Precision Aquaculture: What we need
A data processing and modelling engine
The key objective of precision aquaculture is
moving beyond data – towards decision. At its most basic level, this is a big data problem that requires processing data from disparate sources and extracting actionable insight from these data. The data involved ranges from standard time series datasets at the sensor level, to image or geospatial based datasets from sources such as a camera or HF Radar, to satellite (e.g. ocean color) or numerical models or products (e.g. ocean models or weather datasets).
At a more detailed level, it requires the ability to incorporate information from mechanistic models and human expertise via tools such as physics-based models or human-in-the-loop machine learning implementations. Physics models have had huge applicability in aquaculture considering aspects from fish growth (Cubillo et. al. 2016), environmental effects (O'Donncha et. al. 2013), and fish health (Myksvoll et. al. 2018).
The main challenge of physics-based models in aquaculture is the difficulty of deriving a full mechanistic description of key subsystems. Establishing an understanding of such relationships between environmental or farm conditions and fish response is a prerequisite for being able to derive optimal settings for farm settings (feed, supplementation, veterinary intervention). Moreover, a thorough understanding of the system dynamics is also key in identifying possible Gold Standards to validate such methods (Føre et. al. 2018). Developing mechanistic models of farm systems requires detailed domain expertise and a multidisciplinary approach involving stakeholders from mathematics, fish health, environmental, physics, and biology amongst others.
A precision aquaculture system requires a data platform that is scalable to these disparate needs. Chief amongst these are capabilities to seamlessly ingest data from multiple sources and act on those. A related requirement is the need to apply statistical and machine learning analysis and forecasting to data to inform on critical aspects of farm health. Finally, a close alignment with mechanistic modeling is required in terms of both developing (ease of programming) and deploying (compute scalability) complex models.
Hence precision aquaculture requires programming tools that are easy to develop and maintain, flexible to integrate data from disparate sources, a wide variety of statistical and machine learning libraries, and scalable across CPU and GPU systems.
Combining these capabilities we can accurately interrogate dynamics on farms. Consider one application towards salmon farms here
REFERENCES:
Cubillo, A. M., et al. "Role of deposit feeders in integrated multi-trophic aquaculture—a model analysis." Aquaculture 453 (2016): 54-66.
Føre, Martin, et al. "Precision fish farming: a new framework to improve production in aquaculture." biosystems engineering 173 (2018): 176-193.
Myksvoll, Mari Skuggedal, et al. "Evaluation of a national operational salmon lice monitoring system—From physics to fish." PLoS One 13.7 (2018): e0201338.
O’Donncha, Fearghal, Michael Hartnett, and Stephen Nash. "Physical and
numerical investigation of the hydrodynamic implications of aquaculture
farms." Aquacultural engineering 52 (2013): 14-26.