CEDA Practical Session

1
CEDA Practical Session
FMSP Stock Assessment Tools Training
Workshop Mangalore College of Fisheries 20th
-24th September 2004
2
CEDA Practical Session

• During the session we will look in detail at
• File formats used in CEDA, importing and
  inputting data into CEDA.
• Then using the pre-prepared CEDA tutorial to
  investigate some example catch and effort data
  analysis with the pre-prepared example datasets.
• You will be able to use your own catch and effort
  data with CEDA later on in the course.

3
CEDA Example Datasets

• During the course of the CEDA tutorial we will
  use two different datasets to illustrate
  different fisheries and how CEDA can be used to
  analyse them.
• The first (SQUID.CD3) is a dataset from an annual
  squid fishery (Illex argentius, 1989). We have
  one seasons data showing the gradual removal of
  catch from the population.
• The second dataset (XTUNA.CD3) is a long time
  series of catch and effort data for a Yellowfin
  Tuna (Thunnus abacares) fishery.

4
CEDA Tutorial Contents

• Introduction
• Loading data into CEDA
• Analysis of tutorial dataset 1 SQUID.CD3
• Analysis of tutorial dataset 2 XTUNA.CD3
• Making projections
• see help files Tutorial section

5
Starting CEDA

• As with most Windows software you have a number
  of ways of starting CEDA
• Double-click on the CEDA icon.
• Start gt Programs gt MRAG Ltd gt CEDA3
• Open up Windows Explorer. Find the program
  CEDA3.EXE and double-click.

6
Loading and Inspecting Data

• There are three ways of loading and entering into
  CEDA.
• 1. Import of data from a text file
• 2. Loading a previously created CEDA file .CD1
• 3. Manual data entry into CEDA.

1 NB CEDA version 3.0 will not import files from
CEDA v1.0 only 2.0 and above
7
Manual Data Entry

• It is possible to create a CEDA .CD3 file and
  manually enter data from scratch (try on Friday?)
• However, we will be using pre-generated datasets
  during this tutorial session for CEDA ..

8
Loading the Squid Dataset

• The squid dataset is initially in a text file
  called SQUID.TXT. This data is real catch and
  effort data for the Falkland Island squid fishery
  taken from Rosenberg et al (1990).
• During the loading we will
• Load the file
• Allocate the columns
• Check computed columns
• Save the file as a CEDA .CD3 file.

9
Analysing the Squid Data (1/17)

• The squid dataset is now loaded

10
CEDA Data Requirements (1/5)

• Total Catch (in weight) - Total catch in weight
  taken in the fishery by all gears during the time
  period.
• Total Catch (in numbers) - Total catch in numbers
  taken in the fishery by all gears during the time
  period.
• Effort - In a fishery where only a single gear is
  employed, the effort column should include all
  effort for the fishery. In the case where
  several different gears are used, the effort and
  partial catch (see below) for the type of gear
  that most closely match the assumptions of the
  model or that are considered to be the most
  reliable should be used.

11
CEDA Data Requirements (2/5)

• Mean Weight - The mean weight of individual fish
  in the time period. This is used to convert
  catch in weight to numbers for various models.
• (Partial) Catch in weight The effort column in
  CEDA may contain effort for a single gear type
  rather than the whole fishery. In this case, it
  is necessary to specify a partial catch column
  i.e. a catch corresponding to the specified
  effort.
• (Partial) Catch in numbers -The catch
  corresponding to the effort column where the
  specified effort is not for all gear types.
• Recruitment Index - This is an index whose value
  is proportional to the number of new recruits.
  The recruitment index model requires such an
  index to be specified for each time period.

12
CEDA Data Requirements (3/5)

• Abundance Index (in Weight Only) - As an
  alternative to supplying a catch and partial
  effort series, CEDA allows you to specify an
  Abundance Index. This is an index whose value is
  proportional to the biomass of the population.
  CEDA then converts this internally to catch and
  effort columns.
• Variance of Abundance Index - If estimates of
  variances of each abundance index value are
  available, these can be entered into CEDA and
  used to weight the fitting of models i.e.
  indices with smaller variances are given greater
  weight than those with bigger variances.

13
Data Requirements (4/5)

• Timing - The Time column on the left-hand side is
  very important. In order to cater for the
  various temporal data that you may wish to
  analyse (e.g. monthly, weekly, annual), CEDA
  does not fix the time units for you. All
  datasets being imported in must have associated
  time units, although these can be of your own
  choice. The only constraint is that time periods
  must be denoted by integers and these must be
  consecutive.

14
CEDA Data Requirements (5/5)
15
(No Transcript)
16
Analysing the Squid Data (2/17)

• Now back to our squid data set

17
Analysing the Squid Data (3/17)
18
Analysing the Squid Data (4/17)
19
Analysing the Squid Data (5/17)

• The most suitable method of analysis for this
  type of data from a squid fishery is a depletion
  model with no recruitment. This method of
  analysis is available in CEDA.
• You are now ready to analyse the data.

20
Analysing the Squid Data (6/17)

• Select Fit New Fit from the CEDA menu.
• Select the No Recruitment model.
• Mortality Rate 0.05
• Error model Least Squares (unweighted)
• What outputs do we get?

21
Analysing the Squid Data (7/17)

• We now have a fit.
• Next question to ask ourselves is Is this a good
  fit?
• We need to look at the residuals relating to this
  fit.
• Under the Graph menu you will see two residual
  plots for the residual catches vs time and
  expected catches.

22
Analysing the Squid Data (8/17)
Look at the two residual plots. What is wrong
with them?
Triangular in shape, not the best result for a
residual plot.
23
gtgt Saving the Fit ltlt

• Before carrying on with the analysis you should
  save the work you have done so far. This is done
  by adding the fit to the Fit Manager. Each fit
  is logged separately within the CEDA file and
  saved automatically.
• Select Fit Fit Manager from the menu. This
  brings up the fit manager, a tool which allows
  you to save, delete and reload different models
  you have run on your dataset. We are going to add
  the current model to the fit manager. Press the
  Add Current Fit button. You are then prompted to
  enter a description to identify your work.

24
Analysing the Squid Data (9/17)

• We can now take a look at fitting a different
  model to the data, or excluding some of the data
  points.
• In the case of the squid data we know from
  experience that the first few points of catch and
  effort data collected are not representative of
  the fishery as a whole as the squid are still
  migrating into the area. Therefore we have a
  valid reason to omit these three points from the
  analysis.
• Remember you should not exclude any data points
  from your analysis unless you have a very good
  reason to!

25
Analysing the Squid Data (10/17)

• Re-run the fit without these three points.
• Check the fit against the residuals.
• Are the residual plots better than the previous
  fit?
• Save it and then compare the fit against the
  first fit.

26
Analysing the Squid Data (11/17)

• Comparison of the two fits we have done

27
Analysing the Squid Data (12/17)

• We have now investigated the Least Squares
  (Unweighted) Model.
• Now try doing the same process with the other two
  models, Log Transform and Gamma.
• Run through the same process. Running the model
  at M0.05. Check the fit, the residuals and
  outputs. Check for outliers. Save Log Transform
  fit as Illex 3, and Gamma fit as Illex 4.

28
Analysing the Squid Data (13/17)

• Comparison of the fits we have now done

29
Analysing the Squid Data (14/17)

• We have throughout the analysis so far used
  M0.05. However, what if this is not right?
• You should always test for the sensitivity of a
  model to key input parameters such as M. Try
  running the model with different (but likely)
  values of M in the range (use the Gamma Error
  model)
• 0.01 0.10
• logging the fits as you go (call e.g. M0.01), so
  you can compare them to each other.
• Investigate the changes to N1 and q.

30
Analysing the Squid Data (15/17)

• Example results of sensitivity analysis for M
  (Gamma Error)

31
Analysing the Squid Data (15/17)

• Example results of sensitivity analysis for
  inclusion / exclusion of outliers (Gamma Error
  Model)

32
Analysing the Squid Data (16/17)

• All the analysis so far has given us point
  estimates.
• These are dangerous to use as we all know
  fisheries are highly variable.
• To include some of this variability in our
  analysis we need to add confidence intervals to
  our point estimates.
• This can be done to any fit by selecting Fit
  Generate Confidence Intervals or pressing the
  button on the Parameter Estimates dialog box.
  Try this for a few fits.

33
Analysing the Squid Data (17/17)

• You will see a new box added to the Parameter
  Estimates dialog box.
• This shows the 95 CI of N1 and q.
• You can also view the results of the
  bootstrapping by selecting the graphs from the
  Graphs menu.
• Again, you should now recalculate the CIs for
  sensitivity of different values of M.

34
Summary Squid Analysis (1/2)

• The no recruitment model with gamma error gives
  both satisfactory and useful fits to these data.
  
• The diagnostic plots (residual/percentile plots)
  do not highlight any major problems and the
  confidence intervals are narrow.
• The remaining possible outliers suggests that
  there may still be some problems with the model
  or the data, but the lack of sensitivity of the
  parameter estimates to these data points
  indicates the utility of the model.

35
Summary Squid Analysis (2/2)

• One important conclusion is that the parameter
  estimates are very sensitive to the value used
  for the natural mortality rate M, and that the
  data yield very little information about M.
• Given this sensitivity, a sensible course for
  management might be to consider how to improve
  the estimate of M.

36
(No Transcript)
37
Analysis of Tuna Data (1/12)

• We now move on to the more complex tuna dataset.
• The XTUNA.CD3 dataset only comprises four
  columns, for time (year), total catch (wt),
  effort and catch (wt).
• With these data columns only the three production
  models, Fox, Schaefer and Pella-Tomlinson are
  allowed.
• We will start off trying to fit a Pella-Tomlinson
  model to the data.

38
Analysis of Tuna Data (2/12)
We need to set the initial parameters for the
model Initial Proportion The degree of
exploitation of the stock before the start of the
dataset. Here the default is 1. Z shape
parameter Shape of the curve, again assume the
default of 1. Time This is the time lag between
a biomass creating recruitment and this
recruitment becoming evident in the population.
Enter 0 for preliminary examination.
39
Analysis of Tuna Data (3/12)

• As with the squid data we are looking at which
  error model best fits the data.
• Check the residuals for the fit for Least Squares
  and then repeat for the Log Transform and Gamma
  error models, using the same parameters. Remember
  to save each fit to allow later comparison.
• (call Tuna1, Tuna2 and Tuna3 respectively)
• Gamma model will need to reduce parameter
  tolerance to 0.001.

40
Analysis of Tuna Data (4/12)

• Residuals show better fit for the log transform
  and gamma models than for least squares. Which
  is best?
• Still have two outliers for 1951 and 1953. What
  do we do with these?

41
Analysis of Tuna Data (5/12)

• Outliers will occur at 95 confidence levels 5
  of the time.
• These two outliers are very close to 0 and 1
  though, suggesting a problem with the data or the
  model.
• We have no good reason to exclude them, so we
  must leave them in the analysis.
• Neither error model seems to fit the data well.

42
Analysis of Tuna Data (6/12)

• We need to test the sensitivity of the model to
  the different outliers.
• Run both error models through excluding 1951,
  1953 and both years. Look for the differences in
  K, q and r and if excluding the outliers makes a
  significant difference to the fit (try Gamma and
  Log Transform models).

43
Analysis of Tuna Data (7/12)

• Sensitivity of K to various fits

44
Analysis of Tuna Data (8/12)

• Then repeat the analysis for sensitivity to the
  initial parameters (time lag, initial proportion
  and Z the shape parameter of the Pella-Tomlinson
  production model using just the Gamma model.
• Examine sensitivity to Initial Proportion over
  range 0.8 1.0.
• Examine sensitivity to the Time Lag over the
  range 0 4
• Examine sensitivity to the Z-parameter over the
  range 0.5 2.0

45
Analysis of Tuna Data (9/12)

• Sensitivity to Initial Proportion over range 0.8
  1.0

46
Analysis of Tuna Data (10/12)

• Sensitivity to Time Lag (in range 0 4)

47
Analysis of Tuna Data (11/12)

• Sensitivity to Z parameter (over range 0.5 2.0)

48
Analysis of Tuna Data (12/12)

• Confidence Intervals We now need to generate
  confidence intervals for the models, looking at
  the ranges for K, MSY, q.
• You should consider the sensitivity of the model
  to the input parameters (e.g. Z, Time Lag,
  Initial Proportion) in terms of these confidence
  intervals.

49
Conclusions of the Tuna Data Analysis

• Substantial discrepancies exist between the model
  and the data.
• The residuals do not show a particularly good fit
  with runs of values above and below the expected
  catch.
• Two outliers exist that cannot be explained.
• Wide confidence intervals are shown for MSY.
• Results are sensitive to input parameters.

50
Ways Forward with the Tuna Data Analysis

• The lack of contrast in the data means that
  good parameter estimates would always be
  difficult to obtain.
• Methods of improving the contrast in future data
  from the fishery might also be considered.
• Often your data will not show clear results. As
  frustrating as it is, this is the real world.

51
(No Transcript)
52
Projection Scenarios (1/5)

• CEDA can also be used to make predictions into
  the future given a set of data and a good fitting
  model.
• We will now build a scenario based on future
  effort or catch for the tuna fishery and see the
  effects of this on the population size.

53
Projection Scenarios (2/5)

• Select Projections Set up scenarios. We will
  enter our effort scenario as follows
• However, try different scenarios yourselves.
• Can also set Confidence Intervals about your
  projections.

54
Projection Scenarios (3/5)

• This is projecting a catch of 30,000 tonnes in
  1968, 25,000 in 1969 and 20,000 tonnes between
  1970 - 1980. It shows the stock recovering.

55
Projection Scenarios (4/5)

• Select Projections Set up scenarios. We will
  enter our effort scenario as follows
• However, try different scenarios yourselves.
• Can also set Confidence Intervals about your
  projections.

56
Projection Scenarios (5/5)

• This is projecting a constant catch of 160,000
  tonnes the stock collapses.

57
Comparison of management options (TACs)

• MSY catch 161 Kt, but not sustainable at
  current biomass
• see Section 8.3, p142 for further explanation

58
Projections with confidence intervals

• 50 confidence interval
• for a 140 Kt TAC,
• projected from 1968
• There is less than a 25
• chance that stock size
• should fall below 50 Kt
• after returning to the
• MSY levels
• But there is still some
• chance that the stock
• will crash in future .
• - try projecting with a 95 confidence interval

Section 8.3, p142-3
59
Summary

• CEDA uses Catch (Numbers or weight) and Abundance
  Data
• Six models to choose from. We have looked at the
  No Recruitment Model (Catch in numbers) and the
  Pella-Tomlinson model (Catch in Biomass).
• The Recruitment model gave estimates of N1, q and
  Final population size.
• The Pella-Tomlinson gave estimates of k, r, q and
  MSY
• It is important to choose the correct Error Model
  (Use Residual plots to do this).
• Varying your input parameters may influence the
  estimated parameters, so should perform
  sensitivity analysis on these.
• Results should always be expressed in terms of
  ranges use confidence intervals.