We investigate the impact on remote sensing reflectance by the vertical non-uniformities of water constituents. Reflectance simulated for 210 pairs of in situ measured chlorophyll-a and turbidity profiles (z = 0–20 m) from Lake Geneva are compared to simulations for uniform constituent gradients and non-uniform profiles approximated by Gaussian curves, orthogonal layers and steady gradients. Relevant concentration ranges are between 0 and 17 mg m−3 for chlorophyll-a and 0 and 4.6 g m−3 for total suspended matter within the photic layer. Our results show that mesotrophic lakes are specifically sensitive to non-uniformities with 20% of the 210 samples used in this study showing deviations of the spectral angle > 5° between a uniform assumption and observations which mostly occur for deeper-laying water constituents. By stressing the different use of blue and red parts of the spectrum, we argue further that algorithms are affected by variable vertical structures of algal and inorganic particles. Finally, we demonstrate that approximation models of the vertical structure of water constituents are a good solution to better account for non-uniformities in the development of invertible bio-optical models.