Aquaculture Europe 2021

October 4 - 7, 2021

Funchal, Madeira

Add To Calendar 05/10/2021 11:30:0005/10/2021 11:50:00Europe/LisbonAquaculture Europe 2021DATA DRIVEN INSIGHT INTO FISH BEHAVIOUR AND THEIR USE FOR PRECISION AQUACULTURESidney-HotelThe European Aquaculture Societywebmaster@aquaeas.orgfalseDD/MM/YYYYaaVZHLXMfzTRLzDrHmAi181982

DATA DRIVEN INSIGHT INTO FISH BEHAVIOUR AND THEIR USE FOR PRECISION AQUACULTURE

F. O’Donncha1, Caitlin L. Stockwell2, Sonia Rey Planellas3, Giulia Micallef4 , Paulito Palmes1, Chris Webb5, Ramon Filgueira6, Jon Grant2

 

1 IBM Research – Ireland, Damastown Ind. Park, Mulhuddart, Dublin 15

 Email: feardonn@ie.ibm.com

2Department of Oceanography, Dalhousie University, Halifax, Nova Scotia B3H  4R2 Canada

3University of Stirling, Scotland

4Gildeskål Research Station, Norway

 5 Cooke Aquaculture, Scotland

6Marine Affairs Program, Dalhousie University, Halifax, NS B3H 4R2 Canada

 



Introduction

 Precision aquaculture

 involves a variety of sensors used to gain insight into the farm environment, make decisions which optimize fish health , welfare , growth, and economic return, and reduce risk to the environment. Fundamental to those is monitoring of environmental and animal processes within a cage and processing those data towards farm insight using models and analytics. This paper presents an analysis of environmental and fish behaviour datasets collected at three salmon farms in Norway, Scotland, and Canada. Information on fish behaviour were collected using hydroacoustic sensors that sampled the vertical distribution of fish in a cage at high spatial and temporal resolution, while a network of environmental sensors characterised local site conditions. We present an analysis of the hydroacoustic datasets using AutoML (or automatic machine learning) tools that enables developers with limited machine learning expertise to train high-quality models specific to the data at hand. We demonstrate how AutoML pipelines can be readily applied to aquaculture datasets to interrogate the data and quantify the primary features that explains data variance.

Materials and Methods

 Hydroacoustic methods provide a proxy measure for density and distribution of marine animals in form of acoustic backscattering

. Advantages linked to hydroacoustic sampling techniques include, high spatial and temporal resolution, autonomous long-term sampling duration, range (especially during poor visibility when visual-based methods tend to fail), and a non-invasive surveying approach

 .

 This study considers three salmon cage farms in Norway (NOR), Scotland (SCO), and Canada (CAN). For each site several environmental sensors were deployed monitoring a range of parameters, including temperature, dissolved oxygen (DO) , and current speed. These were complemented with weather data from in-situ weather stations or model generated reanalysis, and open-ocean model  from  the E.U. Copernicus Marine Service Information model repository .

 Data were processed using an automatic machine learning framework. AutoML systems uses a variety of techniques, such as, differentiable programming, tree search, evolutionary algorithms, and Bayesian optimization, to find the best machine learning pipelines for a given task and dataset

 .

Results

 We interrogated relationships between vertical distribution of fish in a cage and environmental variables at the three  geographically  disparate sites. Statistical analysis explored the diel patterns, and how data distributions varied over the duration of the study ,  while  IBM AutoAI was used to quantify the effects of environmental variations on the vertical movement of the fish. For the machine learning interrogation, we provided input features that literature suggests influence salmon behaviour (and were available at the study sites). For our study, these were temperature, DO, current speed, wind speed, and salinity, together with hour-of-day (see Fig 1, Y-Axis). The resultant model explained 59%, 64% and 61% of variance for the NOR, SCO, and CAN sites, respectively .  Figure  1 presents the variable importance computed for the three locations . While there were similarities in the drivers that influenced fish behaviour at the three sites, pronounced variations existed based on the different geography and characteristics of each site.

Results presented in this paper indicate pronounced differences between sites and the need to consider these variations for farm management. One could readily use this approach to quantify the difference between sites, and further to identify the fundamental drivers to these variations. This could be particularly valuable when comparing different farm systems such as inshore and offshore and the associated operational implications.

References

Drori, I., Krishnamurthy, Y., Rampin, R., De, R., Lourenco, P., Ono, J. P., Cho, K., Silva, C., & Freire, J. (2018). AutoML Workshop. AutoML Workshop at ICML. https://www.cs.columbia.edu/~idrori/AlphaD3M.pdf

Foote, K. G. (2009). Acoustic Methods: Brief Review and Prospects for Advancing Fisheries Research. In The Future of Fisheries Science in North America (pp. 313–343). Springer Netherlands. https://doi.org/10.1007/978-1-4020-9210-7_18

Føre, M., Frank, K., Svendsen, E., Alfredsen, J. A., Dempster, T., Eguiraun, H., Watson, W., Stahl, A., Sunde, L. M., Schellewald, C., Skøien, K. R., & Alver, M. O. (2018). Precision fish farming: A new framework to improve production in aquaculture. Biosystems Engineering, 173, 176–193. https://doi.org/10.1016/J.BIOSYSTEMSENG.2017.10.014

Juell, J., Furevik, D., & Bjordal, A. (1993). Demand feeding in salmon farming by hydroacoustic food detection. Aquacultural Engineering, 12(3), 155–167. https://doi.org/https://doi.org/10.1016/0144-8609(93)90008-Y

Scherelis, C., Penesis, I., Hemer, M. A., Cossu, R., Wright, J. T., & Guihen, D. (2020). Investigating biophysical linkages at tidal energy candidate sites; A case study for combining environmental assessment and resource characterisation. Renewable Energy, 159, 399–413. https://doi.org/10.1016/j.renene.2020.05.109