Estimating biomass density in land-based cultivation of Ulva spp. using a low cost RGB imaging system

Seaweed cultivation faces scalability challenges due to labor-intensive biomass monitoring. Here, we demonstrate a low cost RGB imaging system for Ulva spp. biomass estimation (0.5–5.0 g L − 1 ) in land-based raceways. Using a generalized segmentation model, we extracted surface area and color (RGB, CIELAB) features as predictors for linear and log-linear regressions. Surface area proved the strongest predictor (R 2 = 0.99, p = < 0.001, RMSE = 0.18 g L − 1 ), with log-linear models outperforming linear regressions (R 2 = 0.88, p = < 0.001, RMSE = 0.50 g L − 1 ) in per revolution analysis. In validation under cross-device imaging conditions, the linear model produced more stable and less biased biomass estimates (mean error = 0.69–0.82 g L − 1 ) than the log-linear model, despite shifts in sensor resolution and acquisition geometry. Aggregating frame-level data into three-minute revolutions reduced segmentation errors and accounted for random seaweed distribution, enabling accurate biomass density predictions. This accessible, scalable approach offers a practical solution to reduce labor costs and optimize yields in precision aquaculture.