Machine learning-driven nitrogen management in mariculture wastewater: A critical review

With the growth of global demand for aquatic protein, mariculture has become an important source of food supply. Effective nitrogen management is a critical challenge for the sustainable development of mariculture. Mariculture wastewater is characterized by high salinity, low carbon-to-nitrogen (C/N) ratio and complex nitrogen forms, which severely tests the stability and economic viability of prevailing physicochemical, biological and emerging treatment technologies. Current control strategies, which rely heavily on manual operation and static parameters, are inadequate for managing the complex nonlinear disturbances inherent in such systems. This review aims to systematically synthesize the challenges of nitrogen management in mariculture wastewater and explore the potential of machine learning (ML)-driven strategies to address them. It first describes the sources, transformation and fate of nitrogen forms to identify key treatment targets. Subsequently, it critically evaluates the efficiency and inherent limitations of existing treatment technologies, attributing their performance instability to the core contradiction between static control paradigms and dynamic system requirements. Through data-driven approaches, ML enables dynamic prediction of nitrogen forms, intelligent optimization of unit processes and coordinated control across the treatment train, thereby shifting the management strategy from passive response to proactive anticipation. Key challenges for practical ML application, including data quality, edge computing constraints, system integration and economic feasibility, are also discussed. By integrating these facets, this review establishes a foundational framework for transitioning from conventional nitrogen removal to an intelligent, precise and sustainable management paradigm for the mariculture industry.