Spring Frost Damage to Tea Plants Can Be Identified with Daily Minimum Air Temperatures Estimated by MODIS Land Surface Temperature Products

Round 1

Reviewer 1 Report

The results presented seem promising.  However, considerable work on results and then of presentation so that the community may accept and adopt it perhaps is necessary. Thus, apart from the perfect introduction, the paper and the results have many shortcomings. I had to read it many times before finally understood how the study was constructed, what it was about, and even if eventually everything was correct and sound, or the study necessary.

Thus, the authors present a new method based on the lapse rate of MODIS  compared with a recognized and well-renowned RSDAST's method LST method. With it, they determined, apparently their true target, the frost days.

However, there are still many parts of the article which are confusing and obscure. In several aspects, because of the lack of information and figures (for example, DEM, land cover, maps of regional correlations), we get the impression of rushing and that parts of the analysis are missing or skipped and, therefore, incomplete.

In conclusion, the paper deserves a clearly defined presentation strategy for a workflow methodology. Results should follow a previously described order, and each must be correct and complete. This is not the case at present despite the promise of the results.

->

In what follows (called significant comments), general comments on the different sections of the article and what should be improved are provided. Then, in minor, we provide detail throughout the document. Still, although the text will probably change drastically, the latter may be useful.

A: Significant comments:

A1: Introduction section:

Perfect. Why not attach from 132 to 139 a detail of where all is done in the different sections?

A2: Study area and Data section.

Several essential Figures are missing that are necessary to the study: topography (DEM) and the land cover. Interestingly, the authors identified the tea planted area in figure 1-right.

A3: Methodology.

It looks correct, but it is not detailed. This section needs a general flow chart where everything, each section, and precisely presented are specified: a well-presented and implemented workflow.

3.1 and 3.2 should be a unique paragraph.

3.3 is obscure. It needs a drawing that explains the methodology's pipeline (and that is not the general flowchart). For example, look at figure 7 in  https://doi.org/10.3390/rs13040742 

3.4: Is this information going to be used? Where, how? Why is it not said explicitly?

In the end, It all seems so refined and subtle, but the concept is harmful if things are not right afterward.

A4: Results section (section 4).

In the results part, there are two parts. One part (sections 4.1 and 4.2) concerns the analysis of the RSDAST's method LST reconstructed data, and the second part (4.3 and 4.4) the study of the results of the new model (can we call it LR-RSDAST?).

The first part: 4.1 and 4.2 sections.

The authors of the RSDAST technique (Sun et al., 2017; https://doi.org/10.1016/j.cageo.2017.04.007) provide R2 values of the order of 0.9, RMSE of the order of 0.21. The authors of this article provide R's order values, i.e., an R2 of the order of 25 to 35% explained variance, three times lower! How is that possible?

In Figure 2, where global area averages are plotted for all the ground stations and RSDAST. The series is almost the same. How is it possible to have such a low correlation? Is there something wrong or just an impression? Furthermore, it is well known that a particular area may dominate the entire field's average with an overrepresentation of data, or the one with the largest thermal amplitude, thus not representing the whole area. So, are the normalized average also equivalent?

The authors must complement their study on RSDAST's method; put this research results in context. For example, the Sun et al. (2017) study used just three years of data (2000-2002). Is the data length used in the new study (16 years (2003-2018)  instead of 3) a correlation decreasing factor? What would have happened if this study investigated for fewer years (3)?

Or is this due to the region investigated? It is why a  DEM figure and another regarding the land cover in the data section are needed. Are these meaningful to the results found here and also the second part? Sun et al. (2017) provide the DEM, then the mean spatial distribution and standard deviation of the reconstructed LST time series and the local ground stations. It would be interesting to see it here also.

Thus, In that part, we would also have liked to see a comparison point by point (that is, RSDAST with the ground air temperature data) with the best method (3x3 or 5x5), showing sub-regions correlation. Can this help us better understand where and why the method fails? It is also super necessary in the second part, where the new technique is developed. Otherwise, how can one evaluate the whole study?

Finally, use modern statistical representations, like a boxplot of all the correlations, rather than a single value that does not represent al the variability behind. Besides, nowadays, everyone, including Sun et al. (2017), used R2. Why did the authors present R instead? It is not logical.

The second part: 4.3 and 4.4 sections.

It is THE essential part of the work. The results corresponding to the new, improved RSDAST method, say lapse rate RSDAST (can we call it LR-RSDAST?). Then, one hopes that these are shown accordingly to a presentation strategy, which precedes the results. Nope. This is not the case. It is impossible to follow this work easily.  For the reader to correctly follow the text and its development, the results' arrival should not be a surprise.  Sometimes, when starting the results section, some authors tell us that it will be presented first, second, third, etc. This is what we expect here to make it easier for us to read and understand the article.

Now concerning results. We need the authors to provide a map comparing the local ground stations' temperature with the new method product. Inspire yourselves on Sun et al. (2017).

Otherwise, it is not possible to know if, in all subregions, the method works the same. For example, group the ground stations by altitude bands and tell us how the R2 varies according to the altitude (not the R). Then, does land cover play a role? It is very briefly discussed in 5.1.2, but it is not enough. Of course, We do not want a new study taking into account the land cover, but by showing its influence.  Thus, does R2 varying according to landcover significant? Make this study a useful reference.

Finally, consider other statistical measures than the usual and well-known; correlation, RMSE, and MAE. Generally, R2 is preferred to R, standardized mean square error (nrmse) and percentage of the standardized systematic error (pbias)  are complementary to rmse (or mae..). For model evaluation, use Nash–Sutcliffe coefficient (nse). Thus, use these statistics also, as they will only allow a real comparison with other studies later on. For example, the study by Sun et al. (2017) reported values of R2 and RMSE, much lower than your study. However, are those comparable?

A5: Discussion.

Although it seems that all of this was built with the sole purpose of spring frost influence on tea plantations, the authors developed a new and straightforward method that can be easily also implemented elsewhere on earth. Could they again put the new method in context (as in the introduction words) concerning the other methods and how it could still be improved?

->

Minor comments:

23: Talking about correlation analysis is meaningless. It is a comparative analysis.

29: In fact, the comparison is made on 163 daily data ground stations over 16 years (2013-2018), not just 145211 observations.

30: As mentioned above, R2 is already used much more than R. And if the author's idea is that others can use the method, it is better to use standardized statistics.

46: a reference about the six well-known types of tea is missing.

64: a reference illustrating this part is welcomed.

69: Is there something more recent than 2013 ? Have a look at https://www.ipcc.ch/reports/

71: When citing, generally cite the oldest work first and then way up to the most recent.

158: ok, but can the authors tell us what exactly is the percentage of missing data? There has never been more than one day missing ? That would be quite surprising. Unfortunately, the page provided is in Chinese and I cannot verify each of the series. However, would it be possible to have more information on this subject, as it is essential for the study? For example, what happened if there was more than one contiguous missing day?

201: Cite Sun et al. (2017) after RSDAST.

207: It would be good to specify in which region Sun et al. (2017) did their study and in what it differs to the one investigated here.

211: It is "extracted," isn't it?.

232: Is the DEM averaged in the same way, 3x3, 5x5, 9x9? That is not clear. Moreover, the whole process is confusing. That is why section 3.3 requires a proper diagram to facilitate the understanding of the methodology. Would the authors also provide a Python program of the methodology?

235-236: It is not clear how the random values are obtained.

239: all section 3.4.

Why is this not said in the introduction? Why does this come only in the results part? Although very interesting, all this paragraph seems that it should be in the introduction part, or explained in the methodology part, but not here...!  Besides, one expects the authors to tell how it is going to be used. Do we have to assume how? Go straight to that.

253-256: This should also be in the methodology part.

257-259: It does not show anything because there is a table and not a figure.

280-283: Table 2.

Does the correlation vary spatially?  Why is it not shown? The R2 values are low. Where are they geographically acceptable? It is very low; RSDAST MOD11A R2 is 0.27, and MYD11A R2 is only 0.44, while the RMSE and MAE values are enormous. Why?

301: Sample number of what? The authors have to explain where the sample comes from. Why is it varying? The authors must not assume the lector understands and assumes everything. The authors should take the reader by the hand and guide him through results, and not the other way around; the reader is trying to figure out what the authors wanted to do.

382 and 383: Define better the station ratio in the figure caption, so the reader senses the correspondence between the figure caption and the figure.

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

I read the attempt to develop a simple empirical data based model to invert remotely sensed emitted electromagnetic radiation and explain the pattern and spatial distribution low surface temperature in middle and lower reaches of Yangtze River region, China with great interest.

This work has validated the observed low temperature of the model against the low temperatures of the meteorological data and found a strong relationship as evidenced by the 90% coefficient of determination. The model also simulated Temporal characteristics of low temperature in the region, especially for the cold periods and found the historically records of low temperatures. Finally, the model is applied to the larger spatial scale to see if frost damage over tea plantation is detected. Therefore, as far as I am concerned this work warrants publication in your journal. 

The only area I recommend authors to review is discussion and limitations section. While, the limitations made can stay, I advise the discussion reviewed because topics discussed here are new ideas, not part of the outstanding findings of the research. This section was to interpret the outcomes of this research in light of literatures in the area, significance in light of problems/opportunities necessitated the work and the hypothesis formulated. 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx