Detecting and quantifying cyanobacteria algal bloom occurrence plays an important role in preventing public health risks and understanding aquatic ecosystem dynamics. Satellite remote sensing has been used as an important data source to estimate cyanobacteria biomass based on pigment concentration. Phycocyanin (PC) is a unique pigment of inland water cyanobacteria and has been widely used as a proxy for cyanobacteria algal biomass. Based on the PC absorption feature around 620 nm, scientific efforts have been made to develop bio-optical models for orbital satellite observations, but proposed PC models limit the retrievals at different concentration ranges and depend on empirical models calibrated for specific aquatic environments. This study proposes a hybrid machine learning approach for PC retrieval that efficiently adopts the optimal algorithm for specific PC concentration ranges. An in-situ dataset of 165 samples was collected between November 2020 and December 2021 to support full training and validation of the proposed method. First, a Random Forest algorithm was applied to classify PC-low-concentration waters (0 – ∼14 mg.m−3) and PC-high-concentration waters (∼14.1 – 300 mg.m−3). Then, for each defined class, an individual PC estimation algorithm was calibrated. The final PC-hybrid model was applied to atmospherically corrected Sentinel-3/OLCI imagery derived by three approaches (L2-WFR, 6SV, and ACOLITE). The PC hybrid-model performance was evaluated by comparing the estimated PC concentration from satellite and in situ measurements. The hybrid PC model estimates (median symmetric accuracy (ζ) = 25.35%) outperformed the individual PC algorithms calibrated for the entire range of PC concentration, proving the practical applicability for quantifying PC concentration in optically dynamic waters.