Ecosim, Group info
37. Stock reduction analysis
A very useful technique for using long term data in stock assessment is Kimura’s “stock reduc- tion analysis”. In this technique, historical catches are treated as fixed, known quantities, and are subtracted from simulated stock size over time so as to aid in estimating how large (and/
or productive) the stock must have been in order to have sustained those catches and to have been reduced by some estimated fraction from its historical level. In some assessment liter- ature, treating catches as fixed knowns is also called “conditioning on catch”. A drawback of treating catches as fixed values is that catches in fact arise from the interaction of fishing effort and abundance, and ignoring this dynamic interaction amounts to treating the catches as purely depensatory impacts on stock size (when simulated stock size declines, the fixed catches can cause progressively larger calculated fishing mortality rates F, leading to a depen- satory spiral of rapid collapse in the simulated stock, which may or may not have been possible in the real system).
When creating historical reference CSV files for model testing (see Import time series), all or part of a catch time series for any group(s) can be treated as a forcing input (with simulated F calculated each year as (input catch)/(simulated stock size) ) by setting its data type to -6 (rather than the usual 6 for fitting catch data). Note that the catch time series for a group can be entered in two columns, with one column set to data type 6 and one to data type -6, where catches for years to be treated as forcing are placed in the -6 column and catches for years when catch is to be predicted from effort or assessment Fs placed in the 6 column. Most often, this splitting of catches into two columns should be used in cases where there are no inde- pendent assessments of F for some early years.
The Monte Carlo simulation interface in Ecosim can be used to search for Ecopath biomasses needed to have sustained historical catches. We cannot search for such initial biomass values by simple nonlinear search methods, due to the biomass constraints implied by Ecopath mass balance. The Monte carlo simulation interface can do a large number of simulations with ran- domly varying trial values of Ecopath biomasses, and can retain trial values that result in improved model fit; such a search or fitting procedure is known as a “Matyas search”.
Tutorial: Uncertainty in time series data
The Multi-sim tutorial is designed to address uncertainty in time series data, be they envi- ronmental forcing functions, fishing effort, biomass series or other, see the EwE User Guide guidelines for Multi-sim.. Use the Anchovy Bay true.ewemdb database (download). Open the database, and go to Ecosim > Tools > Multi-sim. The EwE Multi-sim can read in time series files with varying forcing functions. You can get the format for the CSV file to use if you click Example on the Multi-sim interface. For your own use, use that file derived from your own model. Here, we will instead download a zipped file that holds a folder with 20 CSV files. Each of these 20 have a time series of environmental productivity that is derived as auto-correlated values based on the original true time series values.
Download the zipped multisim csv.folder, unzip and place it somewhere (where you can find it again).
Figure 1. Top part of one of the Anchovy Bay time series files used for the Multi-sim analysis in this tutorial.
Each CSV file looks like this, only difference is that the values in the Fitting column varies between files, but any of the time series in file can be varied, if so desired.
Now check that the Fitting time series is applied to primary production. Go Ecosim > Input >
Forcing function > Apply forcing (producer) and check that F2 is applied to primary producers (or apply it, if it is not).
Here’s a sample showing how the “Fitting” time series look for four of the CSV files,
Figure 2. Four series of auto-correlated time series for environmental productivity. The original time series is indicated in red on each plot.
On the Multi-sim interface, Input > Source folder, select Choose, and browse to find the folder with the CSV files that you just downloaded. Select the 20 CSV files, by clicking All in the right column.
On Output, select where you want Multi-sim to write output. Select which indicators you want to get output of, e.g., Biomass. Next click Run. Multi-sim will now run Ecosim 20 times, each time with a different time series for the environmental forcing function read in. If saving bio- mass as in this example, it will make a folder for each of the 20 runs, and save two files in each, biomass_annual.csv and biomass_monthly.csv. This structure lends itself easily for analysis in R.
Compare the time series in some of the output files, play!
There’s a log file saved that shows what Multi-sim has done, in Figure 3 is an example.
Figure 3. Log-file from Anchovy Bay Multi-sim run
Tutorial: Fitting time series with true values
In a previous tutorial we fitted the Anchovy Bay model to a simple time series file, mainly in order to explore the time series fitting procedures of Ecosim in a simple manner. Here, we will expand on this, notably by evaluate fitting when considering fisheries, food web, and environ- mental conditions, i.e. we add an environmental forcing function.
The present tutorial is a test to see how well Ecosim can fit a model with known parameters.
The time series ‘data’ for this tutorial were thus derived from a model run with known primary production forcing and with known vulnerability multipliers. Can we retrieve those values?
Open the model Anchovy Bay true.ewemdb (download); then load the anch bay scenario, and then the anchovybay true time series, (which is also available in the anchovybay true.csv time series file).
Reset the vulnerability multipliers to the default: Ecosim > Input > Vulnerabilities. Click the upper left cell in the spreadsheet (above 1 and to the left of Prey\predator), to select the entire sheet, then enter the value 2 in Set: at the upper right, and click Apply to the right.
Run the model, Ecosim > Output > Run Ecosim > Run button. Check the output, notably on the Ecosim Group plots. As an example of what to look for, examine the cod screen. The model shows bigger decline in biomass than the data. What does that tell us about the vulnerability multiplier for cod?
Check the Ecosim > Input > Forcing function form where you should find two forcing functions, 1: True PP, (which we will use later for comparisons), and 2: Fitting, (which we will use for fitting). Check the Ecosim > Input > Forcing function > Apply FF (producer) form. Here there should be an F2 in the (single not blocked entry field) for Phytoplankton (if not, then click the empty field and select the FF. The two forcing functions are, by the way, included in the time series file, and when/if you read it in, you’ll have to specify that you do not want to read in each of these as monthly values.
Now go Ecosim > Tools > Fit to time series. Try running a number of different fits; first with vulnerability search only. E.g., search groups with time series, and search for vulnerability multipliers for the seven groups with time series. Note what the ‘base’ SS is, and what you get with seven estimated parameters, probably a substantial reduction in SS. Try also to estimate fewer vulnerability multipliers. Also search for most sensitive parameters, and try with differ- ent number of parameters. Which groups are most/least sensitive? Are there any implications to be deducted from this? Notably consider if we are likely to have or get time series for the more sensitive groups.
For each run it is always a good practice to check the vulnerability form to see the estimated multipliers.
Also compare vulnerability multiplier fitting with searches by predator and by predator/prey combinations. As a rule, we find that searches by predator are most efficient – and easier to explain.
Reset the vulnerability multipliers. Now search for a primary production anomaly; check the Anomaly search on the Fit to time series form, and uncheck the Vulnerability search, so only search for an environmental signal. Before each run go to Ecosim, Input data, Forcing func- tions, and reset the shape of the Fitting forcing function. Try with different number of spline points, e.g., 2, 3, 4, 6, 10, and finally 0 (i.e. annual PP). Do you see clear environmental signals?
Evaluate not just SS, but also if the anomalies are plausible (there can be a tendency to cut off or wildly increase PP at the beginning or end of time series as this often can be done without time series ‘penalty’). Based on this, select one or several number of spline points to test for the combined Vulnerability and Anomaly search – try such runs.
When you’ve done runs to fit, copy the results from the table on the Fit to time series form to a spreadsheet (so you don’t lose them). Evaluate AICc values, and consider implications for model selection.
• Consider how well the fitting routine is able to find the PP anomalies and vulnerabilities.
• Discuss the findings and their implications.
Optional
There is an additional time series in the model database, “anchovybay true partial”.
It is similar to the previous, but omits a number of early time series values that help identify the cycling pattern. Try this time series to consider the effect of hav- ing time series with contrast.
Vulnerability multipliers for the Anchovy Bay model
Your model version may be a bit changed from what’s below, but should be close
1 Whales 1.14
2 Seals 2.38
3 Cod 1.28
4 Whiting 6.17 5 Mackerel juv 1.10 6 Mackerel ad 1.60 7 Anchovy 1.21 8 Shrimp >1000 9 Benthos 3.22 10 Zooplankton 2.26
Tutorial: Anchovy Bay Ecosampler
Ecosampler is a routine that creates alternative Ecopath models, and then uses each of these models for analysis based on the core EwE routines (i.e. Ecopath, Ecosim, and Ecospace).
Ecosampler is included in EwE Vers. 6.6., and is available from Menu > Tools > Ecosampler in the Navigator.
The Ecosampler guide is in the EwE User Guide.
Open the Ecosampler form and check the Record button (making it red for recording). Go to Ecosim > Tools > Monte Carlo simulation, and run. Notice in the Ecosampler interface that it saves each balanced Ecopath model.
Turn of the Ecosampler recording. Follow the directions in the Ecosampler User Guide.
Ecosampler can run for the basic Ecopath run, Ecosim run and Ecospace run, as well as run loaded plug-ins. But it cannot (yet) run the various modules built into EwE, e.g., policy search, fit to time series. For now, you can load a sample at the time in the Ecosampler routine, then run the various routines in EwE, one after the other. Or, design a plug-in that will do that!
Tutorial: Maximum sustainable yield (MSY)
Learning Objectives
This exercise is designed to,
• Obtain experience with evaluation of single-species vs. ecosystem-level MSY