Integrating image processing and machine learning to enhance microalgal nitrate removal via optimized CO

Efficient nitrate removal in microalgal photobioreactors requires effective control of CO₂ gas holdup, pH, and dissolved oxygen (DO). This study introduces an integrated image-processing and machine-learning framework to optimize these parameters for nitrate removal using Scenedesmus sp. First, CO₂ gas–liquid hydrodynamics were evaluated in a 50-L bubble column without microalgae. Image analysis quantified gas holdup at CO₂ flow rates of 1–7 L/min and identified 5 L/min as the optimal rate, yielding a maximum holdup of 0.35 before bubble coalescence reduced retention. A dataset of 287 paired pH–DO measurements was then used to train four regression models. Gaussian Process Regression (GPR) showed superior performance (RMSE = 0.0519; R² = 0.9966) in capturing the nonlinear pH–DO relationship, confirming that carbonate equilibrium and gas–liquid mass transfer remain the dominant controls even during microalgae cultivation.In the second stage, the optimized conditions (CO₂ = 5 L min⁻¹, pH 6.5–7.0) were applied during microalgal cultivation to maintain bicarbonate availability, minimize CO₂ stripping, and prevent DO oversaturation. Under these controlled conditions, NO₃⁻–N concentration decreased from 50 to 4 mg L⁻¹ within 14 days (91% removal). In contrast, the uncontrolled system achieved only 50–52% removal due to inhibited nutrient uptake caused by elevated pH (> 10) and excessive DO (> 25 mg/L). Enhanced biomass accumulation (0.12–0.13 g L⁻¹) under controlled conditions confirms that optimized CO₂ holdup and pH–DO regulation translate hydrodynamic optimization into improved biological performance. The integrated framework provides a practical, low-cost strategy for microalgae-based groundwater denitrification.