Aquaculture Europe 2023

September 18 - 21, 2023


Add To Calendar 20/09/2023 14:45:0020/09/2023 15:00:00Europe/ViennaAquaculture Europe 2023MAKING RESEARCH ACCESSIBLE USING A DIGITAL TOOL: COMMUNICATING FUTURE SCENARIO PREDICTIONS FROM RECURRENT NEURAL NETWORKS USING A LIGHTWEIGHT WEB APPLICATIONStolz 2The European Aquaculture Societywebmaster@aquaeas.orgfalseDD/MM/YYYYaaVZHLXMfzTRLzDrHmAi181982


L. Veylit*, A. Strand, R. Tiller, T.L. Oftebro, I.H. Ahlquist, A. Misund and T. Thorvaldsen


 SINTEF Ocean, Postboks 4762 Torgarden, NO-7465 Trondheim, Norway 



 Norway is  a global leader in salmon farming , with the aquaculture sector expecting to grow in the future to reach  consumer demand. A dvocates  argue farm raised salmon have a low carbon footprint, and aquaculture can therefore play a pivotal role in reducing the carbon footprint of food production. The detractors, on the other hand, argue aquaculture poses an environmental threat  as escaped salmon  may  hybridize with natural populations,  and  high population densities in pens lead to an increase in local pollution and sea lice occurrence . Furthermore, d elousing of the fish is a challenge for fish welfare, the opponents argue. New production systems  for salmon fish farming  (i.e.,  land-based, floating closed, semi-closed, and open ocean aquaculture systems )  aim to counter these challenges, as well as utilise new areas for production . More knowledge is needed regarding how these new production systems  should be regulated, and how they  will influence the  public  perception of  the salmon farming industry.

In this study ,  variables relevant for exploring  key future scenarios  are  co-developed  during participatory stakeholder workshops using Fuzzy Cognitive Mapping (FCM) , a soft computing framework.  Industry stakeholders decide on relevant variables to include in models , how they influence each other, and  discuss feedback mechanisms in participatory workshops .  Data collected  during workshops  are then  used as the basis for the creation of FCM models (Kosko 1986) , which combine fuzzy logic and  recurrent neural networks,  to produce predictions of  changes to the perception of salmon farming in response to potential future scenarios.  The  results from models (i.e.,  values that represent the strength and direction of change to the  perception of salmon farming in response to different scenarios) were then made interactable and accessible through their integration into a lightweight web application as a  first-generation policy action tool .

 The application is developed to make the results from th is study accessible to diverse stakeholder groups . Th us,

 a concise description of how the model functions and how to interpret results from the scenarios is included in the app. In addition, information on the relevant variables defined by stakeholders  and the fuzzy cognitive map describing the relationship between variables stakeholders defined in workshops  is  provided to users for easier interpretation of results.



Stakeholders included in the participatory workshops included industry representatives working with new aquaculture systems in Norway, researchers, and representatives from  trade organisations.  The variables  that stakeholders identified as relevant to Norwegian salmon farming  and their interactions were visualized in  the freeware  Mental  Modeler throughout workshops to facilitate the co-production of a map of the system. Connections between variables were quantified on a continuous scale between -1 and +1, where a negative value indicated that if one variable increased, the variable on the receiving end of the “negative” connection  would decrease . -1 indicated a  strong decrease in the connected variable ( and it follows  +1 indicated an increase in one variable would also result in the  direct,  strong increase in the connected variable  and  values near zero describe weak relationships between variables). These values served as the basis of a semi-quantitative model we used to describe our system and the basis of the  creation of the  projected future outcomes that were possible to interact with in the  web application. Variables and their relationships were calibrated and validated in subsequent workshops.

 Models were developed using the FCM package (Dikopoulou and Papageorgiou 2017 ) in R.  Models were fit with the rescale inference rule which is preferred  where there is not previous information about a concept-state, and  we did not have information on the initial state of the system (Papageorgiou 2011) .  A sigmoidal transformation function was applied as sigmoidal  (continuous)  FCMs are  recommended for  quantitative and qualitative scenarios with complex feedback structures as we found in our system (Tsadiras 2008) .  The weight matrix consisted of the values obtained from the participatory workshops.  The application was built using Shiny (Chang et al. 2023), a n open-source framework for building interactive web applications.  One interactive element of the  application is in the form of a drop-down menu where users c an select what variables to have a positive value in the activation vector. Visualizations and the table of values predicted for each variable in the application show values reached once the model has reached convergence and the system has reached an equilibrium point .  The model outputs shown in visualisations from the user’s chosen scenario are based on the stakeholders’ perception of the system if there is a change to the current system. Feedback from stakeholders have been and will be implemented to make the app more user friendly. Specifically, s imilar variables were grouped together to simplify the interpretation of the results and the app is being developed in both Norwegian and English to match the language needs of different users (from local industry actors to academics) . Stakeholders will be invited to give feedback on the application to ensure a user-friendly interface  is created  as development of the app continues.


Results & Discussion

 Following the axiom “all models are wrong, but some are useful,” the outputs of the workshops and FCM models  in the web application  will not be used  as predictions of how perceptions will change as Norwegian salmon farming  changes. Indeed, t he data collected from stakeholders is limited due to the small number of actors it is possible to include in workshops .  Outputs from the models  may be useful  1)  in the context of engaging non-experts in research  and 2)  for serving as conversational focal points for ideating scenarios and informing decision making.  The FCM developed in workshops also serves  an important purpose  as a  simplification of the challenges faced by the Norwegian salmon farming  industry,  which allows for the visualisation of important and sometimes surprising relationships between identified variables. The app’s value comes from its ability to bridge gaps between researchers and stakeholders and between stakeholders themselves who may have differing opinions on  how  different variables will affect other variable under future scenarios.  Indeed, the app may be used  as a policy action tool, and to improve the communication between resource managers, regulators, policy makers, and actors from industry who make critical choices and researchers beyond a traditional scientific article or report.  The conversations that result from interaction with the digital tool can thus be used  to inform decision makers on how to develop Norwegian salmon farming in coming decades under different scenarios.


Kosko , Bart. "Fuzzy cognitive maps." International journal of man-machine studies 24.1 (1986): 65-75.

Dikopoulou , Zoumpoulia , and Elpiniki Papageorgiou. "Inference of fuzzy cognitive maps (FCMs)." The Comprehensive R Archive Network (CRAN) (2017).

Papageorgiou , Elpiniki I. "A new methodology for decisions in medical informatics using fuzzy cognitive maps based on fuzzy rule-extraction techniques." Applied Soft Computing 11.1 (2011): 500-513.

Tsadiras, A. K. (2008). Comparing the inference capabilities of binary, trivalent and sigmoid fuzzy cognitive maps. Information Sciences, 178(20), 3880-3894

 Chang W, Cheng J, Allaire J, Sievert C, Schloerke B, Xie Y, Allen J, McPherson J, Dipert A, Borges B (2023). shiny: Web Application Framework for R. R package version,