Supplement to "Ocean Color Algorithm for the Retrieval of the Particle Size Distribution and Carbon-Based Phytoplankton Size Classes Using a Two-Component Coated-Spheres Backscattering Model" Tihomir S. Kostadinov et al. S1 Assessment and propagation of uncertainties The Monte Carlo simulations (N=3000) of the two-population Mie scattering model were used to estimate uncertainty of the retrieved PSD parameters that is due to assumptions of the values of the model input parameters. The input distributions from which the varying input parameters were sampled for the Monte Carlo simulations are shown in main text Tables 1 and 2. This approach resulted in 3000 different realizations of modeled spectral b bp , for each ξ input value. The median of all realizations is 5 used as the SAM end-members, and Kruskal-Wallis analysis of variance is used to determine class similarity (where each input ξ value represents a class). This analysis determined, for each class, the neighboring classes that are statistically not different from it (at the 5% significance level) in terms of spectral angle. This gave a range of possible retrieved ξ values. This range is not always symmetric, but it is approximated as a symmetric 95% confidence interval (CI) and converted to standard deviation for use in subsequent analytical error propagation. The standard deviation of the corresponding N 0 parameter across all similar 10 classes and all Monte Carlo realizations is used as the uncertainty estimate for N 0 . The uncertainties of the PSD parameters and allometric coefficients were used with the first order derivatives of the derived products (absolute and fractional size-partitioned phyto C) with respect to those variables to analytically propagate the un- certainty estimate to the derived products. Phyto C absolute values are functions of both PSD parameters and both allometric coefficients, whereas the fractional values are functions only of ξ and the b allometric coefficient because a ratio of phyto 15 C values is taken (Eq. 5). We use the allometric coefficients obtained by Roy et al. (2017), who use information from three different regressions presented in Menden-Deuer and Lessard (2000). The weighted mean of the confidence intervals of the Menden-Deuer and Lessard (2000) allometric coefficients was used. Uncertainties were also propagated analytically to the composite imagery (monthly and overall climatologies) provided in the OC-CCI-derived PSD/phyto C data set (linked in main text Sec. 4) For more details see the scientific code used to do the uncertainty propagation (linked in main text Sec. 4), as well 20 as Kostadinov et al. (2016) and their Supplement. S1