Abstract. In this paper, we investigate the influence of uncertainties in inherent optical properties on the modelling of radiometric quantities by an ocean radiative transfer (RT) model, particularly irradiance and reflectance. The radiative transfer model is coupled to a 3D physical–biogeochemical model of the Black Sea. It describes the vertical propagation of incident irradiance within the water column along three streams in downward (direct and diffuse) and upward directions, with a spectral resolution of 25 nm in the visible range. The propagation of irradiance streams is governed by the inherent optical properties of four major optically active constituents found in seawater and provided by the biogeochemical model: pure water, phytoplankton, non-algal particles, and coloured dissolved organic matter (CDOM). Sea surface reflectance is then derived as the ratio between the simulated upward and downward irradiance streams, directly connecting the model with remote-sensed data. In this configuration, the coupling is one-way: the radiative transfer model is projecting model variables into the space of satellite observations, working as an observation operator. In the stochastic version of the model, uncertainties are injected in the form of random perturbations of the inherent optical properties of the water constituents. Different ensemble configurations are derived, and their quality is assessed by comparison with in situ and remote-sensed observations. We find that the modelling of the uncertainties in the radiative transfer model parameterisation allows us to simulate distributions of radiative fields that are partially consistent with observations. The ensemble is consistent with remote-sensed reflectance data in summer and autumn, especially in the central parts of the basin. The quality of the ensemble is lower in winter and early spring, suggesting the existence of another major source of uncertainty or that the quality of the deterministic solution is insufficient. CDOM dominates absorption in short wavebands with a relatively high uncertainty that influences irradiance and reflectance outputs. This dominant role calls for better representation of CDOM to improve model calibration. Contributions from phytoplankton and non-algal particles are more significant for (back)scattering. The results of this paper suggest that the integration of a radiative transfer model into a physical–biogeochemical model would be beneficial for calibration, validation, and data assimilation purposes, offering a better link between model variables and radiometric observations.