Estimation of the All-Wave All-Sky Land Surface Daily Net Radiation at Mid-Low Latitudes from MODIS Data Based on ERA5 Constraints

Round 1

Reviewer 1 Report

Synopsis

 

This manuscript tested four RF-based models for Rn, especially in middle and low latitudes. After validation using four different Rn products, the authors conclude that one model outperforms. The value of adding some auxiliary atmospheric information to improve the accuracy of Rn is also discussed. The study is interesting, however, some important information is missing (list below).

Detailed Comments:

• Ln 17: “importance” should be “important”
• Ln 32: spell out “ERA5” and “FLUXCOM_RS” when they appear in the text for the first time.
• Ln 57: “Equation (1)” should be “Equation (1b)”.
• Ln 59: “due to” may be better than “resulting from”.
• Ln 61: could you please specify what kind of “various products”? model simulations or observations?
• Ln 61-62: could you please give some examples of “reanalysis products”?
• Ln 66: what about the accuracy of "remotely sensed products"? Are they better than the "reanalysis products"?
• Ln 73: please add references to “many methods”.
• Ln 74: what does “high-level” mean?
• Lns 76-78, “by considering … is more preferable” please add references.
• Ln 80: please add references to “RTMs”.
• Ln 160: missing “the” in front of “remainder”.
• Ln 167: “and” maybe better than “including”.
• Lns 175-177, “As Rs … used”: please rephrase this sentence. It’s hard to understand.
• 1: could you please add figures showing the pdf of, latitude, climate conditions, and land cover types?
• Lns 196-197, “after strick quality control … were selected.” please elaborate the "quality control" and "manual inspection". What kind of data was discarded? how to define "most reliable"?
• Ln 201, “over a site”: please clarify whether you mean within the 25kmx25km box? or just a point?
• Lns 202-203, “The Rn_daily…Jiang et al. [12].” To clarify, if there is missing for some hours within a day, that day will be discarded. right?
• Lns 203-206, “After matching… collected”: how many for each station? how many for each season?
• Lns 207, what does "similar distributions" mean? in what sense?
• Lns 207-210, “all instantaneous … as training samples”: Can this method guarantee that whenever a Rn_ins is used for training, the corresponding Rn_daily (meaning the daily mean for the day of Rn_ins) will be included in the training group as well?
• Ln 215, “25 km”: in line 144, it's said "31 km". Is this contradicting?
• Ln 219, “region”: centered at the site? or on a regular grid?
• Ln 219: please spell out the acronym when it appears for the first time. what's the resolution of this MCD12C1 data?
• Ln 222-223, “It can be seen … regions.”: please rephrase this sentence, it’s hard to understand.
• Ln 228: missing “of” in front of “less”.
• Ln 229: why not use NVDI from MODIS? which has a finer spatial resolution.
• Ln 232: “relative” should be “relatively”.
• Ln 233-234, “Overall, … spatial size.”: Are there any sites discarded due to inconsistency of the 25 km by 25 km box?
• 2: I have trouble understanding this figure. Suggest adding the x-axis title for each chart. "probability" may be a better word to replace "proportion". x-axis title of (a): percentage of the same land cover type within 25km by 25km box. Is NDVI globally static? Or it changes with year and season?
• Ln 237: missing “the” in front of “standard deviation”.
• Session 2.1.2: accuracy of CI is missing.
• Ln 244: what if there is only Rn for some sites as said in lines 174-175?
• Ln 260: please spell out “MOD” and “MYD” when they appear in the text for the first time.
• Ln 263: why not use the cloud mask to determine the cloud condition instead of CI?
• Fig 3.: please clarify what "frequency" refers to. Do you mean many observations in a day? If so, the color bar should not be continuous, since it should be 0, 1, 2, 3, and 4.
• Session 2.2.2: What are the differences between these products derived Rn_daily? Advantages and disadvantages when used for validation?
• Lns 278-279: could you please clarify what "satisfactory" means?
• Ln 284: could you please explain how you "extracted" the GLASS Rn_daily? spatial and temporal collocation.
• Ln 287: “has” should be “have”.
• Ln 292: could you please explain how you "extracted" the CERES4A Rn_daily? spatial and temporal collocation.
• Ln 304: “same” is redundant.
• Ln 307: in line 300, you mentioned that "the temporal resolution is eight days". now it's saying "daily". please clarify.
• Ln 308: could you please explain how you "extracted" the FLUXCOM_RS Rn_daily? spatial and temporal collocation.
• Ln 314: “have” should be “has”.
• Ln 317: missing “an” in front of “hourly”.
• Ln 320: could you please explain how you "extracted" the ERA5 Rn_daily? spatial and temporal collocation.
• Ln 322: what’s the meaning of “Cd”?
• Ln 337: “time” should be “times”.
• Lns 348-349, “the difference … to ERA5 data.”: please rephrase this sentence. hard to understand.
• Lns 353-355, “the final obtained … overpass times.”: please elaborate more on this.
• Session 3.1.1: what's the spatial and temporal collocation applied to get the MODIS data at the site for the instantaneous case?
• Lns 363-364, “the correlation ... is tigher.”: what does "correlation between ... and ... is tighter" mean? do you mean the correlation is "higher"? could you please quantify the correlation? whether it's statistically significant? especially for the middle-low latitudes that are the focus of the present study.
• Ln 366: missing “the” in front of “land”
• 5: what about SH? How did the authors calculate this correlation? What data was used? From where? Suggest adding a chart showing the data count corresponding to each panel.
• Ln 369: “divided” may be better than “aggregated”.
• Ln 370: “respectively” is redundant.
• Ln 373, “during the daytime”: do you mean "10:00 to 15:00 (local time)"?
• Ln 409, “after experiments”: please clarify what experiments were carried out? Usually, the original data will be grouped into "training", "test", and "validation". The "test" data were used for tuning the model parameter selections. The present study only separates the sample data into "training" and "validation" parts.
• Ln 422: “in” should be “on”.
• Ln 423, the description of the "experiments" is missing.
• Lns 426-427: why "linear or quadratic polynomial"? under what conditions the linear is used and under what conditions the quadratic polynomial is used? need to clarify.
• Lns 436-440, Model 2: in this case, should Local Time be considered as a factor to affect the model performance?
• Ln 438: “the” is missing in front of “output”.
• Lns 449-450: "simulated" by using what model? inputs?
• Ln 454: should “four” be “five”?
• Lns 456-459, “The cloud mask … or “uncertain clear.: why isn't cloud condition considered in the previous three models?
• Ln 461: “of” is missing in front of “the basic”.
• Ln 466 and 716: “number” should be “numbers”.
• Lns 474-475, “Most … to 70.” Please rephrase this sentence. It’s hard to understand.
• Session 4: Discussion regarding the computation time is missing. how much more time is consumed by adding ERA5 as a constraint?
• Ln 506: redundant “the” in front of “those”, missing “the” in front of “other”.
• Ln 537: “model” should be “models.
• Ln 538: missing “the” in front of “cloudy”, “deteriotated” should be “deteriorated”.
• Ln 565: “relative” should be “relatively”.
• 9: suggest improving the size of the charts. Meanwhile, what's the background color of the globe? Land cover type? Or a scatter plot showing the dependence of each parameter on latitude may be optional.
• Lns 624-625, “the newly … models.”: to support this statement, suggest adding a comparison similar to figs. 10, 11 and 12 using the other three models as well.
• Ln 639: “over” should be “in”
• Ln 641: “of” should be “in”
• Ln 642: “more” is redundant in front of “lower”, missing “the” in front of “other”
• Ln 666 and 735: missing “the” in front of “RF-based”.
• Ln 690 and 707: missing “the” in front of “daytime”.
• 14: suggest adding a figure showing the data count in the corresponding category. In the caption, “time” should be “times”
• Ln 694: missing “a” in front of “slight”
• Ln 695: “are” should be “is”.
• Lns 712-714, “it is recommended ... than two per day.”: could you please prove this by applying the RF-based model to high latitudes? will the RF-based adding ERA5 show better or worse performance?
• Caption of fig. 16: “representing” should be “represented”

 

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 2 Report

Review of the paper: “Estimation of the all-wave all sky land surface daily net radiation at mid-low latitudes from MODIS data based on ERA5 constraints”

In this paper a very interesting and important theme of daily net radiation at the land surfaces of different characteristics was analyzed.

The paper is based on a large amount of measured data taken from various projects across the Earth, in different climate zones. Methodology that combine “in situ” measurements and empirical models is promising and may lead to new results in estimating net radiation in research on Earth heat balance system, what is very important for study of climate change.

The construction of the paper is good and correct. The suggestions bellow have the purpose to contribute with the authors.

My general comments and suggestions are as follows:

1. Formulas (1). The source from which the formulas were taken must be stated. If it is not written, the authors of the paper derived formulas!?
2. Page 5. Chapter 2.1.1. I suggest the authors to present in this chapter a figure of the measured data e.g. R_n_daily, as shown in Figure 12, so that the reader is immediately acquainted with the way it changes in places with different climate, land cover and altitude.
3. Page 7. Line 253. It should be noted that the Terra and Aqua satellites belong to NASA!
4. Page 7. Figure 3. At the top of the figure it is better to write: DAY 2005/240. Frequency should be written on the scale shown on the right. Furthermore, in all figures where the scale is shown, the subject parameter and unit should be written!
5. Page 19. Figure 12. The x-axis shows the years, so it should be written!

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 3 Report

The paper is well written, technically correct, and I enjoyed reading it. The subject is of interest to a broad audience, and the methodology is forefront. However, I have some minor issues with the RF approach.

1. It is well known that RF cannot extrapolate outside the range of the training data set. Is that a problem with the current approach? May a few lines explaining the physical range of Rn could help and how it has been accommodated within the training data set could help better understand the performance of RF for their application.
2. The range of hyperparameters is detailed in Tab. 1. By the way, it could help to know the actual values of the hyperparameters for the final optimal RF.
3. It would also be interesting to know if the optimal RF has fewer (more) parameters than samples. That is, is the RF net under- or over-parameterized?

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Reviewer 4 Report

 This study proposes a versatile model for estimation of daily net radiation at the surface applying a supervised machine learning method, random forest. This model relates the net radiation to radiances of several channels of MODIS, as well as radiation simulated from ERA5. Estimation results of daily net radiation are validated using in situ measurement data at many observation sites on land. It is proved that the model based on ERA5 radiation as a constraint performs better than ERA only does, and is sufficient to simulate net radiation at a fine spatial resolution, especially at mid-low latitudes. 

 Estimation of radiation budget at the surface is required not only for investigation of meteoro/hydrological process but also for industrial development. Satellite observation data have been applied to global measurement and forecast of radiation, in order to cover areas where in situ observation data cannot be obtained. This study will help improve remote sensing technique for radiation observation and therefore is appropriate to this journal.  I feel that a little improvement of description may make more readable.  

(1) From a view point of remote sensing study, I have recognized that the comparison among ERA5 radiation and the RF-based model with/without ERA5 is very significant, because this could determine whether combining satellite data (e.g., MODIS) and ERA5 radiation is more effective than applying them individually. The comparison results are shown apart in the article, such as Figure 8, Figure 10, Figure 11, Table 6, Figure 16.  How about summarizing the comparion result (such as RMSE of global statistics) among the three into one table or figure to explicitly assert the superiority of this model?

(2) P5 L196: "strict quality control and manual inspection", What procedure has been actually implemented for quality control?

(3) P5 L204: "After matching with the MODIS observations and manually checking", how was MODIS data checked?

(4) P12 L423: "N has been defined as 1.7 in this study", how N has been determined?

(5) At Equation 6, I feel that the term "ERA5*" is confusing. Replace the other expression (How about Rn_ERA5?). 

(6) Figure 6, the appearance of this figure is not ease to see. Change the map background color to white, and slightly reduce the size of circle. If the authors intend to focus only on the dependence on latitude, 2-D scatter plot with the x-axis of Latitude and y-axis of RMSE (or rRMSE or bias) may be more suitable. 

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I thank the authors for responding to my initial review.

Some minor comments:

• Ln 169: missing “the” in front of “four”
• Ln 212: redundant “a” in front of “similar”
• Ln 295: redundant “a” in front of “better”
• Ln 381: missing “the” in front of “four”, “new” is redundant
• Ln 444: missing “the” in front of “values”
• Ln 466: “in” should be “on”
• Ln 474: “is” should be “are”
• Ln 507: missing “the” in front of “viewing”
• Ln 666: missing “the” in front of “other”
• Ln 808: “superiority” should be “superior”
• Ln 817: missing “the” in front of “daily”

Author Response

Please see the attachment. Thanks.

Author Response File: Author Response.docx