Recirculating aquaculture systems (RAS) are sustainable methods for fish production developed to reduce water consumption and waste discharge. To increase the fish yield in these systems, fish densities and, consequently, feed input are high, potentially leading to high levels of total ammonia nitrogen (TAN) in these systems. The un-ionized fraction (NH3 ) penetrates fish cell membranes via the lipoidal segments and is considered harmful as it might cause unhealthy, suboptimal fish growth and in extreme cases, fish mortality. Therefore, the real-time detection of the un-ionized ammonia concentration is an urgent issue that needs to be addressed in aquaculture. Existing methods for estimating TAN or its fractions , NH3 and NH4+, are usually either offline methods (e.g. molecular absorption spectroscopy), or methods that suffer inaccurate estimations, especially in high turbidity and metal ions (e.g. Nessler’s and hypobromite oxidation ). Alternatively, soft data-based methods are used, yet these methods suffer from low accuracy, large datasets, or "black box" characteristics.
In this study, we propose data assimilation (DA) as a cost-effective method for improved real-time estimation of NH3 and NH4+ (and TAN) in RAS. DA combines information from both simulation models and measurements to obtain optimal estimates of variables of interest. The combination of the measurements and the simulation model is supposed to provide a higher accuracy of the state variables than both the estimation of the simulation model and the measurement. In this work, an extended Kalman filter (EKF) is used to estimate NH3, NH4+ , TAN, and related specifications of the RAS based on simulation estimation and a TAN measurement. A simulation model of a fish tank and a moving-bed-bio-reactor (MBBR) (bio-filter (BF)) is first composed as a set of equations that describes the dynamics of the NH3 and NH4+ in RAS. EKF equations are then formulated based on the simulation model and TAN measurement. We validated our method through synthetic and laboratory case studies and demonstrated its superior estimation capability as compared to the in situ measurements or the simulation models. Additionally, BF specifications of ammonia removal rate were reliably estimated in real time. Furthermore, improved ammonia estimation led to improved current and future fish weight estimations, which can be essential for reliable RAS management. The proposed approach facilitates wider adoption of DA in challenging estimations in aquaculture.