LFDA Practical Session

1
FMSP stock assessment tools Training Workshop
LFDA Practical Session
2
Session Overview

• How this session will run.
• What we will cover during the session.
• LFDA Tutorial Contents.
• LFDA example dataset.
• LFDA Tutorial.
• Summary.

3
LFDA Practical Session

• The LFDA practical session will last the rest of
  today.
• During the session we will look in detail at
• File formats used in LFDA, importing and
  inputting data into LFDA.
• Then using the pre-prepared LFDA tutorial to
  investigate some example length frequency data in
  the example dataset.
• You will be able to use your own data with LFDA
  later on in the course.

4
LFDA Example Dataset

• The example dataset we are using is a simulated
  dataset.
• Data will be loaded in from a text file called
  TUTOR.TXT.

5
LFDA Example Dataset

• Sample Timings
• - Regular sampling
• Length Frequency Classes
• Regular with a good range
• Number of individual fish measured
• Reasonably high samples

6
LFDA Tutorial Contents

• Introduction
• Loading and Inspecting Data
• Estimation of Non-Seasonal Growth Parameters
• Estimation of Seasonal Growth Parameters
• Estimation of the Total Mortality Rate Z

7
LFDA Tutorial Starting LFDA

• As with most Windows software you have a number
  of ways of starting LFDA
• Double-click on the LFDA icon.
• Start gt Programs gt MRAG Software gt LFDA5
• Open up Windows Explorer. Find the program
  LFDA5.EXE and double-click.

8
LFDA Tutorial Starting LFDA

• Start LFDA whichever way you want. Then close it
  down and try another way.
• You should get the screen below

9
LFDA Help File

• There is an extensive help file that details all
  the functionality behind LFDA, some theory behind
  the analysis and the tutorial that we will be
  running through.
• This can be accessed by clicking on HelpgtContents
  and Index on the menu bar.
• Clicking on the Tutorial section and then
  Loading and Inspecting Data, will allow us to
  start following the LFDA Tutorial.

10
Importing Data (1/2)

• Number of ways to load data.
• Commonest way is to load from an ASCII file.
• We will be loading the file TUTOR.TXT.

11
Importing Data (2/2)

• File Open
• Select .txt
• Select TUTOR from the list of files.
• The dataset will then be imported into LFDA. You
  can now save this as an LFDA file by using File
  Save As TUTOR.LF5.

12
Inspecting and Editing Data (1/4)

• You are viewing data in a sort of Spreadsheet
  Mode.
• The data as you see them on the screen cannot be
  edited to avoid accidental mistakes and
  overwriting of data.
• To edit the data. Go to the Edit menu and
  select Edit Mode. This will allow you to edit
  data already imported into LFDA. To stop any
  editing you need to repeat this action by
  selecting Edit Edit Mode again at the end.

13
Inspecting and Editing Data (2/4)

• When you are in Edit Mode you will see a number
  of menu choices appear that were previously not
  available. These are
• Add a distribution
• Erase a distribution
• Combine distributions
• Edit Sample Time.
• These allow you to build LFDA datasets from
  within the program by manually entering the data
  directly into the package.

14
Inspecting and Editing Data (3/4)

• A screen full of figures is not too clear though
  when you want to look at the length distributions
  over time.
• To view the data in a graphical format just
  select
• Data Plot Data.
• This is much clearer and makes tracing the
  progress of length distributions much easier.

15
Inspecting and Editing Data (4/4)
16
Estimation of Growth Parameters (1/10)

• We are going to assume that the data is
  non-seasonal and try to fit a standard von
  Bertalanffy growth curve to the data we have just
  imported.
• The basic idea behind estimating the growth
  parameters is to find the combination of
  parameters that maximizes a specified score
  function.
• A score function in LFDA is something that takes
  a model (e.g. von Bertalanffy) with its specific
  parameters (e.g. values for K and L8), then looks
  at your data and gives you a number which tells
  you how likely it is that your data comes from a
  stock with that growth function.

17
Estimation of Growth Parameters (2/10)

• Maximisation with score grids.
• If 2xy28 (x and y both whole numbers gt0), and
  our score is
• SCORE -1 difference between 28 and our
  estimate from the pair of parameters (x and y).
• Then we have a number of maxima where 2xy28,
  where score 0, e.g. Y14 and x7.

18
Estimation of Growth Parameters (3/10)

• Different score functions in LFDA
• Shepherds Length Composition Analysis
• ProjMat
• ELEFAN
• Details of each are to be found in the tutorial.
• Try fitting each of these now, following the
  tutorial

19
Estimation of Growth Parameters (4/10)

• Results Shepherds Length Composition Analysis
  with TUTOR.LF5

20
Estimation of Growth Parameters (5/10)

• Results PROJMAT Analysis with TUTOR.LF5

21
Estimation of Growth Parameters (6/10)

• Results ELEFAN Analysis with TUTOR.LF5

22
Estimation of Growth Parameters (7/10)

• We have looked at three different methods and
  have widely different results.
• Why is this? Do the models fit the data?

• Suspicion that data may be seasonal, slowing the
  growth at certain times of the year.
• Now try to fit a seasonal model.

23
Estimation of Growth Parameters (8/10)

• Two methods available for the analysis of
  seasonal data only (PROJMAT and ELEFAN).
• Choose the Hoenig model.
• What we are trying to do now is maximise the
  score function not just for two parameters (K and
  L8) but for four (K, L8, C (the strength of the
  seasonality) and Ts (the time the seasonal growth
  starts).

24
Estimation of Growth Parameters (9/10)
PROJMAT Hoenig Seasonal Growth Estimate
25
Estimation of Growth Parameters (10/10)
ELEFAN Hoenig Seasonal Growth Estimate
26

• Return to Theory presentation

27
Estimation of the Total Mortality Rate (1/2)

• Three methods available for estimating Z
• All based on non-seasonal models.
• If you have a strongly seasonal model then you
  should be extremely wary about using these
  models.
• Most seasonal patterns are OK as they closely
  resemble non-seasonal patterns.

28
Estimation of the Total Mortality Rate (2/2)

• So which set of parameter estimates do we use?
• We have three models with non-seasonal sets of
  widely different values of K and L8 and two
  others with closer fitting seasonal data.

• Look at the best fitting non-seasonal models for
  each method and see which looks the best when
  plotted with the data and against the seasonal
  model equivalent.

• ELEFAN looks pretty good. We will use this then
  with the values of K0.841 and L8 180.51.

29
Length Converted Catch Curve

• As you learnt in the theory session, this method
  takes the lengths and converts them into ages.
• For each distribution it then calculates the
  number of survivors to the next length class.
• Plotting the age against the natural log of the
  number of survivors gives us a slope equal to the
  total mortality rate Z for that distribution.
• Averaging overall the Zs calculated for the 10
  distributions will give us an average of Z for
  the year.

30
Length Converted Catch Curve (2/2)

• We only want the descending arm.
• We need to toggle out any points that are not
  part of the descending arm for each distribution.

• The toggling out of points here is quite a
  subjective process but your results should look
  something like this.

31
Beverton-Holt Z (1/2)

• The Beverton-Holt method relies on a simple
  algebraic relationship between the mean length in
  each sample, the length at first full
  exploitation, the von Bertalanffy growth
  parameters and the total mortality rate Z.
• We have not used the length at first capture here
  and have no information on it as this is a
  simulated data set.
• Therefore we will use the following parameters
  K0.84, L8 180.5 and Lc 20 (our first length
  class).

32
Beverton-Holt Z (2/2)

• As for the LCCC method we get an estimate for
  each distribution as shown below.
• The mean Z here is 0.991,

33
Powell-Wetherall (1/2)

• Similar to the Beverton-Holt method.
• Algebraic relationship for the right hand tail
  of a length frequency distribution to calculate
  Z.
• Missing points that have been excluded from the
  calculations are where the number of fish at that
  length class is zero.
• The results show not Z itself but Z / K and
  another estimate for L8.

34
Powell-Wetherall (2/2)
Results from the Powell-Wetherall estimation with
the TUTOR.LF5 dataset
35
LFDA Practical Session - Summary

• What have we covered?
• How to use LFDA.
• Estimation of Growth Parameters from LF data.
• Non-Seasonal
• Seasonal
• Estimation of Total Mortality Rate Z from LF
  data.