A DIGITAL TWIN FOR IMPLEMENTING THE PRECISION FISH FARMING APPROACH IN RAS

Introduction

The Precision Fish Farming (PFF) approach integrates sensors, predictive models and decision support systems in four-phase-cycle which include: 1) Observe: real time data from the sensors; 2) Interpret: predictive models to retrieve information from the system; 3) Decide: artificial intelligence (AI) & decision support system; 4) Act: automation & output correction. According to (Fore&al., 2024), the Digital Twin (DT) concept represents a suitable framework for implementing the PFF in cage culture. In this paper, we extend the approach to Recirculation Aquaculture Systems (RAS) and propose a web-based application for allowing end users to test the Knowledge Based Model component of the DT. This study was developed in the framework of the Italy Slovenia INTERREG project "Circular Rainbow", aimed at promoting the transition from Flow Through System to Hybrid Flow Systems and RAS in trout farming. The latter would lead to enhanced circularity, better control of water quality and hence fish welfare, higher resilience to the effect of climate change on water quantity and quality.

Materials and methods

A DT can be defined as "a virtual representation of a physical asset enabled through data and simulators for visualising, predicting, monitoring, optimising and controlling system states, and improved decision making". It includes three main components: i) an observation system; ii) Knowledge Based Models (KBM); iii) data driven models. Taking into account the typical structure and functionality of a RAS, the KBM component included a Mass Budget (MB) and an Energy Budget (EB) module. The data driven component, thus far, has been implemented only for improving the water temperature prediction of the EB module using the Extended Kalman Filter. The MB module includes biotic and abiotic variables, namely "Mean weight of the fish", "Number of fish", "Total ammonia nitrogen", "Nitrate", "Dissolved oxygen", "Dissolved inorganic carbon", "Alkalinity" and "Total suspended solids". The MB module simulates the evolution of the fish biomass, water circulation, chemical and biochemical processes, deposition of suspended solids. Most processes are affected by water temperature, which is the output of the EB module. The main forcings/controls are: i) feed ration; ii) oxygen supply; iii) air temperature and humidity; iv) quantity and quality of make-up water; v) addition of a base, e.g. NaHCO3, to control pH. The DT was tested on a pilot RAS located in Pagnacco (Udine, Italy). The observation system included automatic sensors for detecting water temperature, dissolved oxygen, pH, CO2 and energy use of the most relevant devices, e.g. heat pump, every 15 minutes. This data set was complemented by the weekly sampling of TAN, nitrate, Reactive Phosphorus, Alkalinity. The KBM was coded using R: from the mathematical point of view, the code solved a system of Dynamyc Algebraic Equations including 8 * 16 state variables, where 16 is the number of components of the pilot RAS, as pH and the speciation of DIC and DIN are simulated assuming chemical equilibrium. A demo version of the KBM is freely available at the web interface DiGiTrout: https://bluefarm.shinyapps.io/DiGiTrout/ , which includes also KBM of FTS and HFS virtual farms. This interface was developed in R (v4.5.1) using the Shiny package (v1.11.1).

Results & discussion

The output of the KBM component of the DT compared well with the data collected in the pilot RAS from October 2025 to February 2026. The growth of rainbow trout was correctly simulated within the weight range 150-500 grams, as shown in Figure 1-a, which compares the model output (line) with the mean weight, averaged over the 12 tanks included in the pilot. The processes affecting the water quality are quite complex: the KBM correctly simulated the trend in nitrate, not shown, related to the increase in fish biomass and, therefore, TAN excretion, thus suggesting the biofilter activity was well reproduced. One of the most relevant model output is pH, which is affected by fish metabolism but depends also on the buffer capacity of the solution: as one can see from Fig. 1b, the trend in pH is correctly simulate both in fish tanks and biofilter: however, the comparison also indicates a slight bias, which should be further investigated. The concentrations of Dissolved oxygen and CO2 depends also on the efficiency of air blower and degassing units, which were tuned to fit the data.

Concerning the EB module, water temperature and air temperature inside the building were correctly simulated. The Kalman Filter adjusts a parameter used to describe the building transmittance with high precision, with small fluctuations, indicating that the pre-simulation calibration was accurate. Reversing these calculations, the heat provided by the heat pump can be retrieved. This information allows the operator to estimate energy costs related to water cooling/heating.

Figure 1: a) Mean weight in all tanks: comparison between model output (line) and manual samples (point). b) pH in a fish tank and biofilter: comparison between model output (blue and red lines) and data from sensors (violet and green lines)

Overall, the comparison between the KBM component of the DT and the data observed in the Circular Rainbow project indicated that the complex physical, biochemical and chemical processes occurring in a RAS were satisfactorily modelled. On this basis, a prototype of a general KBM for a RAS system was designed, coded and made available through the DiGiTrout web interface, an application which allows one to simulate two RAS management scenarios: 1) daily management (7-day ahead simulation); 2) long term planning (6-month ahead simulation). The web application provides a list of Key Performance Indicators (KPIs), e.g. feed demand, Specific Growth Rate (SGR), fresh water used as input, Feed Conversion Ratio (FCR), total suspended solids collected, dissolved oxygen provided by the air blowers. This information can be used by farmers to improve management. In conclusion, a DT can not only provide a virtual RAS with results comparable to the real systems, but also provide actionable information regarding costs and use of key resources such as feed, fresh water and energy. DT can open new scenarios in the aquaculture sector, optimizing productivity and also allowing farmers to upscale their plant, since forecasts of trends and costs can be generated with high confidence.

Acknowledgement

This research was funded by the Italy-Slovenia INTERREG VI-A 2021-2017 project "Circular Rainbow" cod. ITA-SI0600132

References

Føre M. M. O. Alver, J. A. Alfredsen, A. Rasheed, T. Hukkelås, H. V. Bjelland, B. Su, S.J. Ohrem, E. Kelasidi, T. Norton, N. Papandroulakis. Digital Twins in intensive aquaculture—Challenges, opportunities and future prospects. Computers and Electronics in Agriculture 218 (2024) 108676

Rasheed, A., San, O., Kvamsdal, T., 2020. Digital twin: Values, challenges and enablers from a modeling perspective. IEEE Access 8, 21980–22012. http://dx.doi.org/10. 1109/ACCESS.2020.2970143