Spatial Downscaling of NPP-VIIRS Nighttime Light Data Using Multiscale Geographically Weighted Regression and Multi-Source Variables

Round 1

Reviewer 1 Report (Previous Reviewer 3)

This is definitely a better version with some of my concerns addressed.  I thank the authors for their efforts.   Still, this paper is not ready for publication for a few reasons.

1). This paper is still riddled with grammatical errors.   Let us simply look at their abstract.  I do not know what “hinders finer studies” means.  What are “finer studies”?  “This study proposes”.  How can a study propose something?  It should be “we propose”. “The results suggest that the data resolution had been suitably improved after downscaling”.  What is data resolution? Is it spatial or temporal resolution?  What do the authors mean by “suitably improved”? “ the MGWR-based downscaling results produce higher R2 (R2=0.9141)”.   How can results produce something?  “Therefore, this study increases the spatial resolution of NPP-VIIRS NTL data from 500m to 120m by downscaling, thereby facilitating NTL-based applications.”.  Again, how can a study increase something?  Also, the authors seem to suggest that 500m resolution NPP-VIIRS NTL data can’t facilitate NTL-based applications?   The authors should also know that when you state a fact or describe an algorithm, you use present tense.  Those algorithms worked in the past, and they are still functioning, right?   Also, data shall be treated as a plural.  So “data is” shall be “data are”.  Since there are so many places that need to be revised for grammar related issues, I strongly recommend the authors to have this paper carefully proofread by a native English speaker. 

 

2). With the recently added descriptions, I finally know which VIIRS NTL data they are referring to (This is why the authors need to provide as detailed information as possible  so others can duplicate their results).  It looks like they used data from Dr. Elvidge’s group (instead of NASA Black Marble data).  Thus,  they shall remove citations for the NASA Black Marble data and add new citations.

 

Replace:

Román, M.O.; Wang, Z.; Sun, Q.; Kalb, V.; Miller, S.D.; Molthan, A.; Schultz, L.; Bell, J.; Stokes, E.C.; 750 Pandey, B.; et al. NASA's Black Marble nighttime lights product suite. Remote Sensing of Environment 2018, 210, 113-751 143, doi:https://doi.org/10.1016/j.rse.2018.03.017. 752

55. Wang, Z.; Román, M.O.; Kalb, V.L.; Miller, S.D.; Zhang, J.; Shrestha, R.M. Quantifying uncertainties in 753 nighttime light retrievals from Suomi-NPP and NOAA-20 VIIRS Day/Night Band data. Remote Sensing of Environment 754 2021, 263, 112557, doi:https://doi.org/10.1016/j.rse.2021.112557.

 

with

C. D. Elvidge, M. Zhizhin, T. Ghosh, F-C. Hsu, "Annual time series of global VIIRS nighttime lights derived from monthly averages: 2012 to 2019", Remote Sensing (In press)

C. D. Elvidge, K. E. Baugh, M. Zhizhin, and F.-C. Hsu, “Why VIIRS data are superior to DMSP for mapping nighttime lights,” Asia-Pacific Advanced Network 35, vol. 35, p. 62, 2013.

 

Also, the authors shall add more discussions on the data, such as spatial resolution (15 arc second) and the data format (GeoTIFF).  Also, I have no clue about this sentence:  “Here, we used the cloud-free observations file (band-2) for analysis.”  What do the authors mean by band 2?  VIIRS NTL data are derived from the VIIRS DNB.

 

3) I still do not know if the testing and training data are the same?  Can authors be more specific?

 

Other issues:

 

1). Table 1.  The revisit time for DMSP-OLS is not one year. Mentioning that the revisit time for the monthly cloud-free DNB composite as 1 month is misleading.  Also, “1months” shall be “1 month” and “1days” shall be “1 day”. 

 

2). Line 113 “good Spatio-temporal resolution”.  Define “good”.

 

3) Line 248, “irradiance” shall be “radiance”.  Irradiance refers to a hemispherically integrated value.

 

4). Lines 248-249, “However, the radiance of NPP-VIIRS uses the full band radiance expressed in” shall be “However, radiance value as reported in the NPP-VIIRS data is the spectral response function weighed value for the spectral range of 0.5-0.9 µm and is in the unit of “

 

5) line 299, “NPP-VIIRS NTL pictures”  what do authors mean by pictures?

 

 

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 2)

Compared to its last iteration, the improvement has not been substantial. I do see the authors' efforts in improving the quality of the manuscript, however, most my major concerns during the first round review have not been directly or adequately addressed. I would suggest that the manuscript has not been ready for publication yet.

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 3 Report (Previous Reviewer 1)

Orig reviewer 3, minor comments and the authors addressed them adequately. Andy Wheeler

Author Response

请参阅附件。

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report (Previous Reviewer 3)

This is definitely a better version with many grammar related issues fixed.  I recommend publication of the paper after fixing minor issues as listed below.

Table 1.  The revisit time for DMSP-OLS is one day, not 1 year.  If the authors refer to a specific product, then details need to be provided.

Page 5, line 161, [55,56] shall be [53, 55]???  [56] is not a reference for a product.

Page 15,”错误!未找到引用源。”  ??

Author Response

请参阅附件。

Author Response File: Author Response.docx

Reviewer 2 Report (Previous Reviewer 2)

The manuscript has been improved with my major concerns addressed. 

There were some non-English characters at lines 532, 538, and 555, please correct them.

Author Response

请参阅附件。

Author Response File: Author Response.docx

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

Paper is well written -- minor comments:

 - I can't find where you defined RF -> ?Random Forest? in the paper (can just say random forest in abstract and define there)

 - NTL/NLT acronymn are interchanged sometimes (so should always be NTL)

 - line 322, I think saying "uniformily spatially distributed" may be more specific to what you mean here.

About the only substantive comment, I believe the RF model should include X/Y coordinates as additional covariates. So if you were fitting a model for Beijing and say wanted to extrapolate to Guangzhao you wouldn't do this. But, you cannot do that type of extrapolation with the M/GWR models either (they are local slopes idiosyncratic to that data at a specific X/Y coordinate). They have to be a local spatial match. Because of that it is alright to include the XY coordinates into the RF model.

I don't think it is necessary for here, but an interesting future application could apply calibration to the MGWR model (can see the regression line in Figure 3 is not quite 1:1, I imagine a smart train/test framework though can find a mapping). The MGWR is clearly better calibrated out of the box than the other models though.

Andy Wheeler

Reviewer 2 Report

The authors achieved the downscaling of the VIIRS nighttime light data from 500 m to 120 m with information from other related variables. The method used in this study, the MGWR, is theoretically sound, which was evidenced from the comparison of its results with those from other two methods. However, the manuscript needs to be improved before its publication.

Major comments:

1.    “constant relational scale” is a key assumption in the downscaling application in this manuscript. The authors also acknowledged in the discussion that it has only been proved to be successfully applied to certain scenarios. So how can this assumption be evaluated or justified? Any quantitative measure for this purpose?

2.    The authors presented only one case study. Though its results look promising, it is not enough to prove the successfulness of the methodology. Is this case special? Or can this method be applied to other cases?

3.    Land Surface Temperature is one of the input variables for the downscaling, however the estimation method presented in this study (lines 204-211) seems problematic. How are the upwelling and downwelling radiance calculated? To be honest, these are the most challenging part of the proposed method, so that I have not seen such an algorithm has been implemented in any satellites’ operational LST generation.

There are quite many grammatical errors, along with missing white spaces here and there, no spell-out of abbreviations on their first usage.  I suggest the authors can pay special attention to these and have an extensive editing of the English language and style. The followings minor comments include some examples of such issues, there are way more of these in the manuscript.

28: improved of what and compared to what?

29: Please spell out RF

71: Replace “includes” with “include”

108: spell out POI

117: LUCC

144: More detailed figure caption is needed

152: the link provided is broken (white spaces inside a URL?)

243: please spell out GTWR and GWAR

248: variable variables?

258: remove “the” before identical.

325-326: This seems a tricky statement, since your conclusion is solely based on an assumption, which has not been verified in this manuscript at all.  Considering this is the key in your logic of the downscaling, I suggest the authors figure out a way to evaluate it.

 349: Based on (15), the numerator is the variance of the estimates, and the denominator is the variance of the truth. That is by no means a measurement of the accuracy of the estimates?

528: more specifically, the improvement is on its resolution but not the data quality.

Reviewer 3 Report

In this paper, the authors presented a method for statistically downscaling VIIRS NTL data.  This is a poorly written paper, although the idea of the paper may be interesting.  Many key details are missing and the study has flaws.  In addition, this paper needs serious text-editing efforts.  Considering the amount of effort it will take to revise the paper, I recommend the authors to work on the changes and resubmit the paper once it is truly ready.

 

Major issues:

 

Many needed details are missing from the study, including both data and methods for the study.  For example, LUCC data are used but are not explained in the dataset section.  The VIIRS NTL data are mentioned but are not explained in detail (e.g. how are the data created, what are the key points for the dataset?  References for the VIIRS NTL data are not included).  GWR and RF methods are used for inter-comparison with downscaled results from the MGWR method but are not explained in details in the study.  Ordinary kriging interpolation is used but is not explained in the method section.  Quadratic kernel function, AICc, and the golden ratio were used for searching optimal bandwidths; however, details are not provided.  The raw LuoJia -01 data are in 130 m spatial resolution.  The dataset was resampled to 120 m resolution.  How the data were resampled (nearest neighbor, bi-linear interpolation etc), however, are not mentioned.   Some parameters are mentioned, but the methods by which the values are set are not mentioned such as I↑ and I↓ from equation 4.  Those details need to be provided.

 

VIIRS NTL data are atmospheric and surface BRDF corrected data.  The downscaled VIIRS NTL data were inter-compared with LuoJia-01 data in this study.  It seems that the authors downloaded only LuoJia-01 raw data, so atmospheric correction must be implemented to exclude impacts from clouds and aerosols.  However, I do not observe such a step in this study.   At a bare minimum, LuoJia-01 data shall be cloud-cleared; however, such an effort is not presented.  Non-trivial uncertainty can exist from the non-cloud-cleared dataset.

 

Again, as mentioned above, details about their training and testing data are not provided.  The reviewer doesn’t know if the comparison of the two datasets is performed for a selected month or for the whole study period (06-12/2018).  The reviewer also doesn’t know if the testing and training data are for the same study period.   I would like see the application of the downscaling method to a different study period other than the training dataset time period (or apply to a different region) to further prove the developed system is robust.  Or, at least the authors should carefully explain the optimal bandwidth search/ kriging interpolation process to ensure LuoJia-01 data are not used in the process.

In addition, the paper needs serious text-editing efforts.  The reviewer tried to guess the meaning of some sentences.  For example, “1) The Earth Observation Group of the National Oceanic and Atmospheric Administration (NOAA) was used to obtain Version 1 of the NPP-VIIRS DNB cloud-free monthly 151 composite NTL product for June 2018 through December 2018”.  How could NOAA be used to obtain the data?  “A 30 m and 100 m spatial resolution is used”.  This sentence doesn’t make sense.   “The research goal of this paper is to draw NTL data with a spatial resolution of 120 m based on NPP-VIIRS NTL and other related data.”  What do the authors mean by “draw” data?  The reviewer is also confused by this sentence “We spatially summarised the POI densities of Luojia 01 and 30 m resolution in the study area as 482 600 m, 480 m, 360 m, 240 m, and 120 m.”  The reviewer is not sure what is meant by “spatially summarized”.  In short, the paper may need to be proof-read carefully, possibly by a native English speaker.

 

 

Other issues:

 

 

1.      Page 1, line 24, “ Loujia1-01” shall be “Luojia1-01”.  Same typo happened in a few other places.

2.      Page 2, line 108, define POI.

3.      Page 2, lines 128-129.  Need a reference.

4.      Page 4, Figure 1.  Define study period.

5.      Page 5, provide channels/wavelengths of LuoJia-01 and Landsat 8 data, as that information is used later in the discussions.

6.      Page 5, Lines 150-152.  Define VIIRS NTL data.  Note that the VIIRS NTL data used in the study is a product of the VIIRS DNB data.  Discussions need to be provided on how the data are created and some key scientific characteristics of the data.  Also references to the VIIRS NTL data need to be provided (e.g. Wang et al., 2021; Roman et al., 2018). 

 

Roman, M. O., and others (2018). NASA's Black Marble nighttime lights product suite. Remote Sens Environ, 210 , 113, https://doi.org/10.1016/J.RSE.2018.03.017 .

 

Wang, Z. and others (2021). Quantifying uncertainties in nighttime light retrievals from Suomi-NPP and NOAA-20 VIIRS day/night band data. Remote Sens Environ, 263 , 112557, doi: 10.1016/j.rse.2021.112557.

 

 

7.      Page 5, line 161, “Landsat 8 data can be downloaded”  I assume that the authors already downloaded the data?  So it shall be “Landsat 8 data were obtained”

8.      Page 5, line 166, “was less than 1.5%, to assure the data quality”  How did the authors get this number?   The authors need to either derive this number or provide a reference for this number. In either case, details need to be provided.

9.      Page 5, lines 167-169, what are the datasets downloaded from this site?  What are the properties of the datasets downloaded?

10.  Page 5, Table 2, the revisit time for VIIRS is not 30 days.  The monthly VIIRS NTL data are used in this study.  Please rework on the Table to avoid this confusion.

11.  Page 5, Table 5, explain LUCC and discuss the dataset in the dataset section.

12.  Page 6, line 186, “Data preprocessing: NDVI, NDBI, and LST were inverted from Landsat 8” shall be” Data preprocessing: NDVI, NDBI, and LST were derived from Landsat 8”

13.  Page 6, line 190, “was implemented to obtain NTL images with a resolution of 120 m.” shall be “was implemented to statically downscale NTL data to a spatial resolution of 120 m.”

14.  Page 6, lines 191-192, “Accuracy verification: The downscaled NTL results are compared with the LuoJia1-01 NTL to verify its accuracy” shall be “The accuracy of the downscaled NTL data was evaluated through inter-comparing with a much finer spatial resolution LuoJia1-01 NTL data”

15.  In line 198, “The near-infrared reflectance b5 to the red band b4”.  In line 201, “The ratio of the reflectance in the red band (b5) to the near-infrared reflectance (b6)”. The band b5 is defined as both red and near IR channels here.  Please revise.  Please add specific wavelengths.

16.  Page 6, equation 4, I↑ and I↓ values are not provided.

17.  Page 6, need references for equations that compute NDVI, NDBI, LST and other parameters.  If those are not invented by the authors, references are needed.

18.  Page, 7, line 277, “To make data storage simpler”.  How can you make the storage simpler?  I assume the authors intend to say “to reduce the data volume”?

19.  Page 7, line 232, I have no clue what “full-band radiation” is.

20.  Also, page 7, line 232, “the radiation output from NPP-VIIRS is transformed to a central wavelength and represented in nW/cm2/sr”  Equation 7 is for LuoJia1-01 data?????

21.  Page 7, line 235, “these two spacecraft” shall be “these two sensors”??   VIIRS and LuoJia-01 are not spacecraft.

22.  Page 8, lines 287-288, “These auxiliary variables are chosen by considering their correlation with NTL intensity and data accessibility.”   Please show the correlations.

23.  Page 12, line 421, “with those of Rojas”  What is Rojas?

24.  Figures 4 and 5, what are the study periods?

25.  Page 14, Table 5 cation.  Change from “Table 5. Difference between the regression relationships at various resolution scales” to “Table 5. Difference between the regression relationships of NTL and POI data at various resolution scales”

26.  Page 154, line 542, “for testing in th future”  please correct this typo.