Adapting an Existing Empirical Algorithm for Microwave Land Surface Temperature Retrieval in China for AMSR2 Data

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Is it possible to include any discussion about ecological applications of the algorithm?

Author Response

Is it possible to include any discussion about ecological applications of the algorithm?

 

Thanks for your suggestion. LST derived from the proposed algorithm has significant impact on the ecology of the Earth. LST is a reflection of the distribution of heat energy on the Earth’s surface and impacts various ecological elements such as atmosphere, soil, water, vegetation, animals, and microbe, etc. For example, LST affects the heat exchange near the surface, which can create different weather, change the form of water, promote or inhibit the growth of plant and microbe, and impact the activity of animals. We have added the relevant information in the revised manuscript in lines 31-37 of the revised manuscript.

Reviewer 2 Report (Previous Reviewer 3)

I do not have many comments on the article, the authors have treated the topics in a clear and detailed. it would be appropriate to specify the methodology used to conduct field observations such as context, altitude, and other morphological characteristics useful to describe the site.

Author Response

I do not have many comments on the article, the authors have treated the topics in a clear and detailed. it would be appropriate to specify the methodology used to conduct field observations such as context, altitude, and other morphological characteristics useful to describe the site.

 

Thanks for your suggestion. We have added the information describing the attributes of the sites that the reviewer mentioned in section 2.2.3.

Reviewer 3 Report (New Reviewer)

General comments: This paper describes a detailed production theory of long time series LST datasets, which is of high quality. The current written English seems good quality. However, the validation seems overly simplistic, and i think it can be improved. Also, several suggestions should be carefully considered as following.

Comment1: In the second step of Section 3.2, does the GlobeLand30 LC data in this study is updated annually in this work? MCD12Q1 is of low accuracy to some extend, while there are so many LC datasets with their desired accuracy, can the author tell your desired accuracy or your purpose, or the broader advantage of using GlobeLand30 LC,in this section?

Comment2: In Section 3.3, PTS and RMSE are never defined, formulas will be more suitable for reviewing and reading. Also, a flowchart is suggested to clearly describe your training model. The last sentence of this section is useless.

Comment3: Seriously, as far as I know, during climatic and meteorologic field, LST observation is likely taken as possible candidate to fill the spatial gaps in soil moisture observations, and the length of soil moisture data is long enough，what are the benefits of increasing LST length？And the authors should note this during your work to clarify your work’s novelty at least.

Comment4: Soil observation of TPDC has been used without proper quotation. Does this work is validated by using surface soil layer, what is the observation depth？More than 2000 soil stations including grass skin temperature observations have been established by CMA over years, and they might be deserved to be mentioned or discussed at least to guide future work.

Comment5: Is the validation metric using an overall R2 and RMSE on a spatial way or on a temporal way during section 4.2? Does this long sequence LST is obtained at the cost of sacrificing spatial accuracy? This requires at least one result to explain or discuss. Suggest reediting section 4.3 and give a spatially distributed difference/correlation estimation at least.

Comment6: Finally, I think Conclusion section should be before Discussion section.

 

 

 

It's well written,  but grammar issues still need careful check.

Author Response

General comments: This paper describes a detailed production theory of long time series LST datasets, which is of high quality. The current written English seems good quality. However, the validation seems overly simplistic, and i think it can be improved. Also, several suggestions should be carefully considered as following.

 

Thanks for your suggestion. We have revised the manuscript according to the reviewer’s comments.

 

Comment1: In the second step of Section 3.2, does the GlobeLand30 LC data in this study is updated annually in this work? MCD12Q1 is of low accuracy to some extend, while there are so many LC datasets with their desired accuracy, can the author tell your desired accuracy or your purpose, or the broader advantage of using GlobeLand30 LC, in this section?

 

Thanks for your question. The current version of GlobeLand30 LC data consists of three annual datasets including 2000, 2010 and 2020. Therefore, the annual GlobeLand30 dataset closest to the current year of AMSR2 was used to establish the annual CCSEV, based on the assumption that the interannual variation of LC among the adjacent years was negligible at the AMSR2 pixel scale. We mentioned this strategy in Section 2.2.1, where the datasets used in this study are described. The permanent snow and ice (glacier) in annual GlobeLand30 LC data is replaced by the glacier extent data obtained from the Randolph Glacier Inventory version 6.0. We regret that we didn’t clarify the way how GlobeLand30 LC data are used in Section 3.2, so we added the relevant description to make it easier to understand (lines 254-255 in the revised manuscript).

 

As we mentioned in the manuscript, the overall accuracy of GlobeLand30 LC data is 82.4% (line 151 in the revised manuscript), while the MODIS LC data (MCD12Q1) is 75% (line 251 in the revised manuscript). The MCD12Q1 data has misclassified some LC types especially in the transition zones between the bareland and vegetation in northwest China, while the GlobeLand30 data has eliminated this problem significantly. We have added the relevant information in section 3.2.

 

Comment2: In Section 3.3, PTS and RMSE are never defined, formulas will be more suitable for reviewing and reading. Also, a flowchart is suggested to clearly describe your training model. The last sentence of this section is useless.

 

Thanks for your suggestion. We provided the definition of PTS in Section 3.3 (lines 333-334 in the revised manuscript). RMSE is a commonly used indicator for accuracy evaluation. According to the suggestion of the reviewer, we have added the formulas of these two indicators.

 

We have also added a flowchart (Figure 4 in the revised manuscript) to describe the training model as the reviewer suggested.

 

The last sentence in Section 3.3 was added according to the suggestion by a reviewer in last reviewing stage. We have removed this sentence as the reviewer suggested.

 

Comment3: Seriously, as far as l know, during climatic and meteorologic field, LST observation is likely taken as possible candidate to fill the spatial gaps in soil moisture observations, and the length of soil moisture data is long enough, what are the benefits of increasing LST length? And the authors should note this during your work to clarify your work's novelty at least.

 

Thanks for your question. LST can be applied not only in fill the gaps in soil moisture observations, but the data foundation for the research related to climate change (e.g., ablation of glacier, urban heat island, etc.). As we mentioned in Introduction, the main purpose of the retrieval of long time series MW LST is to derive all-weather LST with high accuracy by fusing with TIR LST product (e.g., MODIS LST). Long-time series LST can be used to reappear the variation of the climate change-related parameters in the past and predict their trend in the future. The application of long-time series TIR LST product (e.g., MODIS LST since 2000) is limited by its defect of spatiotemporal discontinuities caused by cloud cover. The fusion of TIR and MW LST could solve this problem. To extend the time span of MW LST (e.g., AMSR-E and AMSR2 LST) and improve its accuracy is benefit the fusion of all-weather LST. That is the reason why we do this study. We have added the relevant description in line 88-89 of the revised manuscript to stress the purpose of long term LST datasets.

 

Comment4: Soil observation of TPDC has been used without proper quotation. Does this work is validated by using surface soil layer, what is the observation depth? More than 2000 soil stations including grass skin temperature observations have been established by CMA over years, and they might be deserved to be mentioned or discussed at least to guide future work.

 

Thanks for your question. We regret that we didn’t express the soil temperature clearly in the manuscript, the 1-5 cm in lines 191 and 194 actually means the depth, but we have forgot the word ‘deep’. We have revised the relevant expression to make it clear.

 

We have searched the official website of CMA and found the products of soil temperature derived by data assimilation at multi-depths. It is regrettable that we didn’t notice this kind of products before. We have added some discussion about them in the last section (lines 567-575 of the revised manuscript) to guide future work as the reviewer suggested.

 

Comment5: ls the validation metric using an overall R2 and RMSE on a spatial way or on a temporal way during section 4.2? Does this long sequence LST is obtained at the cost of sacrificing spatial accuracy? This requires at least one result to explain or discuss. Suggest reediting section 4.3 and give a spatially distributed difference/correlation estimation at least.

 

Thanks for your questions. The validation in Section 4.2 is on a temporal way. For the four sites of DDB, QYG, YMD, and Naqu, the field observations of the soil temperature are timeseries. The field observation timeseries of one site is compared with the timeseries of the AMSR2 LST pixel that covers the location of the site and shares the same observation time.

 

We regret that we may not fully catch the meaning of ‘long sequence LST is obtained at the cost of sacrificing spatial accuracy’. We considered that there is no need to balance the accuracy of LST between spatial scale and temporal scale. They are independent of each other. The field observations are only timeseries, and we are afraid that it cannot be used to verify the AMSR2 LST in spatial way. Therefore, it is hard to give a spatially distributed difference estimation.

 

Comment6: Finally, I think Conclusion section should be before Discussion section.

 

Thanks for your suggestion. As far as we know, most journals require that Discussion goes before Conclusions, including Remote Sensing. The template file that this journal provides specifies the order of the Discussion section and Conclusion section, and we complied with the requirement.

 

 

English language: lt's well written, but grammar issues still need careful check.

 

Thanks for your suggestion. We have checked throughout the text to correct the grammar issues.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Interesting text. I have no major reservations, only small comments. Include list of abbreviations for better clear reading.

Include an info about software which was used.

An electronic appendix with detail description of 62 zones (according to fig. 4) would be good. It is possible to compare these zones with biogeographical division of China?

Reviewer 2 Report

An optimized empirical algorithm for land surface temperature 2 retrieval from AMSR2 data in China

 

Abstract: Lines 19 to 21, explain the implications of these numbers. They are in range? They are logical numbers? Make a better connection to the field values, add some % of error or differences.

Lines 23-24, add numbers (%) for the accuracy in the different time of season.

Last sentences are not well supported by the data presented n the abstract. Please, improve the abstract content. A comparison with the previous methodology is welcome, why the new methodology is better?

Keywords: must be different from title, you must improve the words, some are useless.

Introduction: it is quite broad, and not totally justify the paper content. The objective is quite technical, and very brief. Please, justify better what you do, present a better objective linked to the paper title, and add some hypothesis or questions to be solved with this research. Why we need to optimize the old algorithm? Please justify better, the introduction must present the antecedents (from global to particular) and define the gaps at present.

Methods are not linked with the brief objective... in fact, there are a LOT of analyses that can not be supported by the brief objective presented in the introduction. Every method and analyses presented in the methods must answer to one general or specific objective or some question of the research.

Results and discussion must be presented in separate sections. In fact, results are look like as a technical report, and discussion are few. This is mainly due to the lack of precise and accurate objectives. The study by months or seasons must be included in the objectives.

Conclusions are not conclusions, are mostly results. The conclusions are the new knowledge generated by the research; they are few sentences derived from your results. However, your objectives are not well presented, so, the conclusions have the same weakness.

Reviewer 3 Report

This article is generally well structured and the general design is clear. However, It can be improved by the following comments.

It would be advisable to mention recent studies about LST algorithm retrieval (doi: https://doi.org/10.3390/rs12020294; https://doi.org/10.1016/j.rsase.2023.100921).

In the study area description, the climate classification of the area is missing. It would be also advisable to include information on its climate classification (for example on the basis of the Köppen-Geiger climate classification, doi:10.1127/0941-2948/2010/0430).

In addition, the references to some formulas are missing. Please provide appropriate references to formulas explained in the method section.

Reviewer 4 Report

The authors presented an empirical method to retrieval the land surface temperature retrieval from AMSR2. Given that methodologies presented is very similar to the author's previous work  for AMSR-E, it is basically considered a series of minor updates of the previous method using data from another sensor.  

The authors tried to evaluate the LST retrieval with soil temperature. This is not real validation, as the authors acknowledged their definitions are different. The authors applied calibration which decreased the RMSE by more than 10 degree. Since the difference is so significant, more details regarding the calibration need to be revealed.

The reference data uses MODIS data as a training data source, and the LST retrieval model is based on the MODIS data too. This seems not a fair evaluation.