in logarithmic transform The objective func
14. on yield rather than on fishing effort or relative abundance The theory behind ASPIC and several worked ex amples were first presented in working documents of the International Commission for the Conserva tion of Atlantic Tunas ICCAT by Prager 1992a b Those reference have been superseded by the more formal and complete treatment of Prager 1994 The model and its extensions are also described in Quinn and Deriso 1999 and Haddon 2001 The basic theory of production models is of course also described in many other texts including Hilborn and Walters 1992 and is the subject of a recent FAO publication Punt and Hilborn 1996 The ASPIC computer program as described here has been used by several assessment groups and in many studies including Prager et al 1996 Prager and Goodyear 2001 Prager 2002 Shertzer and Prager 2002 and Williams and Prager 2002 In the course of those studies the program has been exercised on over 100 000 sets of simulated data The resulting experience has been used to improve the program s reliability Versions of ASPIC before 5 0 are no longer main tained by the author They are still available for those who wish to duplicate old analyses exactly However the author urges analysts to use current versions for all new work 2 New in version 5 0 This section gives an overview of changes intro duced between ASPIC 3 x and ASPIC 5 x Although this section will be of most interest
15. programmer s editor such as Windows Notepad Good text editors are avail able as freeware shareware or commercial soft ware A useful feature in any editor used for ASPIC in put files is the ability to cut and paste rectangular blocks of text A relatively simple editor with that feature is ConTEXT which as of September 2009 was available free and locatable through Web search engines e g Google Other well known editors such as emacs are also suited to this task 4 7 3 OUTPUT FILES ASPIC produces several output files Although all are written in plain ASCII some are meant for hu man readability and others for use by ancillary pro grams The main fit and bot output files like Table 2 Files read R or written W by ASPIC and related programs File type Action Used File contents and description by inp R ASPIC Input file with data starting guesses and run settings fitt W ASPIC Main output file from FIT and IRF program modes bot W ASPIC Main output file from BOT program mode biot W R ASPIC Estimated B and F trajectory for each bootstrap trial read ASPICP by ASPICP BOT program mode dett W ASPIC Estimates from each bootstrap trial BOT program mode SUM W ASPIC Optional file with summary of all runs made in a directory prn W ASPIC Estimated trajectories in a table easily read by S Plus R SAS or spreadsheet rdatt W ASPIC Detailed inputs and estimates specially formatted for read
16. starting values The value used by ASPIC 3 x was k 3 increased in version 3 89 to k 6 16 LINE 9 CONTROL OF ITERATIVE COMPUTATIONS Iterative computations are used by ASPIC in several places The two values real integer on line 9 con trol two important sets of iterative computations ts The first value on line 9 is the tolerance 3 for computing the annual fishing mortality rate F When conditioning on yield an iterative method must be used to estimate F it continues until suc cessive estimates are within 3 The recommended value is 3 1x107 4 which is written as 1d 4 in the input file Changing this value is not recommended In EFT optimization mode 3 must be present in the input file but it is ignored t The second value on line 9 is the number of time steps used per year for the generalized model range 2 100 A reasonable default is between 12 and 24 steps The choice affects execution speed 2 2 1 When fitting the logistic Schaefer model this num ber is not required If present it is ignored LINE 10 MAXIMUM ESTIMATED F This line contains a real number specifying the max imum allowable estimate of F This maximum used when conditioning on yield serves to aid the op timizer The recommended default is 8d0 which works well in most cases LINE 11 STATISTICAL WEIGHT FOR B PENALTY IN OBJECTIVE FUNCTION This line contains a real number that controls the influence of the pen
17. statistics not fishing mortality rate The test statistic is _ SSEs SSEe V1 F SSE v2 2 where SSE and SSE are is the error sums of squares of the simple and complex models respectively v is the difference in number of estimated parameters between the two models and v2 is the number of data points less the total number of estimated pa rameters The significance probability of F can be obtained from standard tables of the F distribution with v and v2 degrees of freedom A small program called FTEST is supplied with ASPIC to facilitate making certain such tests This pro gram assumes that the same data are used for both models but are divided into different periods with different estimates of q The weighting for the penalty term line 11 in the ASPIC input file should be set to zero for this F ratio test to be theoretically correct The FTEST program is interactive and takes all input from the screen Three often repeated caveats apply when using the F ratio test for this purpose First hypothesis tests are invalid when suggested by examination of the data Instead the test should be suggested by ex ternal information such as changes in gear Sec ond the significance of a series of tests is less than that of a single test For information on this point consult a reference on multiple comparisons e g Klockars and Sax 1986 Third hypothesis tests generally assume correct specification of the
18. ASPIC conducts an iter atively reweighted fit when two or more data series are analyzed Iterative reweighting of the data series inverse variance weighting pro vides under many circumstances a maximum likelihood solution The modes BOT and IRF cannot be combined In other words ASPIC cannot run a bootstrap on an iteratively reweighted fit A typical analysis might begin with FIT mode in cluding several runs to explore different model structures If questions about series weighting are to be addressed IRF mode might be used instead in the initial analysis After model and data struc ture have been decided BOT mode can be used to estimate the uncertainty in assessment results ASPIC bootstrap runs do not incorporate iterative reweighting and this will cause underestimation of variability when IRF mode has been used to develop a model structure IRF mode was developed in response to requests during assessment workshops but it has not been much used by the author and thus is less thor oughly tested than the other modes Experience has shown that while series weights estimated in IRF mode may be statistically unbiased they can be of high variance This is a characteristic of such weights generally and is not specific to ASPIC For that reason sensitivity to series weights should be examined whenever IRF program mode is used 4 4 Fitting more than one data series ASPIC can fit data on up to 10 simultaneous or serial
19. User s Manual for ASPIC A Stock Production Model Incorporating Covariates ver 5 And Auxiliary Programs w k g Michael H Prager Prager Consulting Portland Oregon USA www mhprager com w w g Last revised April 17 2011 Preface This user s manual describes Version 5 0 of ASPIC a computer program to estimate parameters of a non equilibrium surplus production model from fisher ies data Several utility programs ASPICP FTEST AGRAPH are also described Their purposes are making projections comparing models and quickly making graphs from ASPIC and ASPICP output files The programs together are referred to here as the ASPIC Suite The major change from previous versions of ASPIC is the ability to fit the Pella Tomlinson generalized production model with the Fox exponential yield model included as a special case The Schaefer lo gistic production model the main component of earlier versions is still part of ASPIC and because many of its computations can be done analytically rather than numerically it will be found quicker and its solutions may be more stable The ASPIC Suite is not commercial software and the programs are not warranted in any way either by the author or by any of the sponsors who made de velopment possible The software was developed for use in the author s research and it is used reg ularly for stock assessment and education Distri bution to fellow scientists in made in a
20. alty term on B gt K see 4 5 1 To omit the penalty term set this to 0d0 To use the penalty term enter a positive real number usu ally 1d0 The penalty is useful in analyses showing a sharp decline in relative abundance in the initial years such data sets can otherwise result in an ex tremely high estimate of B The recommended default is no penalty If the re sulting estimate of B K is too high the analyst can try either the penalty term or fixing B K rather than estimating it Either approach can affect es timates of management quantities sensitivity anal yses are useful to examine this The penalty term is described in Prager 1994 fixing B1 in Punt 1990 LINE 12 NUMBER OF DATA SERIES This line has a single integer from 1 to 10 that in dicates how many data series are to be analyzed The types of allowable data series are summarized in Table 1 on page 10 LINE 13 SERIES SPECIFIC STATISTICAL WEIGHTS The program reads as many real numbers from this line as series were specified on the preceding line The statistical weight w for series i is multiplied by each squared residual for that series when the objective function is computed When IRF program mode is used to analyze more than one data series the w are adjusted to implement inverse variance weighting They can all be set to unity best written 1d0 in the input file unless there is reason to set them otherwise LINE 14 STARTING GUESS FOR B K
21. ata S African Journal of Marine Science 9 249 259 Punt A E and R Hilborn 1996 Biomass dynamic models User s manual FAO Computerized Infor mation Series Fisheries No 10 Food and Agri culture Organization of the U N Rome 63 p Quinn T J and R B Deriso 1999 Quantitative Fish Dynamics Oxford University Press New York 542 p Schaefer M B 1954 Some aspects of the dynamics of populations important to the management of the commercial marine fisheries Bulletin of the Inter American Tropical Tuna Commission 1 2 27 56 Schaefer M B 1957 A study of the dynamics of the fishery for yellowfin tuna in the eastern trop ical Pacific Ocean Bulletin of the Inter American Tropical Tuna Commission 2 247 268 Shertzer K W and M H Prager 2002 Least me dian of squares a suitable objective function for stock assessment models Canadian Journal of Fisheries and Aquatic Science 59 1474 1481 Stine R 1990 An introduction to bootstrap meth ods examples and ideas Pages 325 373 in J Fox and J S Long eds Modern methods of data analysis Sage Publications Newbury Park Cali fornia 446 p Williams E H and M H Prager 2002 Comparison of two estimators for the generalized production model Canadian Journal of Fisheries and Aquatic Science 59 1533 1552 26
22. ber of identical returns is now specified in the input file The recommended default is 8 This can improve stability of the fit on some poor data sets The value is specified on line 8 of the input file see 6 3 on page 16 Time steps The generalized model is implemented by numerical integration that approximates a continuous time solution The number of time steps per year is specified on line 9 of the input file see 6 3 on page 16 Length of series Time series in the input file may now be up to 250 years long Updated ASPICP An updated version of the projec tion program ASPICP is compatible with analyses from ASPIC 5 0 It is also backwardly compatible with ASPIC 3 x and 4 x Starting with version 3 15 of ASPICP the projection output filename is derived from the input filename rather than being specified by the user in the input file Windows installer This release is distributed as a self installing binary file for Windows Versions for other operating systems may be available on re quest Drag and drop versions ASPIC has always been a non interactive program that reads from and writes to ASCII files This release includes alternative ver sions that accept drag and drop of input files and that display their output in scrolling windows The original command line versions are also supplied 2 2 Using new features When using the new features of ASPIC 5 0 please consider the following 2 2 1 EXECUTION SPEED
23. ceholder is given instead is to com pute BC intervals and smooth them The author is aware that some BC intervals from ASPICP are very wide in certain years Because that seems illogical the author has checked the corre sponding programming code repeatedly but has not discovered any errors Using PC intervals has avoided the irregularity in the cases tested LINE 5 NUMBER OF YEARS TO SKIP AT START OF PLOTS An integer with recommended values 0 to 3 The first few years of biomass and mortality estimates are especially imprecise Also analysis of certain data sets can give in very high estimated biomasses in the first few years Thus omitting the first few years from the plots can be useful LINE 6 NUMBER OF YEARS OF PROJECTIONS Integer between 0 and 15 The longer projections extend the more speculative they are For that rea son early versions of ASPICP limited projections to only 10 years but some users found that overly restrictive Projections are theoretical constructs and are most useful when comparing management strategies rather than as forecasts of the future FOLLOWING LINES PROJECT MANAGEMENT REGIME TO Each following line has data for one projection year the number of lines should equal the number of years specified on line 6 On each line enter a real number followed by a single character The real number represents the yield or relative fishing mor tality rate to be applied that year and the charac
24. cooperative spirit The software is intended as a set of research tools and those who use them do so at their own risk ASPIC has been used on thousands of real and simulated data sets and all supplied programs are believed to be substantially correct The author ap preciates receiving advice of suspected flaws and he attempts to correct errors promptly By no means is ASPIC the final word in production modeling It is intended as a reasonably flexible pro gram that can serve as a basis for further innova tion Formal description of the theory behind ASPIC is given in Prager 1994 Further references are given in the bibliography The author requests that this manual and Prager 1994 be cited in any report or published article that uses ASPIC Those who have used version 3 x of ASPIC and who now are presented with version 5 x might ask what happened to version 4 x The answer is simple 4 x were test versions It seemed more logical to release the new version as 5 0 rather than some number in the middle of the 4 x series Many colleagues have given valuable technical sug gestions or assistance while ASPIC was being writ ten and as it has been revised through the years I thank S Cadrin R Deriso K Hiramatsu J Hoenig R Methot C Porch J Powers A Punt V Restrepo G Scott K Shertzer P Tomlinson D Vaughan E Williams and the many fishery scientists who have sent data sets to illustrate their applications and
25. d ASPICP 3 3 2 Auxiliary programs 3 3 3 NMFS Toolbox 4 Overview of ASPIC 4 1 Datarequirements 4 2 Program limits 4 3 Program modes 4 4 Several dataseries 4 5 Objective function 4 5 1 Penalty for initial biomass 4 5 2 Conditioning on yield 4 6 Bootstrapping 4 7 Input and output files 4 7 1 Inputfile 4 7 2 Editing input files 4 7 3 Output files 4 8 Starting ASPIC 2 4 2 ee ee ee x 5 Auxiliary Programs 5 1 Overview of ASPICP 5 2 Overview of FTEST 5 3 Overview of AGRAPH 6 ASPIC Input File Specification 6 1 Generating a sample input file 6 2 General format guidelines 6 3 The ASPIC input file line by line 6 4 Starting guesses andbounds 6 5 Questions about data series 6 5 1 Missing values and zeroes 6 5 2 Allocating yield 6 5 3 CAMIIOM oie ace ge a 7 Advanced Options for ASPIC 8 ASPICP Input File Specification 8 1 Line by line 5 26 i 25sw daw 8 8 2 Sample input file 9 Interpretation of ASPIC Results 9 1 Precision of parameter estimates 9 2 Using several data series 9 2 1 Catchability over time O S PKOJECHIONS 2 2 5 8 4 4 Goad ae Gow Bh 9 4 Estimation difficulties 10 Program Changes 11 Source Code Ref
26. d allows duplication of a previ ous run Using a different seed should result in the same answer within expected computation errors if re sults are substantially different at least one of the solutions was a local minimum The user can at tempt to remove sensitivity to random number seed by increasing the number of restarts required sec ond number on line 8 LINE 22 NUMBER OF YEARS IN DATA SET The total number y of years described by the input file including any years with missing values Within the file each data series must be of length y and describe the same specific years Nonoverlapping series can be accommodated by padding each series with missing values or zeroes as appropriate FOLLOWING LINES INDIVIDUAL DATA SERIES There must be one data block group of lines for each data series Each block should include data for all y years thus thus each data block must be the same length y The composition of each block is as follows a On the first line of the block a series title char acter string length lt 40 in quotes Example Spring survey amp total landings b On the second line of the block a character string of length 2 with the type code for the series Type codes are listed in Table 1 p 10 lt Starting on the third line of the block one data line for each year with the following data on each line separated by blanks c c1 First number the year or other ID num
27. d for an observed zero Zero values of the abundance measure CPUE are never permitted because it is assumed that the re source is not extinct during the analysis period If an abundance index calculated prior to using 2 ksd Table 3 Actions taken by ASPIC when data series include data record s with missing value s or zero es Dash indicates normal data neither missing nor zero M indicates a record with missing datum Z with zero datum Series type Index includes IQ I1 I2 BO B1 and B2 series Cond Series CPUE or mode type effort Yield Action by ASPIC YLD CC M Fit estimate missing CPUE YLD CC M Stop missing yield not allowed when conditioning on yield YLD CC M M Stop missing yield not allowed when conditioning on yield YLD CC M Z Fit with F Y 0 no fishing YLD CC Z M Stop missing yield not allowed when conditioning on yield YLD CC Z Stop zero CPUE never allowed YLD CC Z Fit with F Y 0 no fishing YLD CC Z Z Stop zero CPUE never allowed YLD CE M Fit estimate missing CPUE YLD CE M Stop missing yield not allowed when conditioning on yield YLD CE M M Stop missing yield not allowed when conditioning on yield YLD CE M Z Fit with F Y 0 no fishing YLD CE Z M Stop missing yield not allowed when conditioning on yield YLD CE Z Stop for error when F 0 Y must be 0 YLD CE Z Stop zero CPUE never allowed YLD CE Z Z Fit with F Y 0 no fish
28. des a simple method of neatly printing ASPIC output files which are 120 charac ters wide Example Run 4 for Redfin Tuna 1994 An alternative for printing ASPIC output files is to use the freeware utility PrintFile http www lerup com printfile LINE 3 MODEL SHAPE AND OPTIMIZATION CONTROL Note In an effort to make ASPIC 5 0 as compatible as possible with earlier versions the input file has the same general arrangement However additional control values are needed Many of them appear on line 3 which make this section rather long Line 3 has a varying number of items t The first value on line 3 is a character string specifying the model shape program shape mode Value Meaning LOGISTIC Fit the logistic Schaefer model GENGRID Fit the generalized model at grid of values or at one specified value FOX Fit the Fox model a special case of GENFIT below GENFIT Fit the generalized model and esti mate its exponent directly t The second value on line 3 is a character string specifying the conditioning mode for the fit For more information see 4 5 2 on page 9 Value Meaning YLD Condition fitting on yield recommended for most analyses EFT Condition fitting on fishing effort rate ts The third value on line 3 is a character string specifying the objective function 14 Value Objective function SSE Sum of squared errors recommended de fault WTDSSE SSE with annual data weighting
29. eastern tropical Pacific ocean Dissertation University of Washington Seattle Pella J J and Tomlinson P K 1969 A general ized stock production model Bulletin of the Inter American Tropical Tuna Commission 13 3 419 496 Prager M H 1992a ASPIC A Surplus Production Model Incorporating Covariates Collected Vol ume of Scientific Papers International Commis sion for the Conservation of Atlantic Tunas IC CAT 28 218 229 Prager M H 1992b Recent developments in ex tending the ASPIC production model ICCAT Working Document SCRS 92 127 10 p Prager M H 1994 A suite of extensions to a nonequilibrium surplus production model Fish ery Bulletin 92 374 389 Prager M H 2002 Comparison of logistic and gen eralized surplus production models applied to swordfish Xiphias gladius in the north Atlantic Ocean Fisheries Research 58 41 57 Prager M H and C P Goodyear 2001 Effects of mixed metric data on production model estima tion simulation study of a blue marlin like stock Transactions of the American Fisheries Society 130 927 939 Prager M H C P Goodyear and G P Scott 1996 Application of a surplus production model to a swordfish like simulated stock with time changing gear selectivity Transactions of the American Fisheries Society 125 729 740 Punt A E 1990 Is B K an appropriate as sumption when applying an observation error production model estimator to catch effort d
30. ed 6 5 2 ALLOCATION OF YIELD AMONG SERIES When analyzing more than one data series it is not always possible or desirable to associate a KZ yield with each measure of fishing effort rate or rel ative abundance A common example is having sev eral abundance indices for a stock but only the to tal annual yield This section aims to describe how ASPIC assumes yield is allocated among data series Yield is entered in both CE and CC series Because ASPIC derives an abundance index from each CE se ries it is important that the yield in a CE series cor respond to the fishing effort rate in the same series In contrast it is not necessary for the abundance in dex in a CC series to correspond to the yield in the same series For example a valid CC series might have an abundance index computed from one fish ery on a Stock paired with the total catch from all fisheries on that stock Despite the above it is important that yield summed across series represent a constant pro portion usually assumed 1 0 of total removals Changes in that proportion whether due to report ing changes or changes in discarding practices vi olate a fundamental assumption of ASPIC and of most other assessment models The consequences of that violation will depend on its severity 6 5 3 CAUTION ON ZEROS MISSING VALUES The author has attempted to ensure that results of computations including missing and zero values are correct
31. ella Tomlinson system Fishery Bulletin 76 515 521 Fox W W Jr 1970 An exponential yield model for optimizing exploited fish populations Transac tions of the American Fisheries Society 99 80 88 Fox W W Jr 1975 Fitting the generalized stock production model by least squares and equilib rium approximation Fishery Bulletin 73 23 37 Graham M 1935 Modern theory of exploit ing a fishery and application to North Sea trawling Journal du Conseil International pour lExploration de la Mer 10 264 274 Freedman D A and S C Peters 1984 Bootstrap ping a regression equation some empirical re sults Journal of the American Statistical Associa tion 79 97 106 Haddon M 2001 Modeling and quantitative meth ods in fisheries Chapman amp Hall Boca Raton Florida 406 p Hilborn R and Walters C J 1992 Quantitative fisheries stock assessment Chapman and Hall New York NY 570 p Klockars A J and G Sax 1986 Multiple compar isons Sage University Quantitative Applications in the Social Sciences Paper No 07 61 Mohn R K 1980 Bias and error propagation in logistic production models Canadian Journal of Fisheries and Aquatic Science 37 1276 1283 Nelder J A and R Mead 1965 A simplex method for function minimization Computer Journal 7 308 313 Pella J J 1967 A study of methods to estimate the Schaefer model parameters with special ref erence to the yellowfin tuna fishery in the
32. emplate for making new input files It could also be helpful to have it available when reading this section 6 2 General format guidelines The representation of values in the input file must follow certain rules which follow from the use of Fortran list directed read statements to read the data e The exact position of values on a line is not im portant However if a line contains more than one value they must be in the correct order e When a line contains more than one value they must be separated by spaces blanks Using tab characters to separate values is not recom mended The remaining rules depend upon the type of the data item integer real or character 13 e Each real number should contain a decimal point an exponent marked by the letter d or e or a decimal point and an exponent Examples 1 0 2e3 1 3d6 Note that the notation 2e3 means 2 x 10 An integer can be used in place of a whole real number e Integers must not contain decimal points or ex ponents Examples 0 2 94541 e Character strings may be delimited by matched apostrophes or quotation marks However this is necessary only if the string contains embed ded blanks or other special characters Exam ples each on a separate line IRF This is a valid string Another valid string e Each line must have the specified number of values separated by spaces Values may not be otherwise arranged among lines e After the s
33. erences 12 12 12 13 13 13 13 13 18 18 18 19 20 20 20 20 21 21 21 22 22 23 23 24 25 25 List of Tables 1 Data series types and codes 10 2 Files used by the ASPIC suite 11 Model shape codes 14 Conditioning codes 14 Objective function codes 14 Model shape settings 14 3 Missing values and zeroes 19 1 Introduction This user s manual describes Version 5 0 of ASPIC a computer program to estimate parameters of a non equilibrium surplus production model from fisher ies data Several utility programs ASPICP FTEST AGRAPH are also described Their purposes include making projections comparing models and quickly making graphs from ASPIC and ASPICP output files The programs together are referred to here as the ASPIC Suite The surplus production model has a long history in fishery science and has repeatedly proven useful in management of fish stocks The appeal of produc tion models is in large part due to their conceptual and computational simplicity Despite that simplic ity production models incorporate an implicit re cruitment function and thus can be used for stud ies of sustainability Production models have also been found especially useful in stock assessments when the age structure of the catch cannot be esti mated Many early treatments of surplus production mod els assumed that the yie
34. ero A solution might be possible conditional upon that value of B K If this technique leads to a solution a range of fixed values of B K can be tried and the solutions examined Similar values of the objective function among so lutions indicate that the solutions are nearly equiv alent in terms of fit Although the solutions will differ somewhat they still may be useful especially as confirmatory information or if little other infor mation is available for management 24 Although ASPIC has been tested on thousands of simulated and real data sets and is believed to oper ate correctly errors can exist in any computer pro grams Any user experiencing bugs or suspected bugs is asked to send the author copies of the input and output files by email Mike Prager noaa gov The author attempts to correct all errors promptly 10 Program Change History Changes beyond Version 5 00 e Version 5 01 Added error message to output for q out of bounds Version 5 02 Increased internal file name length to 128 characters This has become an issue for the new drag and drop version ASPICPW EXE Version 5 03 Fixed rare problem in initializing simplex when no starting values of K MSY or q given by user Also increased minimum num ber of convergences required in LAV optimiza tion mode from 5 to 8 Version 5 04 Fixed some inconsistencies and errors in handling missing values Verified cor rect action in all 36 possible combination
35. estimate some quantities considerably more precisely than others Among the quantities more precisely estimated are maxi mum sustainable yield MSY optimum effort fusy and relative levels of stock biomass and fishing mortality rate Here relative levels means the bio mass level relative to the level at which MSY is at tained or the level of fishing mortality relative to that at which MSY is attained To provide more precise estimates then it is often useful to divide the stock size estimates provided by ASPIC by the corresponding estimate of stock size at MSY Bysy Similarly the estimates of fish ing mortality rate F are divided by Fysy to obtain relative estimates In its output files ASPIC pro vides such relative estimates The relative estimates present a more precise picture of the condition of the stock because in normalization the estimate of q which is usually imprecise cancels out In contrast absolute levels of stock biomass and related quantities which include uncertainty in the estimate of q are usually estimated much less pre cisely One cannot place nearly as much credence in the absolute estimates of stock size F or any quan tities that depend upon them Absolute estimates of B and F from ASPIC are provided for the mod eler s information and are not intended for use as management guidelines When two or more data series are analyzed es timated ratios of catchabilities are typically est
36. g biomass ratio and of Busy Conditioning options Option names for condition ing on yield or effort have been revised to indicate conditioning rather than residuals The new speci fications are given in 6 3 Fitting criteria An additional objective function least absolute values LAV is available It is recom mended that this robust objective function be used only in conjunction with a regular least squares fit because the optimizer has a more difficult task in finding the best minimum of LAV fits Nonetheless LAV can be valuable where one or more data ob servations are markedly disjoint from the rest For guidelines on appropriate use of LAV please con sult the statistical literature Bounds on catchability coefficient To improve con vergence estimates of q catchability are now bounded to a geometric range around the user s starting guess The bounds are determined inter nally by ASPIC and are not under the user s control If the starting guess for q is severely wrong the es timate may hit a bound and the ASPIC report will indicate whether the starting guess was too low or too high If that happens the user should revise the starting guess of q accordingly and rerun the analy sis Restarts during optimization Previous versions of ASPIC required the optimizer to return to the same solution 3 times in a row to indicate convergence That worked well on most data sets but was not always sufficient The num
37. i mated more precise than estimates of each q Also K may be estimated imprecisely or inaccurately even when MSY and fysy are estimated well Again this reflects the difficulty of translating relative bio mass changes to an absolute scale The starting biomass estimated as B K may be considered a nuisance parameter and its estimate is often imprecise Punt 1990 recommended fixing B K 1 0 rather than estimating it for the Cape hake stock off southern Africa but it is not clear that that approach is appropriate for every stock A similar approach is taken in using the penalty term described in 4 5 1 To stabilize estimates from a particular data set it can be useful to fit the model with B K fixed at a range of values Although the resulting estimates of the biomass trajectory will of course diverge at the beginning they may provide sufficiently consistent estimates of present stock status for management purposes 9 2 Estimating several catchability coefficients ASPIC can use more than one data series in esti mation The analyst should be aware that the un derlying assumption is that each abundance mea sure reflects the entire stock except for random er ror Thus using this feature is similar to deriving an abundance index from each series and averaging them together 22 It is not recommended to use abundance indices that are uncorrelated or negatively correlated with one another unless their overlap is sho
38. imated in the lo gistic fit This ad hoc method is used to increase stability in fitting In simulation studies the value amp 8 0 has proven a reasonable default The pa rameter K is constrained in the same way but using g 605 During bootstrap runs the change in bounds is effective only for the initial fit of the generalized model For remaining bootstrap trials bounds on MSY and K revert to those entered by the user on lines 19 and 20 Thus it is wise to adjust user guesses and bounds after fitting a point estimate and before fitting a bootstrap If the input file specifies 0 0 no change of bounds is made When 0 0 it is required that amp gt 1 1 As noted above 8 0 is recommended as a general default To use a fixed model shape use GENGRID shape mode set the fourth and fifth values equal to the specified shape and set the sixth value the step size to zero EXAMPLES OF LINE 3 Example 1 Specify a grid search for the shape pa rameter between 40 and 60 a moderate range around the logistic 50 Use a step size in of 5 Set bounds on MSY of 1 8x to 8x the logistic estimates Use SSE least squares objective function in log space conditioned on matching the effort in the input file Note that EFT here is the equivalent to the optimization mode CAT in ASPIC 3 x GENGRID EFT SSE 40 60 5 8 0 Example 2 Fit the generalized model conditioned on effort with bounds of
39. ing YLD Index M Fit estimate missing CPUE YLD Index Z Stop zero CPUE never allowed EFT CC M Stop missing CPUE not allowed when conditioning on effort EFT CC M Stop missing yield not estimable in this case EFT CC M M Stop missing effort not allowed when conditioning on effort EFT CC M Z Stop missing effort not allowed when conditioning on effort EFT CC Z M Stop zero CPUE never allowed EFT CC Z Stop zero CPUE never allowed EFT CC Z Fit with F Y 0 no fishing EFT CC Z Z Stop zero CPUE never allowed EFT CE M Stop missing effort not allowed when conditioning on effort EFT CE M Fit estimate missing catch EFT CE M M Stop missing effort not allowed when conditioning on effort EFT CE M Z Stop missing effort not allowed when conditioning on effort EFT CE Z M Estimate with F Y 0 no fishing EFT CE Z Stop if F 0 Y must be 0 EFT CE Z Stop zero CPUE never allowed EFT CE Z Z Estimate with F Y 0 no fishing EFT Index M Fit estimate missing effort EFT Index Z Stop zero CPUE never allowed ASPIC is zero in a given year one could try us ing a small number e g 20 to 50 of the low est nonzero value in its place Use of an extremely small number e g 1 of the lowest nonzero value will usually result in a large residual during the ASPIC analysis such a residual can influence the 19 results strongly Thus converting zeroes to very small numbers is not recommend
40. ing by dget function of S Plus or R gen W ASPIC Summary results from GENGRID mode grd W ASPIC Summary results from LOGGRID mode ctl R ASPICP Control file in which user specifies projection parameters prj W ASPICP Projection results prb W ASPICP Extension to bio file with projection results File types readable by AGRAPH t File types intended for reading mainly by other computer programs all the files meant to be read by the user have line lengths of 120 characters or less Files meant for ancillary programs may have longer lines Names of output files are constructed from the root file name of the run s input file with the appropri ate file extension appended Table 2 Main output files The main ASPIC output file in cludes parameter estimates measures of goodness of fit and estimates of population benchmarks bio mass levels and exploitation levels as well as sim ple character plots Output from bootstrap runs also includes bias corrected confidence intervals on parameters and on other quantities of management interest The file extension of the main output file varies with mode of program operation Table 2 Suppose the data input file is named sword inp Then in FIT and IRF modes the main output would be written to a file named sword fit In BOT mode the main output would be written to file sword bot Bootstrap related files Two additional files are generated after boostrap runs to all
41. irectory e Adds a GINO environment variable pointing to the installation location as required by a graph ics support library used in AGRAPH If a GINO environment variable already exists on the sys tem it is not modified e Adds shortcuts to the Windows Start menu and Desktop including a command window open ing in a user specified working directory e Adds an uninstaller to the installation location and adds ASPIC to the system s Remove Pro grams list The ASPIC uninstaller removes all of the above However any files added by the user are not re moved This User s Manual is supplied with all distri butions of ASPIC as an Adobe PDF file named ASPICMAN PDF It may be distributed freely 3 3 Interface 3 3 1 INTERFACE OF ASPIC AND ASPICP The standard versions of ASPIC and ASPICP do not include graphical user interfaces Instead the pro grams are console mode character programs that 2 ksd read all input from and write all output to ASCH text files The screen is used only for status mes sages For added ease of use the ASPIC 5 0 instal lation includes versions of ASPIC and ASPICP that support drag and drop The executable files of the command line versions are aspic exe and aspicp exe of the drag and drop versions aspicw exe and aspicpw exe 3 3 2 INTERFACE OF AUXILIARY PROGRAMS The auxiliary program FTEST has a text mode user interface is interactive and does not require an in put file
42. ld taken each year could be considered the equilibrium yield e g Fox 1975 However using that equilibrium assumption tends to overestimate MSY when used to assess a declin ing stock and it has been found problematic by sev eral studies Mohn 1980 Williams and Prager 2002 The assumption was a computational convenience that is no longer needed and ASPIC does not use it Earlier versions of ASPIC could fit only the logistic production model Schaefer 1954 1957 Pella 1967 in which the production curve curve of surplus pro duction vs biomass is symmetrical around MSY Version 5 x also fits the generalized model of Pella and Tomlinson 1969 in the revised parameteriza tion of Fletcher 1978 ASPIC incorporates several extensions to classical stock production models One extension is that ASPIC can fit data from up to 10 data series These may be catch effort series from different gears or different periods of time catch abundance index series biomass indices or biomass estimates made independently of the production model This fea ture is described in 4 4 A second major extension is the use of bootstrapping for bias correction and construction of approximate nonparametric confi dence intervals A third extension is that ASPIC can fit a model under the assumption that yield in each year is known more precisely than fishing ef fort or relative abundance in other words fitting can be statistically conditioned
43. lly estimates of relative population trend B Bysy and relative fish ery trend F Fysy from the two models have been similar Occasionally when abundance indices are noninformative the age structured model can esti mate population trends reliably while the produc tion model cannot I have not yet encountered data that supported both types of estimation with rea sonable precision yet give strikingly different re sults from various models Nonetheless one area in which from age aggregated models tend to differ from age structured models is in estimating recovery from overfishing In the au thor s experience projections based on production model dynamics often estimate considerably faster recovery than do projections from age structured methods It is too early to say categorically that one type of projection is more accurate than the other because few stocks so assessed have recovered from overfishing Still it is the author s opinion that results based on production models may be overly optimistic particularly for species that mature at relatively old ages Age structured models can cap ture the lag in recovery that occurs as increased re cruitment and survival propagate through the im mature age classes Non age structured models do not explicitly capture this so in cases where deple tion extends below the age of maturity they may project recovery at a rate not attainable by the ac tual population 9 4 Estimation
44. model Thus significance probabilities of tests on assess ment models are always approximate This caveat is especially important when there is evidence that the model does not fit well A nonparametric test of the null hypothesis q q2 can be conducted from the fitting results This test is constructed by examining the bootstrap es timates of the ratio of the two catchability coeffi cients As an example assume that the alternative hypothesis is that q1 q2 Then the null would be rejected at P lt 0 05 if a bias corrected 95 confi dence interval on q q2 did not include the value 1 0 Like the F test this test is approximate be cause of the possibility of specification error In ad dition bootstrapping residuals may underestimate the true variability present in a time series Freed man and Peters 1984 This has been addressed to some degree in the current version of ASPIC by the adjustment made to the residuals before bootstrap ping is begun 9 3 Projections The ASPIC suite includes the program ASPICP which can be used to make projections of the stock s re sponse to various management measures In inter preting results from ASPICP it is helpful to con sider the differences between age structured and non age structured models The author has participated in many assessments in which both age aggregated production models such as ASPIC and age structured cohort models 23 were used on the same stock Usua
45. n relative abundance The presumption is that the CPUE has been standardized before being used for modeling In addition to data ASPIC requires starting guesses of its estimated parameters Parameters directly estimated are K the stock s maximum biomass or carrying capacity MSY the maximum sustainable yield B K the ratio of the biomass at the begin ning of the first year to K and for each data series i qi the catchability coefficient for that series De scription of the input file format given in 6 in cludes suggestions for starting guesses 4 2 Program limits The array limits of ASPIC are as follows e Number of years of data 250 e Number of data series 10 e Number of bootstrap trials 1 000 Any user with larger requirements is invited to con tact the author 4 3 Program modes ASPIC has three modes of operation here called program modes e In FIT program mode ASPIC fits the model and computes estimates of parameters and other quantities of management interest including time trajectories of fishing intensity and stock biomass Execution time is relatively short In BOT program mode ASPIC fits the model and computes bootstrapped confidence inter vals on estimated quantities Because computa tions are extensive execution time in BOT mode is considerably longer than in FIT mode For example a bootstrap with 500 trials might take 200 500 times as long as a single fit In IRF program mode
46. ng guesses after using FIT mode before using BOT mode Thus it is advisable to generate point esti mates in FIT mode before using BOT mode After examining the results from FIT mode consider ad justing starting guesses and bounds on lines 14 17 and lines 19 20 Using better starting guesses and more restrictive bounds often results in time sav ings or better stability in fitting However bounds should be wide enough to encompass all plausible values 6 5 Common questions about data series 6 5 1 MISSING VALUES AND ZEROES Missing or zero data values are allowed in an ASPIC input file in some cases depending on the condi tioning mode and type of data series All possible cases are described in Table 3 along with the action taken by ASPIC A data line with a missing value or with f O does not contribute to the objective function however the information present on the line is used in the analysis and does influence the estimates Any negative data item in the input file is consid ered a missing value by ASPIC Thus a value can be set missing by inserting a minus sign in front of it and the value can be restored in a later analysis by removing the minus sign When a missing value appears in the ASPIC input file an estimate of the underlying value appears in the output file Missing values are always distinct from true zero values Zero should never be used to indicate a missing value and a negative number should never be use
47. of FTEST A small program named FTEST is provided to per form significance tests when comparing different ASPIC models of a stock The program is designed for comparing pairs of models that differ only in complexity number of parameters The TEST pro gram has a text user interface To run type ftest at a Windows command prompt and answer the pro gram s prompts 5 3 Overview of AGRAPH The Windows program AGRAPH is intended to pro vide quick good quality graphics of ASPIC and ASPICP results Preformatted time series plots of relative benchmarks and of observed and fitted abundance indices are provided Plots can be viewed on the screen sent to a Windows printer or saved as graphics files in several formats The AGRAPH program was not meant to meet all graphics needs of ASPIC users Instead it allows one to examine results quickly and to have graph ics suitable for assessment reports Operation of AGRAPH is similar to that of any Windows pro gram It can also be started from the command line For example to make graphs from results in file sword fit use the command agraph sword fit 6 ASPIC Input File Specification ASPIC reads its input from a single file containing control parameters and data The format of that file is described here 6 1 Generating a sample input file A new feature of ASPIC 5 0 is that a sample input file can be generated with the command aspic help The sample file is useful as a t
48. on of the catch equation is used and computation is slower than when conditioning on effort Table 1 Codes for the eight types of data series allowed in ASPIC Code Data type When measured CE Fishing effort rate catch weight Effort rate annual average Catch annual total CC CPUE weight based catch weight CPUE annual average Catch annual total BO Estimate of biomass Start of year B1 Estimate of biomass Annual average B2 Estimate of biomass End of year I0 Index of biomass Start of year I1 Index of biomass Annual average I2 Index of biomass End of year 4 6 Bootstrapped confidence intervals In BOT mode ASPIC uses bootstrapping to estimate bias corrected confidence intervals on many quan tities of interest In doing this estimated yields if conditioning on effort estimated efforts if condi tioning on yield and residuals from the original fit are saved The residuals are then increased by an adjustment factor Stine 1990 p 338 which is re ported in the output file Bootstrapped data sets are then constructed by combining each saved predicted yield Yi with a randomly chosen adjusted residual to arrive at a pseudo yield value Yi This procedure assumes that the statistical series weights w are correct The model is then refit using the pseudo yields in place of the original observed yields The process is repeated always using the original predicted val ues up to 1 000 times From the bootstrap
49. ow further data analysis The bio file is used by ASPICP described below for projections following a bootstrap run 11 The det file provides information on the individual bootstrap trials It is not used directly by any sup plied program but is provided for the user s conve nience Simple R or spreadsheet output NOTE The more complete rdat file described next is recom mended for R compatible output A file with exten sion prn containing a table of time series data can be output from an ASPIC run in BOT or FIT program mode It should be compatible with such programs as S Plus R SAS and spreadsheets To enable this file see specification of input file line 4 on p 15 More complete R compatible output More com plete output compatible with R is optionally avail able from ASPIC runs in FIT program mode This file extension rdat when read with the R func tion dget becomes an R data object of type list For example sword lt dget sword rdat This author is periodically improving the rdat out put for completeness For information on enabling this file see specification of input file line 4 on p 15 Summary output from simulations To aid in sim ulation studies a summary file sum file can be written in the current directory Enabling this file is described in the specification of input line 4 on page 15 The sum file can be read by S Plus or R with a statement like myres lt
50. pecified number of values have been read from a line the program does not read it further Thus the rest of the line may be used to contain comments Comments are included in the sample input files preceded by pound signs The pound signs are used to make the comments stand out to the eye and do not themselves denote comments to ASPIC After all data have been read from the file as determined by the number of years of data and number of data series any further contents of the file are ignored by ASPIC Thus additional comments may be appended to the file 6 3 The ASPIC input file line by line LINE 1 PROGRAM MODE This is a character string of length 3 with possible values FIT fitting mode BOT bootstrap mode or IRF iteratively reweighted fit mode For further explanation of program modes see 4 3 on page 8 For information on starting values in BOT mode see 6 4 on page 18 LINE 2 TITLE OF ANALYSIS This is a character string of length 110 characters or less The title is written to the main output file to identify the particular analysis The title will also appear on graphs made with AGRAPH and projec tions made with ASPICP C ksd Since the title almost always contains spaces it should be surrounded by quotation marks If the first character in the title is an asterisk the main output file will contain control codes to activate the lineprinter font on many older laser printers This provi
51. results bias corrected BC con fidence intervals can be computed by standard methods Efron and Gong 1983 The statistical lit erature recommends 1 000 bootstrap trials when computing 95 confidence intervals ASPIC com putes 80 confidence intervals and should require fewer trials The author recommends using at least 500 trials for bootstrap runs Some bootstrap trials may produce estimates e g of K or MSY outside the bounds set in the input file Such trials are discarded automatically and replaced by new trials 4 7 Input and output files As noted all ASPIC input and output files Table 2 on p 11 are in plain ASCII format Sample files are provided with the ASPIC distribution 10 4 7 1 INPUT FILE An ASPIC input file contains all data and settings required for a single ASPIC run It is recommended that when series of runs is made that each input file be given a distinct name This will ensure that the resulting output file names also are distinct The input file format is described in detail in 6 on page 13 The simplest way to generate an ASPIC input file is to run the command aspic help from the command line to generate the file sample inp That file can then be renamed and edited to the user s specifications It may be use ful to save an extra copy of the resulting file for use as a template 4 7 2 EDITING INPUT FILES To edit ASPIC input files and ASPICP control files use a plain text editor
52. rt When abundance indices present different pictures CPUE might instead be standardized with a model to re move effects of vessel type area gear season etc before fitting an assessment model The result ing index of yearly abundance can then be used as a SCC series with the total catch This provides quicker and more reliable estimation from ASPIC but more importantly it removes explainable varia tion from the data which would otherwise become noise Nonetheless using several abundance measures di rectly in ASPIC or any assessment model allows ex amining the departure of each series from model predictions 9 2 1 CATCHABILITY OVER TIME The user can estimate separate catchability coeffi cients for different periods of time This is accom plished in practice by putting the periods of time in separate data series each padded with zeroes or missing values as appropriate This procedure can be used to examine hypotheses about changing catchability with time perhaps as a result of chang ing fishing gear or changing environmental condi tions In interpreting such models there are several considerations One concern is estimating whether the improve ment in fit obtained from a more complex model is statistically significant An ASPIC model with time varying catchability can be tested against the base model i e the simpler model with constant catch ability with an F ratio test Here F is the F distri bution of
53. s of series type missing and zeroes See Table 3 Version 5 05 Added LRP and CV LRP to boot strap output for use with REPAST Version 5 06 Further increased length of file name variables Version 5 07 Improved output to PRN file Was not printing series gt 1 correctly Version 5 08 Further increased length of file name variables Version 5 09 Changed FIT to correct situation with MC search mode 2 repeated search Pre viously in initial fit of a bootstrap run MC re peats were fewer than for same data in in FIT program mode Version 5 10 Eliminated printing of AIC and F test for Fox model Added printing of revised bounds on MSY and K during generalized fit Changed logic to restore the user s bounds on MSY and K before bootstrapping a generalized fit Version 5 11 Revised ASPICP to version 3 16 which allows either BC or PC confidence inter vals and writes a PRB file Revised manual to explain those changes Version 5 12 Added option for RDAT file Version 5 13 Fixed bug in which some biomass index series were handled wrong and estima tion failed completely e For further changes please see change log sup plied in PDF format 11 Source Code The Fortran source code for this software uses cer tain proprietary routines from the book Numeri cal Recipes by Press et al and for that reason can not be freely distributed Numerical Recipes Soft ware has kindly granted their permission ID n
54. successful result is most often obtained when the catch and index data are scaled so that all qj lt 0 01 This of course does not apply to Bn data series for which by definition qi 1 See also 6 4 on page 18 LINE 18 FLAGS TO ESTIMATE OR FIX INDIVIDUAL PARAMETERS If line 12 specifies I data series the program reads I 3 integer values flags from this line The flags refer in order to B K MSY K and qj i 1 2 J Set the flag to 1 to estimate the cor responding parameter or 0 to keep the parameter constant at the starting guess No flag should be set to any value other than 0 or 1 Although q is not estimated for some series types I 3 flags are always required LINE 19 BOUNDS ON MSY This line contains maximum and minimum bounds on the estimate of MSY These two real numbers are used to limit the solution to reasonable values The user defines what is reasonable by setting these values If final estimates are at either constraint an error message is printed on screen and in the output file Bootstrap trials falling outside these bounds are discarded See also 6 4 on page 18 LINE 20 BOUNDS ON K This line contains maximum and minimum bounds on the estimate of K They are used in the same way as the bounds on MSY See also 6 4 on page 18 LINE 21 RANDOM NUMBER SEED Use a large 7 digit positive integer Different num bers result in different random number sequences Using the same see
55. t random number seeds If results cannot be dupli cated within a few percent usually less a fitting failure is indicated and such results should not be considered valid estimates The author will appreciate receiving reports of suc cessful or unsuccessful use of the features itemized above He will attempt to fix all bugs promptly 3 Installation and Interaction 3 1 Compatibility The ASPIC suite is compatible with personal com puters running Microsoft Windows 9x including Windows 95 98 and Me or Windows NT including Windows NT 4 0 2000 and XP lUse of tradenames does not imply endorsement by NMFS NOAA or the author ASPIC is written in standard Fortran 95 and is portable to other operating systems Please consult the author if you would like to use ASPIC under op erating systems other than Windows 3 2 Installation This version of ASPIC is available as a self installing executable file for Windows The installer performs the following tasks Installs binary files for ASPIC ASPICP FTEST and AGRAPH to a location specified by the user Installs this User s manual and a Quick Refer ence Card to the doc subdirectory of the instal lation location e Installs sample input and output files to the samples subdirectory of the installation loca tion e Adds the installation location to the user s PATH specification so that ASPIC and related pro grams can be executed from a command win dow open to any d
56. ter tells which type of value the number is For exam ple line 7 of the CTL file might read 1 456d3 Y to indicate that in the first projection year a yield of 1 456 units will be taken Thus lines ending in Y are used for making projections conditioned on quota TAC management measures As another example line 8 of the CTL file might read 0 85d0 F 21 to indicate that in the second projection year the fishing effort rate will be 85 of the rate in the fi nal year of the original data Thus lines ending in F are used for making projections based on propor tional reductions in fishing mortality rate This use of relative values allows F based projections to be made with reasonable confidence even when the es timated fishing mortality in absolute terms is quite imprecise F lines and Y lines can be mixed in the CTL file That might be done e g when yield in the first pro jection year is already known and management in subsequent years is to be by control of fishing ef fort 8 2 Sample ASPICP input file Case with YO2 Y01 F03 to F07 FCMSY test bio XX BC 1 0 6 1200 0 55 0 55 0 55 0 55 0 55 moan lt 9 Interpretation of ASPIC Results This section explains some features of ASPIC esti mates and reviews considerations important when using ASPIC Prager 1994 and Prager et al 1996 contain additional discussion 9 1 Precision of parameter estimates Production models tend to
57. tion Q minimized then is IN y Q Dd wi m 7 1 for residuals accumulated in yield EFT optimiza tion mode 6 3 or a similar expression for residu als in effort YLD optimization mode In equation 1 i indexes the data series j the year w is the series statistical weight Y is the observed yield or biomass index or estimates from series i in year j and Vij is the corresponding predicted value 4 5 1 PENALTY FOR INITIAL BIOMASS A penalty term can be added to the objective func tion to discourage estimates in which the first year s biomass B is greater than the carrying capacity K This penalty can affect the estimates of other pa rameters so when this term is used the results should be compared to those obtained by setting the term to zero The penalty term is described in more detail in Prager 1994 and its use is described in the section describing the input file format 4 5 2 CONDITIONING ON YIELD ASPIC can consider yield known exactly and accu mulate residuals in effort Yield is usually observed more precisely than effort or the abundance index and it is usually preferable on statistical grounds to compute residuals in the more imprecise quan tity Thus conditioning on yield is recommended for most analyses An additional advantage is that estimation of missing effort values is quite simple and is included in this version of ASPIC When con ditioning on yield an iterative soluti
58. to users of previous versions new users should also review it briefly for information on running ASPIC 5 0 pro ductively 2 1 Major changes Generalized production model Earlier versions of ASPIC could fit only the logistic form of the pro duction model Graham 1935 Schaefer 1954 1957 Pella 1967 Prager 1994 As well as that form ASPIC 5 0 can fit the generalized production model Pella and Tomlinson 1969 Fletcher 1978 in one of three ways by direct optimization by a grid of fits on the model shape or with fixed model shape to imple ment the Fox 1970 model or other pre determined shape Parameterization change The generalized model requires parameterization in terms of MSY and car rying capacity K rather than MSY and intrinsic rate of increase r This occurs because when the ex ponent 7 in the generalized model is in the region n lt 1 thenr As a result ASPIC 5 0 requires a starting guess for K not for r Starting biomass parameterization A parameter estimated by ASPIC is the biomass in the first year of the analysis In previous versions this was ex pressed both in the input file and in ASPIC reports as a ratio to the biomass providing MSY i e as B Bysy In version 5 0 it is expressed as a ratio to the carrying capacity i e as B K This change is required because in the generalized model Bysy is no longer a fixed proportion of K The change re duces correlation between estimates of the startin
59. ues t The first value on line 5 is an integer 0 lt n lt 1000 which specifies the number of bootstrap tri als A reasonable default is 500 Although this value is used only in BOT program mode it must be in cluded in all input files t The optional second value on line 5 is an inte ger 30 lt n2 lt 95 which specifies user confidence interval levels in BOT program mode ASPIC pro vides two sets of confidence intervals on estimates The first are 80 confidence intervals The second by default are 50 intervals but others can be spec ified by setting n2 When specifying nz gt 80 please set the number of bootstraps to 1000 LINE 6 MONTE CARLO SEARCHING This line contains two integers to control the op tional Monte Carlo MC search during fitting t The first value on line 6 may be 0 to disable the Monte Carlo search during fitting 1 to enable MC searching or 2 for repeated searching Turning MC on can help when a repeatable solution is otherwise difficult to find Unfortunately when strong local minima are present the MC search can cause more problems than it solves The author recommends leaving it off unless it is definitely needed t The second value on line 6 sets the initial num ber of Monte Carlo trials when repeated searches are enabled this number is reduced by the program in searches after the first Even if the first number on this line is 0 the second number is needed as a placeholder
60. um ber V95038 for the author to supply the source code to users upon specific request However any source code so supplied must not then be redis tributed to others The author also wishes to be aware of all distribu tion of the source code so that any useful modifica tions or error corrections can be made in the master copy of the software to benefit all users If you require a copy of the Fortran source code for this software please request it from the author In your letter or email please include the following 1 Your true name institutional affiliation phys ical address and email address or telephone number 2 Your agreement that you will not redistribute the source code to others 3 Your agreement that if you modify the source code you will not distribute any resulting pro gram or programs nor the modified source code beyond your immediate working group at your own location 4 Your agreement that if you modify the source code you will ensure that your users do not redistribute either the modified source code or any resulting program or programs 5 Your agreement that if you identify errors in the software you will contact the author promptly so that the errors can be fixed for all users 25 References Efron B E and G Gong 1983 A leisurely look at the bootstrap the jackknife and cross validation American Statistician 47 36 48 Fletcher R I 1978 On the restructuring of the P
61. under all combinations of data series type conditioning mode and model shape To that end a simple test has been done of every combination shown in Table 3 Still some cases occur infre quently in real data and so have not been tested repeatedly Users are urged to examine results criti cally when missing and zero values are used and to advise the author if any problems should arise 7 Advanced Options for ASPIC As of version 5 34 of ASPIC using an aspic ini file to control advanced options is no longer sup ported Several output files can be turned on and off with the verbosity setting line 4 of input file as described on page 15 User confidence intervals are specified on line 5 of the input file see page 15 8 ASPICP Input File Specification The control file for ASPICP is relatively short it should have the file extension CTL For the cor rect way to represent different data types in the file 20 see 6 2 A sample file is provided with the ASPIC distribution The following file format will work with all ASPICP versions An expanded input file format is available for use only with version 4 0 and up The new for mat allows added stochasticity in the projection and several additional ways of specifying management in the projection period Documentation is not yet complete Please see a version 4 0 CTL file for an example 8 1 Line by line LINE 1 PROJECTION TITLE This is a character string length lt 70
62. verview of ASPICP The auxiliary program ASPICP can be used following an ASPIC bootstrap run It provides estimated time trajectories of population biomass and fishing mor tality rate with bias corrected confidence intervals ASPICP is also used for making population projec tions beyond the observed data set When making projections the user can specify future harvests or effort levels and the program projects biomass and fishing mortality trajectories for up to 15 years past the original data Printer plots of the trajectories are also provided ASPICP reads information recorded in the BIO file of the corresponding bootstrap run The user con trols the program with a simple control file default ASPICP CTL Thus the first step in using ASPICP is to create a proper control file with a text editor De tails of the file contents are given in 8 and sample ASPICP input files are included in the ASPIC distri bution When starting ASPICP the control file name is given on the command line for example the command aspicp sword or equivalently aspicp sword ct starts ASPICP as described in control file sword ctl The output report from ASPICP is written to a PRJ file whose name is derived from the CTL file As of ASPICP 3 16 detailed projection results from each bootstrap trial are written to a PRB projection BIO file essentially an extension of the BIO file Table 2 that includes the projection years 5 2 Overview

ASPIC 5.0 User`s Manual - R

Contents

Download Pdf Manuals

Related Search

Related Contents