FishMet: A Digital Twin Framework for Appetite, Feeding Decisions and Growth in Salmonid Fish

Abstract

Salmonids are important fish species in aquaculture in countries in the temperate zone. Optimisation of feeding in next-generation precision fish farming requires developing models for decision support and process control. Black box ML and AI models are often very efficient but have drawbacks, such as requiring large amount of training data and reduced performance in novel situations where no data are available. Thus, developing realistic process models of fish appetite, feeding decisions, feed intake, energetics and growth is necessary. Such models are essential for predicting fish performance, for example, feed intake, waste from uneaten feed and faeces, growth, in novel ‘what if’ scenario testing. We have built a conceptual model based on a review of major neurophysiological mechanisms and feedback loops controlling appetite and food intake in fish. Building on this, we have developed the FishMet model: a new extensible stochastic simulation framework that represents the basic feedback loops controlling appetite, feeding decisions, energy budget and growth in salmonid fish. The appetite and feeding decision model in FishMet is the novel advance, while the bioenergetic part follows the established theory. The model is supported by server-based components and open API for data assimilation and on-demand model execution that allows to use FishMet as a digital twin. We demonstrate relatively good prediction of stomach and gut digesta transit and food intake in the rainbow trout Oncorhynchus mykiss. The digital twin also demonstrated good prediction of growth and feeding efficiency in a pilot scale experiment on the Atlantic salmon Salmo salar. We discuss the concept of the digital twin and the directions of further development of the model as an applied predictive tool.