| 16 Oct 2025
Three-stream modelling of radiative transfer for the simulation of Black Sea biogeochemistry in a NEMO framework
Abstract. In this paper, we propose a three-stream ocean radiative transfer (RT) module as an extension of the NEMO ocean modelling framework. This module solves the subsurface irradiance field in 1D water columns, discriminating between two downward streams, direct and scattered, and a backscattered upward stream. The module solves 33 wavebands ranging between 250 and 4000 nm, with a finer 25 nm resolution in the visible range. The sea surface reflectance is also calculated as a model output, based on the ratio between the upward and downward irradiances at the air-sea interface. An optional feedback towards NEMO is presented, enabling the use of irradiances to compute temperature in the hydrodynamics. The module also includes a stochastic version in which the inherent optical properties of the main optically active components of seawater can be perturbed. This mode is meant to account for uncertainty in the modelling of marine optics. This module is can be plugged to any NEMO configuration, with the computation of optical properties either driven by a biogeochemical model or directly forced into the RT module.
We apply this module in a test case for the Black Sea, coupled with the physical-biogeochemical framework NEMO 4.2.0-BAMHBI. We find that substituting the existing radiative transfer scheme with our model unlocks the ability to simulate radiometric variables that can be compared more truthfully to observations, both in situ and from remote-sensing. We also find that using irradiances to compute the temperature and PAR in the model maintains consistency in the calculation of physical and biogeochemical variables in the model, such as temperature or chlorophyll concentration, while enabling additional capabilities in the model in the simulation of radiometric quantities.
General appraisal
The paper by Macé and co-authors describes an effort to improve the modelling of light propagation within the NEMO ocean modelling framework (NEMO stands for “Nucleus for European Modelling of the Ocean”).
Improving modelling of light propagation without introducing excessively computationally expensive parameterisations is a long-standing question in the community of developers of global circulation models coupled with some biogeochemical (BGC) components. This is an important aspect that drives the profile of heat deposition in the top layers of the ocean and also determines how much energy is available for photosynthesis, as input to the BGC component.
The authors acknowledge that this modelling is often oversimplified. The work they detail in this manuscript intends to improve this.
The authors should be commended for such an attempt.
However, I did not see much novelty here. The 3-stream model that they incorporate, based on Aas 1987 in particular, was already selected and used by Dutkiewicz et al. (2015) (as indeed cited by the authors). I could not find a clear explanation of what is new here. But maybe this is not what this journal expects, so presenting the implementation of an existing method and how it modifies the outputs of the model could be enough. I leave this to the appreciation of the editor.
Also, there is a sense that this model is made of odds and ends, if I may use this expression, and the overall coherence of the various parameterisations is hard to fathom.
Besides this, I do not think that testing this new RT model implementation in the Black Sea is the most relevant choice. Most of that sea is made of “optically complex” waters, making the modelling of optical properties quite difficult (e.g., comments about coccolithophorids lines 420-425), and also making the uncertainties of the satellite products (used for validation) higher than in an open ocean setting. Definitely not the best framework. The Western Mediterranean Sea could have been better, as any other open ocean area (I understand that the choice of the Black Sea is essentially made for practical reasons because an implementation of the model already exists for this area). There is for instance an open ocean field site in the Western Mediterranean (BOUSSOLE site) where most of the optical properties and radiometric quantities are measured and could help validating a model run in that area.
I also have some doubts about the robustness (validity) of the argument that modelling the reflectance gives a better opportunity for validating the model against satellite-derived reflectances than validating against the chlorophyll (Chl) (as alluded to on top of page 3).
When validating the modelled Chl with satellite-derived Chl, the BGC model uncertainties in the calculation of Chl come into play as well as the uncertainties of the satellite-derived Chl. When validating (comparing) the reflectances, the uncertainty of the reflectance-to-Chl calculation is removed on the satellite side, yet this uncertainty is “transferred” into the model calculation of reflectance from optical properties, themselves calculated from some of the model state variables. Therefore, the benefit (other than practical) is unclear to me in terms of reducing the level of uncertainty in the validation process. I have tried to summarise this in the attached sketch.
Otherwise, when looking at the results in Figs. 3 to 6, the differences between the “simple optics” and the new “RT model” seem really minute. So, it is not obvious to me what improvements the new modelling brings. Then Figs 7-11 show quite poor results for the validation against satellite products. It is then hard to be convinced of the usefulness of the new model. There is no stated goal in terms of acceptable RMS or bias, which makes improvements, if any, hard to qualify as either significant or simply within the uncertainty of the modelling.
In any case, the quality of the outputs of the best possible RT model entirely depends on the relevance of the inherent optical properties (IOPs, i.e., absorption and scattering) that are used as input. This aspect is barely addressed in the manuscript, when I think it should be the very first step to check. For instance, the Chl-specific absorption coefficients for phytoplankton displayed in Fig. 1 do not show much difference for the three phytoplankton groups, when the size should normally matter to determine this coefficient. It is also supposed to be depending on Chl as well. So, why these values, where do they come from?
I would expect a revised version of the manuscript to discuss these points. Apart from this, I think this paper is publishable after some minor changes, however.
On a more general note, I do not understand why the modelling community does not try to work more closely with qualified RT and bio-optics specialists when trying to improve the modelling of light propagation. There might be some good justifications to use the parameterisations that have been chosen here but this is not made clear (neither it was in Dutkiewicz et al., 2015, by the way). For instance, the separation of the direct and diffuse components of the downward solar radiation is not really needed here. Modelling the diffuse attenuation coefficient for downward irradiance (Kd) does not require this separation (neither for heat deposition nor for photosynthesis).
A few detailed comments