Integrating adsorptive pre-treatment and deep learning modeling for nitrogen management in marine recirculating aquaculture systems

Excess total ammonia nitrogen (TAN) and its oxidized forms, nitrite and nitrate, pose significant risks to marine aquaculture. This study evaluated the feasibility of adsorption-based technologies for nitrogen removal under seawater aquaculture conditions. Three synthesized adsorbents—chitosan hydrogel (Cht-HG), p -hydroxybenzoic acid-modified Cht-HG (M-Cht-HG), and alginate–hydrogel–metal–zeolite (Al-H-M-Z)—were benchmarked against a moving bed biofilm reactor (MBBR) and commercial zeolite. In pilot-scale fish tanks, Al-H-M-Z achieved rapid physicochemical TAN removal, with 75.6% efficiency within 24 h, reducing concentrations from 0.023 mg L –1 to 0.006 mg L –1 . Furthermore, it maintained TAN and nitrite levels below 0.20 mg L –1 for 7 days, demonstrating stable performance. In contrast, M-Cht-HG was effective for only 2 days, after which TAN levels exceeded 1.62 mg L –1 . At an initial concentration of 300 mg L –1 , a long short-term memory (LSTM) deep learning model predicted that Al-H-M-Z would remove 80.1 mg L –1 TAN (26.7%) and 86.5 mg L –1 nitrite (28.8%) within 24 h. This study provides a foundation for integrating advanced adsorbents with predictive modeling to improve nitrogen treatment strategies for sustainable marine aquaculture. • Various adsorbents, including hydrogels and coated zeolites, were developed. • Al-H-M-Z outperformed other adsorbents in aquaculture applications. • Al-H-M-Z exhibited 2.14- and 1.11-fold higher removal of TAN and nitrite vs. MBBR over 7 days. • LSTM predicted Al-H-M-Z can adsorb 80.1 mg L –1 and 86.5 mg L –1 of TAN and nitrite in seawater.