C-band Scatterometer (CScat): the first global long-term satellite radar backscatter data set with a C-band signal dynamic

Tao, Shengli; Ao, Zurui; Wigneron, Jean-Pierre; Saatchi, Sassan; Ciais, Philippe; Chave, Jérôme; Le Toan, Thuy; Frison, Pierre-Louis; Hu, Xiaomei; Chen, Chi; Fan, Lei; Wang, Mengjia; Zhu, Jiangling; Zhao, Xia; Li, Xiaojun; Liu, Xiangzhuo; Su, Yanjun; Hu, Tianyu; Guo, Qinghua; Wang, Zhiheng; Tang, Zhiyao; Liu, Yi; Fang, Jingyun

doi:https://doi.org/10.5194/essd-2022-264

Preprints

https://doi.org/10.5194/essd-2022-264

Preprints

15 Aug 2022

Status: a revised version of this preprint is currently under review for the journal ESSD.

C-band Scatterometer (CScat): the first global long-term satellite radar backscatter data set with a C-band signal dynamic

Shengli Tao¹, Zurui Ao², Jean-Pierre Wigneron³, Sassan Saatchi⁴, Philippe Ciais⁵, Jérôme Chave⁶, Thuy Le Toan⁷, Pierre-Louis Frison⁸, Xiaomei Hu¹, Chi Chen⁹, Lei Fan¹⁰, Mengjia Wang¹¹, Jiangling Zhu¹, Xia Zhao¹², Xiaojun Li³, Xiangzhuo Liu³, Yanjun Su¹², Tianyu Hu¹², Qinghua Guo¹³, Zhiheng Wang¹, Zhiyao Tang¹, Yi Liu¹⁴, and Jingyun Fang¹ Shengli Tao et al.

Show author details

¹Institute of Ecology, College of Urban and Environmental Sciences, and Key Laboratory for Earth Surface Processes of the Ministry of Education, Peking University, Beijing 100871, China
²Beidou Research Institute, Faculty of Engineering, South China Normal University, Foshan 528000, China
³ISPA, UMR 1391, Inrae Nouvelle-Aquitaine, Université de Bordeaux, Grande Ferrade, Villenave d’Ornon, France
⁴Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 91109, USA
⁵Laboratoire des Sciences du Climat et de l’Environnement/IPSL, CEA-CNRS-UVSQ, Université Paris Saclay, Gif-surYvette, France
⁶CNRS, Université Toulouse 3 Paul Sabatier, IRD, UMR 5174 Evolution et Diversité Biologique (EDB), 31062 Toulouse, France
⁷Centre d'Etudes Spatiales de la Biosphère, CNRS-CNES-UPS-IRD, Toulouse, France
⁸LaSTIG, Université Gustave Eiffel, ENSG, IGN, F-77420 Champs-sur-Marne, France
⁹Climate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720
¹⁰Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station, School of Geographical Sciences, Southwest University, Chongqing 400715, China
¹¹School of Geoscience and Technology, Zhengzhou University, 450001, China
¹²State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing, 100093, China
¹³Institute of Remote Sensing and Geographical Information Systems, School of Earth and Space Sciences, Peking University, Beijing 100871, China
¹⁴School of Civil and Environmental Engineering, University of New South Wales, Sydney NSW 2052, Australia

¹Institute of Ecology, College of Urban and Environmental Sciences, and Key Laboratory for Earth Surface Processes of the Ministry of Education, Peking University, Beijing 100871, China
²Beidou Research Institute, Faculty of Engineering, South China Normal University, Foshan 528000, China
³ISPA, UMR 1391, Inrae Nouvelle-Aquitaine, Université de Bordeaux, Grande Ferrade, Villenave d’Ornon, France
⁴Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, 91109, USA
⁵Laboratoire des Sciences du Climat et de l’Environnement/IPSL, CEA-CNRS-UVSQ, Université Paris Saclay, Gif-surYvette, France
⁶CNRS, Université Toulouse 3 Paul Sabatier, IRD, UMR 5174 Evolution et Diversité Biologique (EDB), 31062 Toulouse, France
⁷Centre d'Etudes Spatiales de la Biosphère, CNRS-CNES-UPS-IRD, Toulouse, France
⁸LaSTIG, Université Gustave Eiffel, ENSG, IGN, F-77420 Champs-sur-Marne, France
⁹Climate and Ecosystem Sciences Division, Lawrence Berkeley National Laboratory, Berkeley, CA 94720
¹⁰Chongqing Jinfo Mountain Karst Ecosystem National Observation and Research Station, School of Geographical Sciences, Southwest University, Chongqing 400715, China
¹¹School of Geoscience and Technology, Zhengzhou University, 450001, China
¹²State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing, 100093, China
¹³Institute of Remote Sensing and Geographical Information Systems, School of Earth and Space Sciences, Peking University, Beijing 100871, China
¹⁴School of Civil and Environmental Engineering, University of New South Wales, Sydney NSW 2052, Australia

Hide author details

Received: 01 Aug 2022 – Discussion started: 15 Aug 2022

Abstract. Satellite radar backscatter contains unique information on land surface moisture, vegetation features, and surface roughness, and can be acquired in all weather conditions, thus has been used in a range of earth science disciplines. However, there is no single global radar data set that spans more than two decades. This has limited the use of radar data for trend analysis over extended time intervals. We here provide the first long-term (since 1992), high resolution (~8.9 km) satellite radar backscatter data set over global land areas, the C-band Scatterometer (CScat) data set, by fusing signals from European Remote Sensing satellite (ERS, 1992–2001, C-band, 5.3 GHz), Quick Scatterometer (QSCAT, 1999–2009, Ku-band, 13.4 GHz), and the Advanced Scatterometer (ASCAT, since 2007, C-band, 5.255 GHz).

The six-year data gap between C-band ERS and ASCAT was filled out by modelling an equivalent C-band signal during 1999–2009 from Ku-band QSCAT signals and climatic information. Towards this purpose, we first rescaled the signals from different sensors, pixel by pixel, using a new signal rescaling method that is robust to limited overlapping observations among sensors. We then corrected the monthly signal differences between the C-band and the scaled Ku-band signals, by modelling the signal differences from climatic variables (i.e., monthly precipitation, skin temperature, and snow depth) using decision tree regression.

The quality of the merged radar signal was assessed by computing the Pearson r, Root Mean Square Error (RMSE), and relative RMSE (rRMSE) between the C-band and the corrected Ku-band signals in the overlapping years (1999–2001 and 2007–2009). We obtained high Pearson r values and low RMSE values at both the regional (r ≥ 0.93, RMSE ≤ 0.16, rRMSE ≤0.37) and pixel levels (median r across pixels ≥ 0.80, median RMSE ≤ 0.38, median rRMSE ≤ 0.64), suggesting high accuracy for the data merging procedure.

The merged radar signal was then validated with a continuous ERS-2 data set available between 1995 and 2011. ERS-2 stopped working in full mode after 2001 but observations are occasionally available for a subset of the pixels until 2011. Because the period of 1995–2011 fully overlaps with the working period of QSCAT (1999–2009), comparing the merged radar signal against the ERS-2 data in 1995–2011 is the most direct validation available. We found concordant monthly dynamics between the merged radar signals and the ERS-2 signals during 1995–2011, with Pearson r value ranging from 0.79 to 0.98 across regions. These results evidenced that our merged radar data have a consistent C-band signal dynamic.

The CScat data set (https://doi.org/10.6084/m9.figshare.20407857, Tao et al. 2022a) is expected to advance our understanding of the long-term changes in, e.g., global vegetation and soil moisture. The data set will be updated on a regular basis.

How to cite. Tao, S., Ao, Z., Wigneron, J.-P., Saatchi, S., Ciais, P., Chave, J., Le Toan, T., Frison, P.-L., Hu, X., Chen, C., Fan, L., Wang, M., Zhu, J., Zhao, X., Li, X., Liu, X., Su, Y., Hu, T., Guo, Q., Wang, Z., Tang, Z., Liu, Y., and Fang, J.: C-band Scatterometer (CScat): the first global long-term satellite radar backscatter data set with a C-band signal dynamic, Earth Syst. Sci. Data Discuss. [preprint], https://doi.org/10.5194/essd-2022-264, in review, 2022.

Download & links

Preprint (PDF, 5062 KB)

Supplement (878 KB)

Shengli Tao et al.

Status: final response (author comments only)

RC1:
'Comment on essd-2022-264', Anonymous Referee #1, 12 Sep 2022

The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2022-264/essd-2022-264-RC1-supplement.pdf

Citation: https://doi.org/10.5194/essd-2022-264-RC1
- AC1: 'Reply on RC1', Shengli Tao, 14 Sep 2022
  
  This work developed a monthly global long-term satellite radar C-band backscatter data set (CScat) by fusion of ERS-1(C-band), QSCAT (Ku-band) and ASCAT(C-band) observations using a new rescaling method. Maybe the CScat data set has useful in analysis and understanding of some global surface parameters (e.g., vegetation and soil moisture). But the temporal resolution is little low. And, there are some main problems of this manuscript:
  Response: We thank Referee #1 for reviewing our manuscript. All the received comments have been carefully considered. The review is overall beneficial for improving our research, and we very much appreciate this chance to discuss with Referee #1 on some potentially unclear parts of our manuscript. Below please find a point-by-point reply. A full revision will be available once comments from all referees are received.
  
  1) The signals of Ku-band (13.4GHz) and C-band (5.3GHz) microwave is different. Theoretically, comparing the Ku-band, the X-band and C-band have more similar frequency. Authors choose the Ku-band to fill up the six-year gap of the C-band scatterometer, not choose the X-band, L-band. It is no reasonable explanation here. In addition, authors did not choose data of the same C-band satellite radar data for fusion. It is better using same C-band radar data for fusion. For example, ERS-1/2, ASCAT, Sentinel-1 and GF-3 et al. The results of microwave data merging using the same microwave C-band have greater application significance compared with different microwave bands.
  Response: Thanks. We fully agree that Ku-band and C-band signal dynamics are different, but we believe this is exactly why our research is potentially valuable: we successfully developed an approach to adjust the Ku-band signals into C-band signal dynamics.
  Referee #1 asked why X-band or C-band data were not used for filling the six-year (2001-2007) data gap between ERS and ASCAT. We appreciate that this point might have not been made clear enough in the manuscript, thus will address it in full during revision.
  The answer to Referee #1’s question is: as far as we know, there is no such data at the global scale. Referee #1 first mentioned X-band data. However, the only X-band sensor covering the entire period of 2001-2007 is TRMM TMI. Unfortunately, TMI is only available for tropical regions. Since we aimed at producing a global dataset, TMI was not used.
  Referee #1 then mentioned C-band Sentinel-1 and GF-3. However, Sentinel-1 and GF-3 are available since 2014 and 2016, respectively, thus cannot be used to bridge the data gap of 2001-2007.
  
  2) For the developed new rescaling method, the comparison analysis in Figure 3 is not enough with CDF method in only two sites. And, Is the new rescaling method developed by authors only applicable to Ku-band correction? Can X-band and L-band data also be fused with C-band using this new rescaling method？
  Response: Indeed, we showed only two examples in Fig. 3. We will provide more examples during revision to address this concern. Meanwhile, we emphasize that we have placed an utmost emphasis on transparency, including making all the results fully available for anyone to inspect the merged signals.
  The new scaling method relied only on the mean and the std of the overlapping observations between two microwave sensors. It’s therefore theoretically applicable to other microwave sensors as long as overlapping observations are presented. Because our research deals with C-band and Ku-band signals, we tested the new scaling method on these two kinds of signals. Referee #1 mentioned X-band and L-band data but didn’t specify the name of such data. We will be glad to test further our method if Referee #1 could provide more details here.
  
  3) I think the validation of CScat data set is not sufficient if authors only used ERS-2 data as validation data for CScat. I suggested that the authors consider using the C-band observation data of airborne or other satellite/sensor different ERS-1/2 as comparison data. And, I doubt the reliability of the validation results of CScat data set. Authors used the ERS-1 observation radar signals to correct the Ku-band signals of QSCAT, and used the ERS-2 signals to validate the corrected Ku-band data. Because the satellite parameters and sensor parameters of ERS-1 and ERS-2 are the quite same, the observation radar signals of ERS-1 and ERS-2 are very similar at the same place and time. This may be the reason for the very high correlation coefficient in Figure 9.
  Response: We believe there is a misunderstanding of our method here, thus respectfully disagree. Referee #1’s statement, “Authors used the ERS-1 observation radar signals to correct the Ku-band signals of QSCAT”, is unfortunately incorrect: we did not use ERS-1 to correct the Ku-band signals.
  Instead, we used C-band ERS-2 (1996-2001) and ASCAT (2007-2020) to adjust the Ku-band QSCAT (1999-2009) into C-band signal dynamics, based on overlapping observations in the years of 1999-2001 (between ERS-2 and QSCAT) and 2007-2009 (between ASCAT and QSCAT).
  To check whether Ku-band QSCAT signals have been well adjusted into C-band dynamics, the best validation data should be a continuous C-band time series extending through our study period---This is exactly what we did with Fig. 9: although ERS-2 stopped working in full mode after 2001, observations are occasionally available for a subset of global pixels until 2011. Comparing our merged radar signal against this long-term but spatially incomplete ERS-2 dataset is the strictest validation we can perform.
  Referee #1 suggested “…the authors consider using the C-band observation data of airborne or other satellite/sensor different ERS-1/2 as comparison data”. Again, we will be glad to perform the validation if Referee #1 can specify where to obtain such validation data.
  
  4) The English language of manuscript needs to be polished. The abstract of this manuscript is too long. For the introduction of this manuscript, the research background for active microwave fusion or rescaling study is not sufficient. In 110 lines, is there any other studies that show that the Ku-band QSCAT signal can be adjusted to the ERS observations except the author's own research (i.e., Tao et al.,2002b)? I suggest that the abstract and introduction of this manuscript need to be rewritten.
  Response: Thank you. Regarding English writing, we would appreciate if Referee #1 can point out directly which sentences are unclear. We will address them during revision.
  Referee #1 pointed out that the abstract is too long. Indeed, the abstract is longer than requested by other journals. However, before submitting our paper, we double-checked the submission guideline of ESSD, and found no limitation on the length of the abstract. We therefore choose to write a relatively long abstract to make our methods and results as clear as possible. Per suggestion of Referee #1, we will try to shorten the abstract.
  Regarding the Introduction, we appreciate the suggestion that more background for fusing active microwave data is needed, and we will revise accordingly. We would also like to mention that, as far as we know, our research is the first to merge active microwave signals at the global scale. We have successfully merged signals from the same sensors (ERS, QSCAT, and ASCAT) for global tropics (Tao et al. 2020b). Referee #1 questioned “is there any other studies that show that the Ku-band QSCAT signal can be adjusted to the ERS observations except the author's own research (i.e., Tao et al.,2002b)? ” We appreciate this question, the answer to which is yes: recently, Frolking et al. (2022a & b) have been published which merged signals from exactly the same sensors but for global metropolis. Their research therefore confirms that QSCAT is one of the best options for gap-filling the six-year data between the ERS and ASCAT. We will revise our manuscript to make sure Frolking et al. (2022 a & b) are referred.
  Frolking, S., Milliman, T., Mahtta, R., Paget, A., Long, D. G., & Seto, K. C. (2022a). A global urban microwave backscatter time series data set for 1993–2020 using ERS, QuikSCAT, and ASCAT data. Scientific Data, 9(1), 1-12.
  Frolking, S., Mahtta, R., Milliman, T., & Seto, K. C. (2022b). Three decades of global trends in urban microwave backscatter, building volume and city GDP. Remote Sensing of Environment, 281, 113225.
  
  Above-mentioned, I am in a difficult position to reject the manuscript for publication
  Response: Once again, we thank Referee #1 for commenting on our manuscript. We hope our reply have clarified the unclear points of our manuscript, as well as the potential misunderstandings. Hopefully, we will have the chance to prepare a fully revised manuscript to further exchange with Referee #1.
  
  Citation: https://doi.org/10.5194/essd-2022-264-AC1
RC2:
'Comment on essd-2022-264', Anonymous Referee #2, 04 Oct 2022

This work extended the C-band data set to the previously missing period by rescaling the QSCAT Ku-band dataset during 2001-2007. Data-rescaling was used to unify the backscatter values from different sensors and then the machine learning method was used to address the monthly values differences. This is a quite useful dataset for further detecting forest structure and resilience dynamics. I have some minor comments as below:

To compare the linear regression, CDF and new data rescaling method, the author should compare their performances at global scale, i.e. a map showing the pearson r and RMSE pixel by pixel.

Similar for Fig 4, the author can show the spatial map of the performance of scaled Ku-band and corrected Ku-band pixel by pixel.

It seems that such rescaling method can also apply to other merging tasks. Can you discuss a bit of its potential usage to benefit the big data environmental science field?

The author could include a table mentioning the specific information of available microwave dataset, i.e. their time and spatial coverage, time and spatial resolution, etc, to prove the uniqueness of constructing the time series over non-overlapped period with QSCAT Ku-band data.

As you used the decision tree regression, have you checked whether the over-fit issue exist or not?

For Fig 8, there is large overlap between type 1 and type 2 pixels. If the author just compared the corresponding pearson r values between corrected Ku-band and C-band values to find the appropriate regressor, the type of each pixel can be determined. Why are some pixels assigned by two type?

Citation: https://doi.org/10.5194/essd-2022-264-RC2
- AC2: 'Reply on RC2', Shengli Tao, 06 Oct 2022
  
  Thank you greatly for your very helpful comments. We will make sure to address them in full during revision.
  
  Citation: https://doi.org/10.5194/essd-2022-264-AC2
RC3: 'Comment on essd-2022-264', Anonymous Referee #3, 25 Nov 2022

In this study the authors presented a monthly global C-band backscatter data record by combining ERS (C-band), QScat (Ku-band) and ASCAT (C-band) data for the time period 1992-2021. QScat data has been used to fill a six year gap between the C-band backscatter datasets (1999-2009). For this reason the Ku-band dataset has been rescaled using the overlapping period with ASCAT (2007-2009). The presented rescaling method was found to be robust to both signal trends and sudden changes. Monthly signal differences have been corrected after rescaling based on a decision tree regression. ERA5-land data (monthly rainfall, snow depth and skin temperature) was used to model signal differences in C- and Ku-band. Two types of quality assessments have been carried out. The first one is based on a comparison between the C-band and scaled Ku-band signal on a pixel by pixel bases reporting the distribution of Pearson R, RMSE and rRMSE for the periods 1999-2001 and 2007-2009 before and after the monthly signal correction for 13 regions. The second quality check is using ERS-2 data for the time period 2001-2011 reporting Pearson R for 10 regions. The results overall show that the rescaling and correction method are doing reasonably job fitting the Ku-band data in the C-band data space generating a homogeneous dataset.

Major comments:

1. While it is clear that this "C-band" dataset is one of its kind, I doubt the novelty of the presented "new data scaling method". It is a simple mean-std rescaling and part of "standard data rescaling techniques". See e.g. 10.1201/b15610-21, 10.1016/j.rse.2008.11.011, 10.5194/hess-14-1881-2010

2. I can see the importance of long-term C-band radar data, but a monthly temporal resolution is a big disadvantage and perhaps a no-go criteria for certain applications. The study doesn't explain why this temporal resolution has been chosen in the first place and is also not discussed in chaper 4.3. What is the reason? Would it be possible to get a 14-day, 10-day or lower temporal resolution? Please discuss possible applications and limitations of monthly C-band radar data. E.g. how is it possible to describe/separate vegetation and soil moisture (trends), also taking long-term land cover changes into consideration?

3. The manuscript is missing essential background information on decision tree regression. The authors describe that they performed three separate regressions (against monthly rainfall, snow depth and skin temperature) and used MSD to decide on the optimal regression. The term "decision tree regression" is far-fetched and not correct in this context. A decision tree regression would separate the feature space using nodes/leaves thereby selecting the optimal regression/parameter. See e.g. 10.1007/s00704-019-03048-8

Minor comments:

- Title: It is a "C-band" dataset so it should certainly have a C-band signal

dynamic. I'd suggest to highlight the fact that a Ku-band dataset is

used to fill a gap and create a long-term "C-band" data set.

- p2 - l38: remove "and can be acquired in all weather conditions"

- p2 - l53-54: No unit for RMSE/rRMSE in abstract, is it dB? Also missing in the rest of the article and graphics

- p4 - l102: Metop-SG

- p5 - l131: Please add references

- p7 - l180: wording

- Figure 4: remove connection of Ku-band time series for the temporal break

- Figure 8: why is there an overlap? shouldn't it be one map indicating type 1,2,3?

- Figure 9: why two different y-axis?

- p22 - l450: typo

On this basis, I found the topic of the paper interesting, but I suggest a major revision and after that reconsider a possible publication.

Citation: https://doi.org/10.5194/essd-2022-264-RC3

Shengli Tao et al.

Supplement

https://doi.org/10.5194/essd-2022-264-supplement

Data sets

CScat Shengli Tao https://doi.org/10.6084/m9.figshare.20407857

Shengli Tao et al.

Viewed

Total article views: 795 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
603	176	16	795	45	11	7

HTML: 603
PDF: 176
XML: 16
Total: 795
Supplement: 45
BibTeX: 11
EndNote: 7

Views and downloads (calculated since 15 Aug 2022)

Month	HTML	PDF	XML	Total
Aug 2022	221	67	6	294
Sep 2022	140	37	3	180
Oct 2022	113	31	4	148
Nov 2022	74	18	3	95
Dec 2022	40	16	0	56
Jan 2023	15	7	0	22

Cumulative views and downloads (calculated since 15 Aug 2022)

Month	HTML	PDF	XML	Total
Aug 2022	221	67	6	294
Sep 2022	140	37	3	180
Oct 2022	113	31	4	148
Nov 2022	74	18	3	95
Dec 2022	40	16	0	56
Jan 2023	15	7	0	22

Viewed (geographical distribution)

Total article views: 742 (including HTML, PDF, and XML) Thereof 742 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 28 Jan 2023

Short summary

We provide the first long-term (since 1992), high resolution (8.9 km) satellite radar backscatter data set with a C-band (5.3 GHz) signal dynamic for global lands, the C-band Scatterometer (CScat) data set. CScat was created by fusing signals from ERS (1992–2001, C-band), QSCAT (1999–2009, Ku-band, 13.4 GHz), and ASCAT (since 2007, C-band). CScat has been validated against independent ERS-2 signals. It could be used in a variety of studies such as vegetation monitoring and hydrological modeling.


Total:	0
HTML:	0
PDF:	0
XML:	0