Machine learning–based decision support for harvest optimization in commercial microalgae photobioreactors: A field validation study

Operational decision-making in commercial microalgae cultivation remains largely heuristic despite the increasing availability of high-frequency farm data. In photobioreactor-based production systems, harvest fraction selection is a key management decision that affects short-term biomass recovery, culture stability, and overall production efficiency. This study presents a machine learning–based decision-support framework for optimizing harvest management in commercial tubular photobioreactors cultivating Nannochloropsis sp. The framework integrates three years of heterogeneous farm data, including operational logs, meteorological variables, and FlowCam (a digital imaging system used to measure microalgae concentration and cell morphology) data. A supervised regression approach was used to predict next-day post-harvest cell concentration, and several learning algorithms were evaluated. A Random Forest model achieved the best predictive performance (R² = 0.705), demonstrating robust prediction under noisy commercial production conditions. The trained model was implemented in an inference module that evaluates candidate harvest fractions and recommends the fraction expected to maximize short-term biomass recovery. The system was validated in a four-week farm-scale experiment across 12 commercial photobioreactors (96 planned harvest events), comparing model-guided and operator-guided harvest decisions. While the machine learning–guided strategy achieved comparable productivity levels to expert human decision-making, it reduced variability. These results provide a successful proof-of-concept for the operational feasibility of a data-driven framework to stabilize and scale commercial microalgae management. These findings demonstrate the practical feasibility of integrating machine learning into daily farm operations as a hybrid human–AI decision-support layer. The proposed framework highlights the potential of data-driven decision systems to enhance stability and scalability in commercial microalgae photobioreactor production.