Emergency action for calculating stock size and reference points in case of massive data deficiency (BREFdata)

This EMFF-project “Emergency action for calculating stock size and reference points in case of massive data deficiency” has delivered the tasks specified in the project description.

This project was a timely response to the easily predictable gaps in data caused by the corona pandemic. The project has developed and implemented statistical rigorous methods for dealing with such data gaps. One consequence of gaps in the data series is that the uncertainties become larger --- even when the gaps are handled in the best way possible. The project has also developed methods to calculate reference points within the assessment model because that ensures that the (now larger) uncertainties are propagated correctly to all the results used by managers.

The SAM model, which is used for more than 30 analytical assessments in ICES, is now capable of handling data situations where individual observations are becoming more uncertain (due to partially impacted data collection) and situations where individual observations are completely missing. This works for all types of observations (even of biological parameters) and for partially observed observations it works when the larger uncertainty is a known quantity and when it is unknown and needs to be estimated by the model. The SAM model can now also estimate a range of different reference points (e.g. Fmsy, Fmax, Fspr, Fb0, Fext, and Flim) internally and these can be based on wide selection of stock-recruitment relationships (e.g. Ricker, Beverton-Holt, hockey, spline, Shepherd, power, and Hassel/Deriso). Internal estimation of these reference points ensures that they are calculated consistently with the assessment model. In combination, this gives the working groups the tools needed to correctly handle data gaps in assessment and management.

The project has thoroughly validated the implemented methods to study their performance and to strengthen confidence that they have been implemented correctly. The project has documented the developed methods in clear text, code examples, and mathematical formulas and further made them easily available in software (R package https://github.com/fishfollower/SAM) and online platform normally used to conduct such assessments (http://stockassessment.org). The project has presented these methods and helped apply them at all relevant expert and benchmark working groups.

In the initial stages of the project, it was found that data gaps were also frequently caused by other factors than a covid pandemic, so this project turned out to be even more useful and applicable than anticipated. The combination of the techniques from this project and a previous project on treating biological quantities (e.g. weights and maturities) as observations gave the added benefit of being able to predict biological quantities where they are missing. Finally, the ability to predict any observation opens a lot of possibilities of using prediction-validation or cross-validation to compare the performance of models.