Although, the paper has been improved following the first round of comments, there are still issues that must be addressed before the paper can be accepted for publication.

Please consider the following recommendations:

1. Retrieve the 75 sample filters collected in the original experiment, and measure the reflectance and transmittance of each one on both sides: Rc and Rs, and Tc and Ts. If this effort would be difficult or uninformative for some reason, note such constraints in the paper to provide context for the choice of experimental design.
2. Then replot Fig. S5 using these measured values rather the two-stream model outputs for
a. ODs = ln((1-Rs)/Ts),
b. ODc = ln((1-Rc)/Ts), and
c. ODimprove = ln((1-Rc)/Tc).
3. Discuss how scattering by the filter substrate can yield Ts > Tc for measured values, even though equality is implied by the two-stream approximation.
4. Explain what value was used for Tb when modeling ATN = ln(Tb/Ts) in Fig. S5. Was it the mean or median measured over a set of 20 blanks? Pre-sampling measurements on each of the 75 actual sample filters?
5. Discuss the variation of Tb observed in your individual blank filters, and compare it with that reported by White et al. (2016, Fig. 1) and Presler-Jur et al. (2017, Fig. 4).
6. Calculate and plot the standard deviation to mean ratio of modeled filter OD measures over the observed range of Tb and Rb values, versus AOD. This is the analog to Fig. 3, looking at sensitivity to blank optics in place of sensitivity to SSA.