This paper conducts research on a new temperature profile retrieval method using simulated data from China’s new-generation geostationary microwave satellite scheduled for launch. Channel optimization is performed based on information contribution, and an all-weather temperature profile retrieval algorithm is finally established. The proposed algorithm fully leverages the advantages of the geostationary microwave satellite and meets the requirements for rapid operational retrieval. Meanwhile, this paper intensively analyzes the improvement brought by the hyperspectral channels of the geostationary microwave satellite to temperature retrieval results, which provides valuable references for further instrument optimization research. The research outcomes possess strong practical value. Nevertheless, this study needs further elaboration on the variation characteristics of retrieval product accuracy under different weather systems. In addition, the current retrieval accuracy should be compared with existing operational retrieval methods to better clarify the application prospect of the proposed algorithm.

Specific comments：

Line 33-35: “The closer the frequency is to the center line of the oxygen absorption band, the stronger the absorption, and the higher the peak altitude of the weighting function.”

I fail to understand this sentence. Is there a direct correlation between the peak height of the weighting function and strong absorption?

All figures are relatively blurry, which affects the analysis of the results. It is recommended to adjust the figure types.

Line 135-137: “the Global Precipitation Measurement (GPM) Microwave Imager (GMI) Level 1C-R Version 05 TB product (Wentz and Draper, 2016) and GMI Level 2 retrieval product (Version 07) derived via the Goddard profiling algorithm (GPROF) (Kummerow et al., 2015).”

Specific descriptions should be provided to illustrate the applications of these two datasets; otherwise, the introduction of the two datasets will appear abrupt.

Line 158-159: “Sa and Sy denote the error covariance matrices of and, respectively, with off-diagonal elements set to zero.”

How are these errors defined in this study? Is it static or dynamically variable? The authors are suggested to discuss these issues.

Line 162-164: “In this study, the observation vector includes TBs from nearly all GMI channels, except for the horizontally polarized channels at 89 GHz and 166 GHz.”

Does this mean that these two channels will no longer be used in subsequent retrieval processes? If so, what are the reasons?

Figure 3: The interpretation of Figure 3 is unclear. Does the black straight line represent the mean profile? What do the error bars stand for?

Line 270: “with the findings of Aires et al. (Kangas et al., 2015)”

The citation format adopted in this manuscript is unconventional, and revision is recommended.

Line 339-340: “the temperature bias profiles for different channel configuration schemes are generally distributed around the zero line, with root mean square error (RMSE) below 1 K.”

If the study only focuses on typhoon-affected areas, can the temperature retrieval still achieve such accuracy? The authors are suggested to discuss these points to further clarify the application scenarios of the proposed method.

Line 346-347: “the RMSE values of NHC95 and NHC90 increase by approximately 19% and 47% relative to the full NHC set, respectively. Especially”

Does the value of the optimal channel selection method mainly lie in cloudy and rainy conditions? Is it because the eliminated channels would introduce more errors under cloudy and precipitation circumstances?

Figure 8: The biggest challenge of retrieval research based on simulated data lies in the inability to reasonably take instrumental errors into account. Will the retrieval errors in this study be affected by instrumental observation errors? The authors are suggested to discuss this issue.

General Assessment

This manuscript presents a timely and valuable contribution to the field of satellite remote sensing of thermodynamic profiles of the atmosphere, focusing on the upcoming FY-4M geostationary microwave satellite. The authors propose a complete retrieval algorithm with an information-content-based channel optimization scheme, specifically designed for all-weather atmospheric temperature profiling over the ocean. The work is methodologically sound, well-structured, and addresses a critical operational challenge, balancing retrieval accuracy with computational efficiency for a sensor with over 100 channels. The novelty lies in the comprehensive analysis of the FY-4M's unique hyperspectral temperature channels and several high-frequency channels, and in the comparative assessment of their performance under clear, cloudy, and precipitating conditions. The results are promising, showing that the optimized AC95 channel configuration can achieve a favorable balance. I recommend the manuscript for publication after the authors address the following  comments and suggestions.

Specific Comments

1) The sequential forward selection method in Sec. 2.3 is clearly described. However, the process of treating all hyperspectral channels as a superchannel (Section 3) is a simplification that merits more discussion. While it demonstrates the overall importance of this band, it does not provide a ranking of individual hyperspectral channels. This is a minor limitation, as the operational algorithm (AC95) includes the entire band. I suggest the authors briefly acknowledge this point in the discussion or conclusions, noting that future work might explore internal optimization within the hyperspectral set.

2）Treatment of observation error (sec. 2.3 & 2.4) needs some discussion. The authors state that Ratm​ and Rsurff​ are diagonal matrices, yet the final Rf​ is constructed using a Gaussian correlation matrix with a 1 GHz correlation length. This creates confusion for the reader. Are the forward model errors assumed to be correlated or uncorrelated? I suspect the authors' actual intent is that the diagonal matrices are intermediate constructs used solely to store the estimated error variances for each channel, while the 1 GHz correlation matrix is subsequently applied to impose spectral error correlations, yielding the final full covariance matrix Rf​. If this is the case, the manuscript would benefit from explicitly describing this two-step construction process to resolve the apparent contradiction. In addition, please clarify the basis for selecting 1 GHz as the correlation length. Is this value physically motivated (e.g., based on the spectral spacing of the channels or the oxygen absorption line width), or is it adopted from previous studies? A brief justification or reference would suffice.

3) The retrieval RMSE for the AC95 scheme during precipitation is reported as within 0.8 K, which is a very good result. However, the text mentions that retrieval errors increase noticeably due to complex scattering. The Bayesian algorithm relies on a comprehensive a priori database that includes simulated brightness temperatures from a plane-parallel MWRT model. This model has known limitations in representing 3D scattering effects. How might these model limitations affect the generalizability of the retrieval to more intense or convective precipitation systems that are not well represented in the training database? A short note on this would be valuable for the reader.

4) The 14% reduction in computational time for AC95 compared to the full AC configuration is a key finding. However, it is not entirely clear whether the training phase or the retrieval phase is the main cost driver. In operational scenarios, the retrieval is the time-critical step. Could the authors clarify which part of the algorithm benefits most from the channel reduction? This would help assess the real-world operational advantage.

Minor Technical Corrections:

1) The formula of Eq.4  is correct, but it would benefit from a definition of the integration time (t). Is this the dwell time per channel, or the overall integration time? I assume it is the dwell time, but clarifying this in the text would be helpful.

2) The caption of Fig. 7 says "Similar to Fig. 6, but including hyperspectral channels..." This could be more explicit. I suggest: "Similar to Fig. 6, but ranking all FY-4M channels, with the entire hyperspectral band (52.6-57.3 GHz) treated as a single predefined channel with the highest priority.

3) The English is generally fluent and readable. A few sentences are slightly convoluted; a minor stylistic edit by a native English speaker would further improve clarity.

As the world's first passive microwave instrument uploaded on Chinese geostationary meteorological satellite FY-4M, it will open a new era of continuous three-dimensional observations of various key geophysical parameters under all skies. In this study, channel optimization for atmospheric temperature profile retrieval over ocean was performed based on information entropy theory, and then the retrieval was done using a Bayesian algorithm. The topic of this paper has great practical value. Although this research is based on simulation research, it is urgently needed for China's meteorological satellite operations. This new PMW sounder adds some channels never be used before, especially the hyperspectral and Y1&2 bands is uniquely available on meteorological satellites. An original study on its contribution to temperature profile retrieval is conducted in this paper.

Overall, the paper is written smoothly, and the content is organized reasonably. It is recommended for acceptance after minor revisions. Specific suggestions and comments are as follows.

1 The emphasis in the title is on the all-sky atmospheric temperature profile retrieval over ocean.

2 In abstract: “several high-frequency channels in the high-precision detection of atmospheric temperature profiles over ocean”, gives the specific center frequency.

What is the ‘true value’ used for temperature profile validation?

3 In table 1: channel 5&6 the characteristics are the same. It is suggested to add one column to specify the purpose of each channel. Based on Table 1, provide the distribution of the weight function for all channels.

4 Line 138“… within the full-disk coverage of FY-4M, resulting in approximately 200 million collocated samples”, in this paper you just use the samples over ocean according Line 108-109, right? If it is true, the later prior probability distributions of Bayesian and so on will be explained over ocean. 

Is this database newly developed by this paper or already exist? The aim of section Line 151-169 is to adjust the retrieved GPROF hydrometeor products, if this work is not be done by this paper, suggest deletion。

5 Line 158: Sa is the error covariance matrices of Xa state vector, state variables are correlated, so it is a non-diagonal matrix with non-zero off-diagonal elements.

6 In Equ.(5) what is the meaning of A and B?

7 In Fig.4 the temperature Jacobian spectrum is calculated from just one sample under different weather conditions or the average of many samples?

8 Fig.8 is not clear, especially label of 90、95，AC.