Estimation and Evaluation of 15 Minute, 40 Meter Surface Upward Longwave Radiation Downscaled from the Geostationary FY-4B AGRI

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper proposed a method to downscale coarse spatial resolution SULR data to a finer resolution. Multiple Linear Regression (MLR) model and five surface factors are used to establish the relationship between SULR and related surface factors. A step-by-step downscaling strategy was applied to reach the 100-folds increase in spatial resolution, transitioning estimated SULR from 4 km of AGRI) onboard FY-4B satellite to 40 m of VIMI in infrared spectrum onboard GF5-02. Finally, the downscaling results were evaluated by comparing the downscaled SULR values with in-situ measured SULR and GF5-02 calculated SULR. It’s a good effort, but some questions still remain:

Minor comments:

1.       The manuscript used downscaling methods of LST on the downscaling of SULR. Are there any previous work on SULR downscaling?

2.       Because multiple linear regression model has not been used in the downscaling of SULF, how about the effectiveness of the five surface factors (MNDWI, NDBSI, NDVI, NMDI, UI) when constructing the regression model of SULR, and which is the most influential?

3.       In Figure 5(a-f), the label on the plot is not clear and the color bar of Figure 5(u-x) does not show the maximum and minimum values.

4.       In Figure 6(o)，the colors used to represent the points of each day are too similar.

5.       The x-axis of Figure 7(b3-d3) and Figure 4 (g2-g4) does not show the maximum and minimum values.

Comments on the Quality of English Language

The language need minor editing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The Surface Upward Longwave Radiation (SULR) is an important component of the surface net radiation. This study estimated SULR using geostationary FY-4B AGRI data, and then downscaled it from 4 km to 40m with the help of 40m-GF5-02 VIMI data, which provides 15-mintute, 40 m SULR products. The methods are robust and the results are reliable. Minor revisions should be conducted before been published.
1. Some details and discussions are missing. For example, the evaluation results for other resolutions in section 4.3, the comparisons with other SULR estimations in section 5, and the introduction about GF5-02 surface reflectance data in Table 2. See following items for more detail.
2. Line 25, it should be “Statistical”, not “Statisticalal”.
3. Line 39, should be plural: “Root Mean Square Errors (RMSEs)”.
4. Line 162-163, the reason why Fig. 2d is filtered is unclear. Simply state that three GF5-02 VIMI images were chosen based on data availability.
5. Fig.3, it should be “Aggregation”, not “Aggragation”.
6. Table 1, The symbol “°” is missing.
7. Table 2. The sources of GaoFen5-02 surface reflectance data should be stated.
8. Line 267, why simultaneously, do you mean “both”?
9. Section 4.3, Of which resolution was validated in this section, 40m, 200m or 1 km? What about the evaluation result of other resolutions? What about the evaluation results at every site?
10. Line 379 and Fig. 7, It should be root mean square difference (RMSD), but not “difference” or “RMSE”.

Comments on the Quality of English Language

No comments

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

 

The study entitled “Estimation and evaluation of 15-minute, 40-meter Surface Upward Longwave Radiation downscaled from geostationary FY-4B AGRI” presents a new methodology to downscaling satellite imagery to obtain high-spatial resolution values of Surface Upward Longwave Radiation (SULR). The method consists of an 4-steps process in which the coefficients of a MLR model are obtained, SURL values are re-calculated and corrected for each spatial resolution. The study area has been properly selected, including different surface types and land uses, and the validation process shows a detailed comparison with both in-situ measurements and other satellite measurements. The text is well organized and written.

I really appreciate the authors’ reasonable decision to apply a MLR model instead of a machine learning model knowing the limitations of their data set, as they discuss in lines 217-219: “Here, we chose MLR as the regression model considering that our study area consists of only 120 pixels (10 rows × 12 columns) at a 4 km scale which is insufficient for training a RF model“. This shows a great coherence and wisdom on the part of the authors.

I recommend this article for publication in Remote Sensing although some questions should be reviewed and improved. Next you can find the issues to be addressed before publication.

 

Comments:

Question 1.- In my opinion, the Abstract should be shorter. It should go directly to the objective and show the main contributions and results. In line with this, I think that the introduction given in lines 17-25 is not necessary. Such types of sentences are more suitable in the Introduction section.

Question 2.- Number of data employed for each part of the analysis should be clarified. In different parts of the text, different information about the data used are mentioned. Thus:

.- In line 146 four days are mentioned, the same that appear in Figure 2.

.- However, in line 165 only three days are mentioned.

.- Which data has been employed to obtain the RMSE values in Table1?

.- In line 338 authors say: “downscaling strategy was applied to other imaging time”. What time?

Maybe a table in Section 2 including for each instrument type and each part of the analysis the dates, number of images and/or grid cell numbers, could be useful to clarify this question.

Question 3.- Information about the in-situ radiometers (manufacturer, accuracy, …) installed in the 14 stations and the quality control applied to the experimental data should be included.

Question 4.- Why “July 24, 2022, September 27, 2022, November 17, 2022 and March 6, 2023” have been selected? What season and characteristics represent each of them? Why has more data not been employed? The selection of these dates should be stronger justified.

Question 5.- Figure 3 could be improved including the number of the equation employed in each step and using the same nomenclature as in equations 1-5, that is, TFY-48 SULR should be replaced by SULR_FY; Corrected SULR by SULR’, Calculated SULR by SULR_GF, etc. Additionally, if I properly understood the process, there is a mistake in “Fitted Coefficient” boxes where coefficients a0, a1,...(coefficients in Eq.1) appear instead of p0, p1,...(coefficients in Eq.3-4)

Question 6.- Be careful with nomenclature in Equations 1 and 2. Is band 12 in the same range of wavelengths for both satellites? If not, you should avoid using the same symbol R₁₂ in both equations. You can use R in one equation and R* or R’, for the other.

Question 7.- In lines 343-346 authors say: “As shown in Figure 5, the high-spatial-resolution data and high-temporal-resolution data was infused, downscaling coarse-resolution SULR from FY-4B geostationary satellite to fine-resolution align with GF5-02 satellite was achieved,...” However, at this point any comparison with respect to the GF5-02 has been performed and, therefore, this conclusion can not be included here.

Question 8.- In section 4.3 I miss the discussion of two important questions: 1) Respect which in-situ station the downscaled SULR show the highest differences? Are there any reasons to explain that?, and 2) In which time interval differences are higher? Why?

Question 9.- Figures:

9.1.- Check legends in figures (mainly scatter plots and daily plots) to not overlap the data represented in them.

9.2.- In Figure 5, different scale colors have been used for each 6-plots group. How convenient is this with respect to the use of the same scale color for all plots? I can imagine that using only one scale color could not allow to differentiate textures in some plots due to saturation but using different scales makes the plot comparison difficult. If possible I would like to see Figure 5 using the same scale color for all plots.

9.3.- Moreover, Figure 7 can be improved: 1) using the same color scale for a1-d2 plots and adding only one colorbar, 2) using a common y-axis for all a3-d3 plots and including the corresponding numbers only in a3 (similarly than in Figure 6), 3) include color bar only in d3 because it is the same for a3-d3, 4) remove legend “line 1:1” in a3-d3 or include it only in one plot, 5) because legend with RMSE, BIAS and R is too small remove them from the plot and include them in a new table

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf