Predicting Key Substance Levels in Aquaculture through AI-Based Water Quality Monitoring

Abstract

Land-based aquaculture farms use seawater transported from nearby seas instead of large amounts of freshwater. A seawater recirculating filtration system is essential for sustainable fish farming; however, this system has limitations in improving the levels of ammonia, nitrite, and nitrate, which are directly linked to fish mortality. Therefore, most land-based aquaculture farms periodically exchange a certain amount of seawater to maintain optimal water quality. Despite these efforts, managing water quality remains a significant challenge due to the fluctuating levels of these harmful substances.This study aims to address this challenge by predicting the levels of ammonia, nitrite, and nitrate—the primary causes of fish mortality in land-based aquaculture—using AI models. The training data were collected from various sensors installed in the farms, including those measuring water temperature, dissolved oxygen, dissolved solids, pH level, oxidation-reduction potential, salinity, nitrate, and ammonia. By leveraging this comprehensive dataset, we evaluated the performance of multiple models, such as Random Forest (RF) and K-Neighbors Regressor (KNN).The study demonstrated that these models could achieve remarkable performance metrics, with the Random Forest model recording an MAE of 0.0150, MSE of 0.0008, RMSE of 0.0289, R² of 0.9999, RMSLE of 0.0039, and MAPE of 0.0024. Such high accuracy levels indicate that AI-based water quality prediction models have significant potential for effectively monitoring and predicting fish health in aquaculture farms. Implementing these AI models can lead to more proactive and precise management of water quality, ultimately reducing fish mortality rates and enhancing the sustainability and profitability of aquaculture operations.

Keywords: Water Quality Prediction, Aquaculture, Fish, AI, Monitoring

DOI: 10.54941/ahfe1005480