Ask An Expert
Taking Stock of New England Fish: Part 2
Mike Palmer is Research Fisheries Biologist in the Population Dynamics Branch of the Northeast Fisheries Science Center. This is the second post in TalkingFish.org’s “Taking Stock of New England Fish” series; the first post in the series can be found here.
TalkingFish.org: Okay, so it’s time to conduct a stock assessment. What are the basic steps you take to create a stock assessment from start to finish?
Mike Palmer: Stock assessments are time consuming. The first step, data preparation, is the most time intensive and perhaps the most important. This is where we evaluate the accuracy and sufficiency of the raw data. For example: many of the models need data on how many fish were caught at a particular age. Before we can estimate this, we first have to screen age and length information to make sure that the raw data that has been collected from port samplers does not contain any erroneous information — such as age 1 Atlantic cod that has been reported as 100 cm long, when a typical age 3 cod is usually around 50 cm. Once we’ve screened the raw data, we have to process it into the final data products that can be inputted into the assessment model. Typically, before data are used in an assessment a public data meeting is convened where all of the data that will be used in the assessment are presented and reviewed by a working group.
Next, we construct the assessment model. The models used for most groundfish assessments are relatively standard, but how the models are configured varies among individual assessments. The type and amount of data available dictates which models can be used. Once a model is selected, then other decisions must be made such as what year to start the model in, how many ages to include, how best to model the selectivity of fishing gear, and so on.
There are numerous ways to set up, or configure, a model. Models are basically mathematical tools that attempt to replicate the natural processes of birth, growth and death in fish populations. When we perform an assessment we start by updating the model used in the previous assessment, if there is one. The update simply adds additional years of data into the model. Based on the diagnostics of the model we evaluate whether alternate configurations of the same model type would improve the model performance or whether there are properties of a different model type that would better handle the available data.
After a suite of candidate models have been assembled, a public meeting is typically conducted where a working group will discuss the merits and uncertainties of the various models and select one or more model formulations as the ‘best’ model(s) and determine which formulations should be carried forward as ‘sensitivity’ runs. Sensitivity runs are model configurations that are not considered the ‘best’ model, but can provide an evaluation of how the results could be affected by alternate assumptions (like whether assuming that the mortality of discarded fish is 100% or some lower percentage will have perceptible impacts on the assessment results).
Assessment modeling is an incremental process; we are trying to constantly test and improve the model in order to provide the most realistic picture of the stock.
TF: We hear the word “models” used a lot in discussions of stock assessments. Can you describe a basic stock assessment model? If there are multiple models, how do you determine which is most appropriate for a particular assessment?
MP: Models are basically mathematical tools that attempt to replicate the natural processes of birth, growth and death in fish populations. So we look at the birth rate (annual recruitment), the growth rate of the fish in the population, and how many fish die each year through capture or natural mortality. If we know a lot about all three drivers, we can easily determine the total size of the fish population at any point in time. The growth rate can be estimated from information collected on fishery surveys or from port samplers but there is no easy way to estimate annual recruitment or total mortality. Instead we rely on stock assessment models to estimate these quantities and compute the total population size over time.
There are two types of models that are widely used to assess New England groundfish stocks: virtual population analysis (VPA) and statistical catch-at-age models. Both models have the same objective, to estimate total population size over time, but differ in their approaches. A VPA works backward in time attempting to reconstruct the historical fish numbers-at-age based on the assumed natural mortality and an estimate of fishing mortality. Conversely, statistical catch-at-age models, work forward by using the proportion of fish caught at each age to predict the relative abundance of fish in each of the age classes. These models often achieve similar answers, but the statistical catch-at-age models tend to have more flexibility and can better account for the uncertainty in the underlying data.
While the goal is for a model to produce a picture of the stock that is close to reality, model stability and good model diagnostics are also important. Model stability is critical to understanding how much the assumptions made are affecting the model results. Complicated models with a poor statistical fit raise serious concerns about how accurately they are interpreting the data, and whether the results are solid enough to use in providing management advice. Models with poor statistical fits are almost invariably rejected by peer-review panels. For example, if we have a lot of confidence in the landings that have been reported by industry, but the model we are trying to use doesn’t do a very good job estimating landings, it would suggest that there may be some aspect of the model which is misspecified. In other words, the model does not provide an accurate description of the data.