Main comments

• The paper contains many interesting insights in an interesting topic. The paper however could be much improved in its presentation, as it is at several points unclear in its goals (I will explain below), and also contains many interesting results in the figures and tables that are not explained in the text. Also, the observational strategy is not generic, as it only applied to cumulus, it would be nice of some deeper reflection on other cloud types is added to the paper.
• The paper is in fact a two part paper, first it explains ideas and theory on how to observe the detailed spatial distribution of surface solar irradiance, and in the second part details about radiation under cumulus fields are explained. It got a little bit confused while reading the second part, as I expected there that an elaborate study would be presented on how to optimally observe radiation under cumulus clouds, but instead it presented a (very nice) sensitivity study to how the distributions of SSI change with cloud properties. I would like to ask the authors to rethink the exact purpose of this paper and to give the paper a clear and well-defined purpose. In my view, the best would be to focus the entire paper on the measurement strategy of SSI and create a dedicated study on SSI distributions and cloud properties, or to see how this information can be used to optimize measurement strategy.
• Table 2 contains many interesting insights in the impact of cloud properties on surface radiation, but only very few results are mentioned in the text. Please reflect deeper on this Table and the figures in the section, there is much more to learn from it.
• Probably my biggest concern on the usefulness of the results of this paper is the absence of aerosols in the analysis and the radiation computations. As, for instance, Gristey has shown in his papers, aerosols largely change the probability distributions of SSI, and the direct/diffuse partitioning. If this paper is aiming for designing optimal measurement strategies, but at the same time performs computations on a situation that will not occur in reality, how applicable are the strategies then? I suggest the authors include at least some computations on aerosols to show the sensitivity.

Detailed comments

• Why did the authors select cumulus clouds to do this study? Are they most relevant, or most appealing?
• In my view, the selection procedure is unnecessarily complex. Why did the authors choose for such a detail level in selecting the case study?
• The exact resolutions of the LES and ray tracer computations are a little unclear. It is stated that the RT is ran using 5x5m2 squares, but the LES is much coarser, so why is this high resolution used?
• How are the photon paths distributed over the different spectral computations? There are 15000 per 5x5m2 pixel, but is this per quadrature point, per band, or in total?