Stray Light Correction Algorithm for High Performance Optical Instruments: The Case of Metop-3MI
Round 1
Reviewer 1 Report
Please find my comments in the attachment.
Comments for author File: 
Comments.pdf
Author Response
Thank you for this very detailed review. It is particularly pleasant when reviewers pay so much attention to details. All the requested improvements, english and formatting, have now been made in the manuscript.
Lionel Clermont
Reviewer 2 Report
The manuscript by Clermont et al. presents a framework to correct for the straylight of the Metop-3MI instrument. The methodology includes spatial binning, field binning, and interpolation methods to populate the full straylight matrix from measurements at limited detector pixel positions. The interpolation methods in particular will be useful for straylight correction of other remote sensing instruments featuring 2D detector arrays and complicated straylight patterns. The manuscript is well written, technically sound, and can be recommended for publication.
My only general comment is to consider setting the stage clearer and early in the manuscript. It might be helpful to include a brief summary of the scientific objective and general design of the 3MI instrument, especially how these drive the straylight requirement. Moreover, the entire study appears to be based on simulations without any laboratory measurement data. It should be clarified what data (e.g., the SPST and FPST images) are simulated and what are measured. If there are laboratory measurements, the details in experimental setup should be provided.
Technical comments:
Line 43: “max” in Imax should be subscript.
Equation 2: I’m a bit confused about the boundary between nominal signal and SL signal. Since the PSF is sub-pixel, a no-SL input at (i,j) will be a delta function of one at (i,j) and zeros otherwise. Because the straylight SPST redistributes photons, the observed I_mes should have signal level of less than one at (i,j) and non-zero values otherwise. In that case, I_mes and I_SL are the same, and equation 2 should be removed. Please clarify.
Equation 9: does the power of p+1 mean each element in A_SL is raised to the power of p+1?
Line 160: “SL” in ASL should be subscript.
Lines 162 and 182: please correct the subsection numbers.
Line 205: “2” in (512/100)2 should be superscript.
Line 256: for field binning, although signals from the same pixel are averaged, they are from different collects and hence different realization of noise, aren’t they? Then the noise can still reduce with averaging.
Section 5.1: Have any calibrations been done for 3MI straylight? It would be useful to discuss the calibration techniques here.
Line 328: what is exactly the “higher resolution grid”? native (512 by 512)?
Line 378: how are those 4 neighbors used exactly? Weighted average?
Figure 11: there are two words “in” in the caption. Besides, please clarify what exactly “radial distance of pixel (x,y)” is, as (x,y) seems to be only a single detector pixel.
Lines 413-414: it would be useful to discuss the computing time needed to loop through the full A_SL matrix (N^2 by N^2) and whether/how this can be made feasible for operational use.
Line 455: should 102 and 104 be 1e2 and 1e4?
Line 457: just wanted to confirm that although the collimated beam covers the full entrance aperture (line 442), the nominal signal still stays within a pixel. For sensors with a slit, the nominal signal usually covers 2-3 pixels.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
- In Section 2.1, the first paragraph of page 4 of the manuscript, it is described that the limitations of the inversion process, such as error amplification. However, as mentioned in the introduction part of the manuscript, inversion matrix based SL correction has been applied to hyperspectral instruments. Is the iteration method suggested in this manuscript also promising for hyperspectral instrument? Also, did you compare the performance of iteration and matrix methods for Metop-3MI SL correction?
- The convergence speed can be increased with the Gauss Seidel or with the multigrid method. However there is any comparison between two methods. For a better understanding, please clarify the convergence speed depending on the method using the Gauss Siedel and multigrid.
- Typically, a smooth and high dynamic range PSF (point spread function) dataset is obtained after stitching process.
(image with the shortest exposure/highest level of filtering produce an unsaturated image with good SNR near the peak, but poor SNR at the wings.
(image with high exposure time/lowest level of filtering, resulting in an extremely over-saturated image around the peak, but with good SNR very far out in the wings)
Was the SPST measurement made with stitching process?
Minor corrections are required.
- page 3: 410nm to 410 nm
- page 5: ASL to ASL, Inom to Inom
- page 17: In the future to In the future
Author Response
Please see the attachment
Author Response File: Author Response.pdf
