Photo: Colourbox

Open source fisheries management

Marine research
A statistical model originally developed to allow fishermen to study the work done by researchers, has now become one of the most highly acclaimed tools for estimating fish populations in Europe.

By Lene Reeh and Christian Blomgreen

Every year, fishery biologists from all parts of Europe and North America meet under the auspices of the International Council for the Exploration of the Sea (ICES) to determine the status of the fish populations and to shape their biological recommendations for the coming year. The advice from ICES is part of the EU negotiations and helps set quotas and other regulations for fishery in various member states.

By their very nature, fish in the sea are hard to count, so statistical models play a crucial role in estimating populations. At the same time, the processes governing the development in the populations are dynamic, because fish eat fish, for example, and because fishermen are constantly developing their equipment and methods. All in all, statisticians employed in fisheries management are working in a challenging field of non-linear relationships, diverse data and multiple parameters. Moreover, the consequences of the modelling results can prove serious for both fishermen and the fish populations of the future when the quotas are subsequently set. 

When fishery biologists meet every year across national borders in ICES to decide how much fishing for herring or cod the North Sea can withstand, they use a model developed in Denmark by two researchers from DTU Aqua. Senior Research Scientist Anders Nielsen and postdoc Casper W. Berg are the people who developed the model, which has been named ‘SAM’ (State-space Assessment Model). It is used today to estimate development in at least ten of the most economically important fisheries in Europe.

Transparent processes
The major difference between SAM and earlier models is that because it is an open source solution, everyone can see—down to the smallest detail—precisely how an estimate for a fish population calculated, because all the data and parameters used for measurement are transparent.

This level of transparency has proved a huge advantage for the researchers themselves, as Postdoc Casper W. Berg explains:

“You can ‘rewind’ all the results and see which data are used to reach a specific conclusion. As a result, it is easier for peers to review each other's work and ask questions about the parameters used. This, in turn, increases transparency and helps assure the quality of the estimates.”

Fishery powering development
SAM was developed for fishery analyses, but when Anders Nielsen is invited to places all over the world to teach the statistical methods that lay the foundations for the model, the courses attract a wide range of professionals, not just those who work with fish. Anders Nielsen is not surprised that marine research is inspiring bird researchers and other ‘landlubbers’:

“Fishery research is one of the areas that is powering development in statistical modelling in general. The reason for this is that the statistical models we work with to estimate fish populations are unique because they involve so many model parameters—hundreds or even thousands, in fact, which is many more than in most other branches of applied statistics. As a consequence, the statistical tools suitable for handling large models have largely been developed in the field of fishery modelling.”

State-space models

State-space models such as SAM describe situations where something ‘un-observable’—i.e. something that cannot be measured directly and completely—changes over time. This may, for example, be the number of mature fish in a population, or the actual fishing pressure (how intensively the population is being fished).

The models can estimate this ‘un-observable’ quantity through its relationship to something that is observable, such as the catches registered by fishermen, or the catches and measurements performed scientific surveys.

The use of state-space models is a modern statistical approach, utilized in many areas of applied statistics. A significant strength of these models is that they provide a more finely resolved estimation of the uncertainty of the estimates than more conventional models. With regard to a fishery model, this would have to do with the level of uncertainty concerning the volume of mature fish.

State-space models obtain these more precise estimates of uncertainty by separating uncertainties into two types of errors: ‘process errors’, which express how good your model is at describing reality; and ‘observation errors’, which have to do with how much noise—and thus how much uncertainty—is attached to your observations. This is why SAM can take into account factors such as observation errors on fish catches included in the calculations.