Spectral ocean-colour reflectance is increasingly used to constrain marine ecosystem models, yet its potential to provide direct observational constraints on process-based biogeochemical models remains underexplored. Here we calibrate multi-wavelength remote sensing reflectance (R_RS(λ)) using a genetic algorithm within a one-dimensional configuration of the Biogeochemical Flux Model. The approach is applied to two contrasting Mediterranean environments: an open-ocean ecosystem in the Ligurian Sea (BOUSSOLE observatory) and an optically complex coastal site in the Northern Adriatic Sea (Acqua Alta oceanographic tower – AAOT). At both sites, the calibration successfully reproduces observed R_RS(λ) at six visible wavelengths, demonstrating that the model captures the main features of the R_RS(λ) spectra across contrasting trophic regimes. However, the mechanisms used by the model to match R_RS(λ) differ markedly and reveal distinct controls on particle scattering. At BOUSSOLE, R_RS(λ) calibration primarily adjusts phytoplankton and small-particle optical properties, improving phytoplankton biomass estimates and increasing particulate backscattering. This highlights the dominant role of small and algal particles in shaping backscattering in oligotrophic waters. In contrast, at AAOT, R_RS(λ) calibration mainly modifies suspended particulate matter (SPM) optical parameters, with limited changes in the biogeochemical state. Overall, R_RS(λ) calibration reproduces observed optical signals while revealing site-dependent controls on particulate backscattering. Sensitivity and calibration exercises improved predictive capability within the calibrated parameter space, but residual error patterns revealed potential structural limitations or simplifications in the model's optical representation. This work demonstrates the potential of R_RS(λ) observables to guide improvements in the representation of particle optical properties and biogeochemical processes in marine ecosystem models.