See attached pdf.

The manuscript describes an algorithm to derive cloud optical thickness, particles effective radius, and cloud top effective temperature using EarthCARE’s multi-spectral imager (MSI). Cloud water path is also derived using optical thickness and effective radius. The retrieval uses 7 channels of MSI. The retrieval is done though a forward model in an iterative way to minimize a const function. The const function is the sum of the difference of modeled radiances and observed radiances and retrieval cloud properties and their a priori estimates. The const function takes into account of measurement uncertainty based on signal-to-noise ratio and forward model errors. The forward model errors are assumed to be independent. The algorithm is tested with EarthCARE simulator HALIFAX scene and MODIS data. Also, M-CLD properties are validated within the framework of the CGMS internatonal cloud working group.

The manuscript is well written and results are easy to understand. I only have minor comments or clarifications.

Equation 1. Is this for all 0.67, 0.865, 1.65, and 2.21 channels?

Line 141. What is theta_c?

Line 242. S changes weights to sum up the elements to compute the cost function. How do you estimate diagonal term of Sa? Does this covariance matrix depend on region or cloud type? If variances are fixed, could you present the size of variances in a table? This is nice, in principle, if we know the covariance matrix. But in practice, we do not really know Sa. How sensitive the final solution is to the covariance matrix?

Table 1. Change “cloud particle size” to “cloud particle radius”.

Figures. Generally, axis labels and legends are too small. For example, Figure 2 (legend), Figures 3, 4, 5 and 6 (labels for the color bar).

Figure 7 and 8 do not mean very much to readers who are not a member of the cloud working group unless all title or legends are explained.

Figure 9. Could you provide mean difference and RMS difference between MODIS M-COP values in a table? It is even better If the authors have compared more scenes and provide robust statistics.