Water-leaving radiance is subject to depth variability of the water constituents. The optical penetration depth is strongly dependent on the wavelength λ, which allows to retrieve a non-uniform vertical profile of an optically-active constituent CTSM(z) from remote-sensing reflectance Rrs(λ,Cz). We define the apparent particle concentration CTSM,app(λ) of a vertically homogeneous water column whose Rrs(λ,Cconst) matches Rrs(λ,Cz). Subsequently, we define a vertically-weighted averaged particle concentration CTSM,ave(λ), only dependent on CTSM(z), and retrieve CTSM(z) by minimizing the error between CTSM,app(λ) and CTSM,ave(λ) with genetic algorithms. We conclude that the retrieval is excellent if the sub-surface maximum lays close to the surface or the background concentration of CTSM(z) is low. Conversely, results worsen for opposite conditions, due to insufficient signal strength from superimposed sub-surface maxima.