5.1. Accuracy of Numerical Computations
A numerical integration over 
 is required to derive 
 and 
 from the normalized phase function (
Section 2). Due to the sharp increase of the normalized phase function in the forward scattering directions (
Figure 3), the selection of the relevant angular step for the numerical integration is crucial. For that purpose, the impact of angular step (
) on the calculation of 
 is studied using Lorentz-Mie simulations in DS1 (
Figure 1, N
oM2, M3). The normalized phase function of polydisperse particles 
 exhibits a maximum around 
 = 0
o [
26]. For small 
 value, that is when the proportion of large-sized particles compared to smaller particles increases, the forward peak is sharper. Indeed, for particles with a large diameter as compared to the wavelength, 
 displays a sharp forward peak [
26] due the concentration of light near 
 = 0
o caused by diffraction. The presence of the peak in 
 requires several integration points large enough to provide the desired numerical accuracy. The numerical integration over 
 (
Figure 1, N
oM2) is performed using the “Trapz” function from the Numpy package with Python. The “Trapz” function performs an integration along the given axis using the composite trapezoidal rule. To test the accuracy of the integration and to find the correct integration step, 
, we compare the result of the numerical integration of 
 between 0 and 
 to its theoretical value (=2) (
Figure 1, N
oM3). When 
 = 0.05
o, corresponding to a total number of integration steps (N
) of 3600, the numerical integration value of 
 is in the range [1.999–2.000] for small 
. For larger 
, it is in the range [1.800–1.999]. When the value of the numerical integration is in the range [1.800–1.999], a renormalization factor is applied to 
 to ensure that the result of the numerical integration is exactly 2. We could also increase the number of integration points, but it would increase the computation time. Using a renormalization factor for large 
 is a good compromise to guarantee the accuracy and save computation time.
For two-layered spheres (i.e., phytoplankton cells), the ScattnLay code provides only normalized phase functions for monodisperse particles (
Figure 1, N
oS1), so the numerical integration over the particle diameter range (Equation (
2)) is realized as a separate calculation with the Python “Trapz” function (
Figure 1, N
oS2). For monodisperse particles, the normalized phase function displays a forward peak as explained above but can also display a sequence of maxima and minima due to interference and resonance features [
26,
42]. The frequency of the maxima and minima over the range of 
 increases with both increasing 
 and size parameter (=
). To test the accuracy of the numerical integration over the particle diameter range (
Figure 1, N
oS3), we ran the ScattnLay code for DS1 case studies and compared 
 and 
 rebuilt from Equation (
2) with Lorentz-Mie computations as the Lorentz-Mie code provides the polydisperse phase functions and cross section as outputs (
Figure 1, N
oM1). Note that even a narrow polydispersion washes out the interference and resonance features, which explains why most natural particulate assemblages do not exhibit such patterns [
26,
42] (
Figure 3). A perfect match is obtained between the ScattnLay-rebuilt-polydisperse and Lorentz-Mie-polydisperse 
 and 
 values when the integration step (
) is set to 0.01 
m for 
D in the range [0.03, 2 
m]; 0.1 
m for 
D in the range [2, 20 
m]; 2.0 
m for 
D in the range [20, 200 
m]; and 10.0 
m for 
D in the range [200, 500 
m].
Using the DS1 data set, the impact of the angular integration on the backscattering ratio 
 is examined as a function of the hyperbolic slope 
 for different values of the real refractive index and two values of total angular steps (i.e., N
 = 750 and 3600) (
Figure 4). The impact of the integration is noticeable only for 
 values lower than about 3 and relatively high 
 values. When the number of angular steps increases, the curves become flatter at low 
 values. Differences in the curve shape are reduced if we increase the angular step. For 
 = 0.24
o (N
 = 750), the present results of the Lorentz-Mie calculations (solid lines in 
Figure 4) perfectly match those previously obtained by Twardowski et al. [
5] (not shown). However, in this case (N
 = 750), the numerical integration is not accurate enough as the integration of Equation (
1) gives values between 1.999 (
 = 4.9) and 1.04 (
). In the following, 
 is set to 0.05
o (N
 = 3600) and 
Figure 4 (dashed lines) will be the reference figure for homogeneous spheres.
  5.2. Impact of the Structural Heterogeneity of Phytoplankton Cells on the Bulk Particulate Backscattering Ratio
The impact of phytoplankton cell structural heterogeneity on 
 is examined as a function of 
 for the three previously described water bodies (oligotrophic-like, phytoplankton bloom, coastal-like) considering the 80%–20% phytoplankton morphological model (
Figure 5a). The real and imaginary bulk refractive indices for oligotrophic-like, phytoplankton bloom, and coastal-like case studies, vary with 
 as the relative proportions of the different particle components, having different 
 and 
, vary with 
 (
Table 1, 
Table 2, 
Table 3, 
Table 4 and 
Table 5 and 
Figure 5b,c). For the no-mineral water bodies (oligotrophic-like and phytoplankton bloom), 
 stays around 1.04 ± 0.007 (
Figure 5b). In contrast, 
 shows large variation with 
 for both oligotrophic-like and phytoplankton bloom water bodies (
Figure 5c). In bloom conditions, 
 increases as the relative proportion of phytoplankton increases as compared to the no bloom conditions. In agreement with typical values of the oceanic bulk imaginary refractive index [
43], the 
 values for the particulate populations considered here are always lower than 0.002. In the presence of mineral particles (coastal-like), 
 increases as MIN have a higher 
 than VIR, BAC, PHY and DET. Its values are between 1.103 (
) and 1.145 (
). Values of 
 vary between 9.79 × 10
 and 9.44 × 10
.
The impact of the structural heterogeneity of phytoplankton cells is evaluated by comparison with Lorentz-Mie calculations (particulate assemblages composed of homogeneous spheres only, regardless of the particle group) performed for low and high bulk refractive index. These case studies with homogeneous spheres only are named “Homogeneous reference cases”. The real and imaginary values of the bulk refractive indices are 1.045 and 9.93 × 10
 for the “Homogeneous reference case 1”, 1.043 and 1.36 × 10
 for the “Homogeneous reference case 2”, and 1.131 and 1.37 × 10
 for the “Homogeneous reference case 3”, respectively (
Figure 5b,c). These values of 
 and 
 were chosen to be equal to values of 
 and 
 obtained for the oligotrophic-like, phytoplankton bloom, and the coastal-like case study when 
 = 4. “Homogeneous reference cases 1 and 2” with low 
 represent phytoplankton-dominated Case 1 waters and are compared with the oligotrophic-like and phytoplankton bloom water body, respectively. “Homogeneous reference case 3” with high 
 represents mineral-dominated Case 2 waters and is compared with the coastal-like water body. The variation of 
 due to structural heterogeneity of phytoplankton cells is evaluated using the relative absolute difference calculated between the homogeneous reference cases (named 
x in Equation (
13)) and oligotrophic, phytoplankton bloom or coastal-like water bodies (named 
y in Equation (
13)):
		
Even if the numerical relative abundance of phytoplankton is very small for the oligotrophic-like water body (=8.4 × 10
%), the structural heterogeneity increases the 
 value by 58% compared to the homogeneous case (“Homogeneous reference case 1”). This is consistent with previous studies showing the large contribution of coated spheres to the backscattering signal [
15,
16,
18,
19,
22,
44]. The value of 
 calculated between the oligotrophic-like and phytoplankton bloom water bodies is smaller (=22% at 
 = 4) even if 
 is different (9.93 × 10
 for oligotrophic-like against 1.36 × 10
 for phytoplankton bloom). This latter pattern provides evidence that the structural heterogeneity (coated-sphere model) has a greater impact on the particulate backscattering ratio than the tested increase in the bulk imaginary refractive index. The relative absolute differences between the phytoplankton bloom and “Homogeneous reference case 2” is 41%. When mineral particles are taken into account, 
 is 12% between the “Homogeneous reference case 3” and the coastal-like water body. This smaller difference is because phytoplankton have a smaller impact on the bulk scattering when highly scattering particles such as minerals are added.
The impact of the relative volume of the cytoplasm on 
 is now evaluated by comparing the change of 
 as a function of 
 for the 80%–20% and 70%–30% models for the oligotrophic-like and phytoplankton bloom water bodies (
Figure 6). The mean relative difference in 
 is about 5.41% with a maximum value of 11.5 % (
 = 3) for oligotrophic-like case study. In bloom conditions, the mean relative difference reaches 13.0% with a maximum value of 23.5% (
 = 3.2). 
Figure 7 compares simulated 
 when phytoplankton cells are modeled as two-layered spheres (80%–20%) or three-layered spheres (80%–18.5%–1.5%). For the oligotrophic-like waters, relative absolute differences are small. They range between 0.0174% and 1.81% with a mean value of 0.444%. For phytoplankton bloom case study, they are between 9.84 × 10
 and 2.86% with a mean value of 0.894%.
Regardless of the morphological model used to optically simulate phytoplankton cells, the 
 reaches an asymptote when 
 decreases for phytoplankton bloom water bodies (
 = 0.005 for 
 = 3.5 and 0.004 for 
 = 2.5). The value of the asymptote is consistent with previous observations [
5], which showed the lowest backscattering ratio (about 0.005) in waters with high chlorophyll-a concentration.
The contribution of the different particle groups to the backscattering ratio is presented in 
Figure 8 for the 80%–20% model and 
 = 4. For coastal-like waters, the minerals contribute more than 80% of the total 
, whereas they contribute only 6.5% to the total particulate abundance. This percentage agrees with the results of Stramski et al. [
25] (Figure 12 in their paper). Such high contribution to backscattering is due to the high real refractive index of minerals. As in Stramski et al. [
25], these results show the important role of minerals even when they are less abundant than organic living and non-living particles. In oligotrophic-like waters, the contribution of heterotrophic bacteria ranges from a few to about 30% with a maximum for 
 between 3.5 and 4, which agrees with Stramski and Kiefer [
24]. The contribution of viruses is quite high, about 40–60% for 
 between 4 and 5. This high contribution is explained by the extreme value of viral abundance (around 1 × 10
 particles per cubic meter) used in this study [
24]. As for bacteria, the contribution of phytoplankton ranges between a few and 30% with a maximum around 
 = 3. For a 
 value of 4, typical of oligotrophic waters, the contribution is around 10%. For the phytoplankton bloom study case, the contribution of phytoplankton cells is between 10% and 60%; maximum values are reached for a PSD slope between 3 and 3.5. Such high percentages are due to the higher backscattering cross section of phytoplankton as compared to the other particles (
Figure 9). The low phytoplankton abundance is offset by the high 
 so that the backscattering coefficient of phytoplankton represents a significant contribution to the total backscattering.