Aquaculture America 2026

February 16 - 19, 2026

Las Vegas, Nevada

Add To Calendar 19/02/2026 13:45:0019/02/2026 14:05:00America/Los_AngelesAquaculture America 2026REAL-TIME MACHINE LEARNING PREDICTIVE WATER QUALITY SENSING AND BIOFLOC-DRIVEN WASTE RECOVERY IN A RECIRCULATING AQUACULTURE SYSTEMLoireThe World Aquaculture Societyjohnc@was.orgfalseDD/MM/YYYYanrl65yqlzh3g1q0dme13067

REAL-TIME MACHINE LEARNING PREDICTIVE WATER QUALITY SENSING AND BIOFLOC-DRIVEN WASTE RECOVERY IN A RECIRCULATING AQUACULTURE SYSTEM

Robel K. Adane*, A. Gross, Elad Levintal

Ben Gurion University,

Sde boker, Israel

kahsu@post.bgu.ac.il

 



This work presents an integrated, closed-loop recirculating aquaculture system (RAS) that pairs a dedicated side-stream biofloc reactor with low-cost edge analytics to recover waste and stabilize fish water quality (Fig.1). Effluent from the fish tank is directed to the biofloc reactor, where solids accumulate and microbial communities convert dissolved and particulate nitrogen into biomass. The clarified water then returns to the fish tank. The reactor was operated under controlled oxidation–reduction potential (ORP) regimes in microaerophilic conditions (- 100 to 100 mV), the system captured up to 43% of the feed nitrogen as microbial protein, maintained total suspended solids (TSS) at 3000 – 10,000 mg/L, and returned water with a total ammonia nitrogen (TAN) concentration below 0.3 mg N/L. To enable autonomous operation, we built a sensing node with a cost of approximately $1,100–$1,800 USD, utilizing commercial probes (pH, DO, EC, ORP, turbidity, and temperature), an Adafruit Feather M0 microcontroller, and cellular telemetry. The node streams data once per minute and runs TinyML models on board to predict TSS and TAN in real-time. Models were trained on an 8-week dataset, quality-controlled for pump downtime and sensor drift, and validated. The TSS model achieved an R² of 0.83 (RMSE = 561 mg/L), and the TAN model achieved an R² of 0.88 (RMSE = 2.06 mg/L), enabling direct operational decisions on aeration, carbon dosing, and solids harvesting without the need for laboratory delays. This framework links process understanding to automated control and production.