The SARSense Campaign: Air- and Space-Borne C- and L-Band SAR for the Analysis of Soil and Plant Parameters in Agriculture

Round 1

Reviewer 1 Report

I have completed the review of this paper on the SARSense campaign. The design of the campaign is very good. It is disappointing that the corner reflectors were not aligned resulting in the loss of ability for absolute calibration of the airborne data. However, the paper presents the analysis very well.

An overall review:

I would like to see more conclusive findings in the paper. In the current version, there are a lot of descriptive results presented. It would be good if a few bullet points summarize each subsection in the results and discussion section for the readers to get a quick notion of the major findings.

Minor comments:

1. Introduction, Page 3: The authors can add the papers on the 2019 UAVSAR AM-PM campaign conducted in preparation of the NISAR mission. Suitable papers suggested are: Chapman et.al IGARSS 2019 and Khati et al Remote Sensing, 2020
2. Introduction, Page 3: The NASA-ISRO ASAR campaign is L- and S-band. It is incorrectly mentioned as P- and S-band
3. Page 7, first paragraph: Why is the L-band data histogram mean made 2.5 dB below that of the C-band. Can you provide some references
4. Page 10: How was DEM generated from the RGB data? Was this a stereo-pair image? Kindly elaborate
5. Page 13, Equation 5: Kindly change it to "intensity" as you use 20log10. Else if the amplitude was used, then change the equation to 10*log10
6. Page 15, Figure 8 and 9: Mismatch between caption and graph frequencies
7. Figure 9: Why don't we see any temporal trend over the 3-month period due to crop cycles?
8. Page 17, Section 5.2: how did you take 11m radius with 10m square pixels. Was sub-pixel weighted average taken, or any other method used? Kindly elaborate
9. Page 19: Line 737-739: the use of the sentence structure "in the co- and R2=0.20 ....." is confusing. Kindly consider repharasing
10. Figure 11: Kindly see that the -ve sign in the third row is visible
11.  

Author Response

Reviewer 1:

I have completed the review of this paper on the SARSense campaign. The design of the campaign is very good. It is disappointing that the corner reflectors were not aligned resulting in the loss of ability for absolute calibration of the airborne data. However, the paper presents the analysis very well.

An overall review:

I would like to see more conclusive findings in the paper. In the current version, there are a lot of descriptive results presented. It would be good if a few bullet points summarize each subsection in the results and discussion section for the readers to get a quick notion of the major findings.

We appreciate your suggestion very much. You are correct that due to the four analyses we discuss in the paper, the most important results are not highlighted and directly identifiable to the reader. In this regard, we reedited the manuscript in a way, that the main findings of each correlation analysis are explicitly stated in the end of the related results & discussion chapters. In this regard, the reader is directly lead to the major findings. (Line 662 – 670; 774 – 781; 852 – 859, 966 -973)

Minor comments:

1. Introduction, Page 3: The authors can add the papers on the 2019 UAVSAR AM-PM campaign conducted in preparation of the NISAR mission. Suitable papers suggested are: Chapman et.al IGARSS 2019 and Khati et al Remote Sensing, 2020

Thank you very much for mentioning this recent mission. We added the mission to the to the introduction chapter. (Line 125 - 129)

1. Introduction, Page 3: The NASA-ISRO ASAR campaign is L- and S-band. It is incorrectly mentioned as P- and S-band

Thank you very much for noting this mistake. Indeed, the NASA-ISRO ASAR campaign is L- and S-band SAR, we changed it in the manuscript accordingly. (Line 119)

1. Page 7, first paragraph: Why is the L-band data histogram mean made 2.5 dB below that of the C-band. Can you provide some references

We highly appreciate your comment. The 2.5 dB was chosen based on the research of Ulaby and Dobson 1989, indicating, that the backscattering signal of L-band is in the mean 2.5 dB lower than C-band. We edited the reference in the manuscript accordingly. (Line 288)

1. Page 10: How was DEM generated from the RGB data? Was this a stereo-pair image? Kindly elaborate

Thank you very much for indicating this lack of explanation. In principle you are right. An intermediate step to generate orthomosaics is the generation of a DEM by photogrammetric principles. The procedure is currently known as structure-from-motion. We elaborate about this in the new version. Now it reads:

“During the standard procedure to generate orthomosaics, a Digital Elevation Model (DEM) was generated from RGB data by a structure-from-motion procedure. Characteristic features were identified in multiple images and by this photogrammetric multi-view approach the 3D position and orientation of each feature is retrieved. The resulting 3D point cloud can then be converted into a DEM, which is provided here with a spatial resolution of < 8 cm.” (Line 375 – 382)

1. Page 13, Equation 5: Kindly change it to "intensity" as you use 20log10. Else if the amplitude was used, then change the equation to 10*log10

Thank you very much for noting this wrong wording. As we used intensity, we changed it accordingly in the manuscript (Line 543). To be consistent in our wording, we also changed the term “amplitude” to “intensity” in chapter 5.4 (Line 874).

1. Page 15, Figure 8 and 9: Mismatch between caption and graph frequencies

Thank you very much for your comment. We edited the caption of figure 8 and 9, that it is now clear, multiple radar signals are displayed within the time series graph. The different frequencies of Sentinel-1A/B and ALOS-2 are caused by the different repeat cycles. (figure 8, figure 9)

1. Figure 9: Why don't we see any temporal trend over the 3-month period due to crop cycles?

Thank you very much for your question. For the temporal comparison, the overall mean was calculated for each acquisition from the whole flight tracks A, B and C. As the tracks cover multiple landcover types (forest, urban areas, agricultural areas) and various crops are cultivated on the agricultural fields, the impact of the crop cycle of individual fields are strongly attenuated. This effect was intended, as the focus is on investigating the calibration induced difference between air- and spaceborne SAR data, independent from individual surface parameters. Nevertheless, if plotting the SAR data of a single field or a crop for the whole investigation period, crop cycle (e.g., harvest) induced backscattering changes can be observed. Please find further information in the SARSense Report, also investigating the temporal backscattering behaviour for individual crops (https://earth.esa.int/eogateway/documents/20142/37627/Final-report-SARSense-final.pdf) We changed the manuscript accordingly, that this point is clearer. (Line 585 – 588)

1. Page 17, Section 5.2: how did you take 11m radius with 10m square pixels. Was sub-pixel weighted average taken, or any other method used? Kindly elaborate

We highly appreciate your question. Indeed, the mean was calculated using a buffered point shapefile with 11m radius, calculating the area-weighted mean of the covered pixel values. We edited the manuscript, being more precise on this issue (Line 674)

1. Page 19: Line 737-739: the use of the sentence structure "in the co- and R2=0.20 ....." is confusing. Kindly consider repharasing

Thank you very much for noting this confusing sentence. We changed it in the manuscript, that it is now easier to understand. (Line 752 – 754)

1. Figure 11: Kindly see that the -ve sign in the third row is visible

We appreciated your comment very much, but unfortunately, we did not quite understand what is meant by “-ve sign”. Still, we increased the transparency of the legends within the graphs, that they are not covering the scattering points. (figure 11).

Author Response File: Author Response.pdf

Reviewer 2 Report

This manuscript presents an interesting airborne campaign carried out to provide a publicly available SAR dataset acquired at C and L band over agricultural crops, plus the associated ground campaign. This sort of initiatives is always welcome, since different groups and institutions could use this dataset for algorithm testing and benchmarking. With this aim, the manuscript serves to advertise the existence of the data (which could be overlooked otherwise).

The text is well written and there is a comprehensive analysis of some initial comparisons between the airborne data and analogous satellite data (from Sentinel-1 and ALOS-2) and in situ data (soil moisture, plant water content).

Therefore, the manuscript is worth to be published. Before that, there are some aspects which need detailed clarifications. I also have some short suggestions for trying to improve the paper. All of them are explained next.

-L89: PAZ (identical to TSX and TDX) and the Radarsat Constellation Mission should be also cited here as current spaceborne SAR’s.

-L100-104: AgriSAR2009 was not a ‘flight’ (airborne) campaign and was not aimed at ‘supporting the RADARSAT-2 mission’. It was a satellite campaign in which Radarsat-2 images were used to test the potential and performance of the future Sentinel-1.

-L164: -scale -> multi-scale.

-Section 3.1. There is no mention to the polarimetric calibration. Was it carried out in order to get the right phase differences between channels? This aspect is completely skipped in the text, but the authors state that the data are “fully polarimetric” (e.g. in line 285), so the data must have been calibrated from the polarimetric point of view. Please add the description of this calibration. Otherwise, if the data are not calibrated, then it should be pointed out that only the backscattering coefficients of the 4 channels (HH, HV, VH, VV) are available.

-Section 3. Which speckle filter was used for each data type? Please indicate the filter and kernel size.

-Discussion on variability comparison between data (airborne and satellite) in section 5: in my view, a fair comparison between the two types of data can only be performed once they have the same equivalent number of looks (ENL). Otherwise, the data may have been estimated under different conditions, hence showing different variance. Please comment on that and clarify accordingly.  

-There is no mention to the potential use of interferometry and/or time series with the future L and C satellites for retrieving soil moisture and vegetation parameters. Please fill this gap.

- L729: ‘significantly’ Is the right word (statistically speaking)?

- The final report of the campaign (accessible at the website indicated for the dataset) should be also referred to in the manuscript. I found it and downloaded it, but other readers may overlook its existence.

Author Response

Reviewer 2:

This manuscript presents an interesting airborne campaign carried out to provide a publicly available SAR dataset acquired at C and L band over agricultural crops, plus the associated ground campaign. This sort of initiatives is always welcome, since different groups and institutions could use this dataset for algorithm testing and benchmarking. With this aim, the manuscript serves to advertise the existence of the data (which could be overlooked otherwise).

The text is well written and there is a comprehensive analysis of some initial comparisons between the airborne data and analogous satellite data (from Sentinel-1 and ALOS-2) and in situ data (soil moisture, plant water content).

Therefore, the manuscript is worth to be published. Before that, there are some aspects which need detailed clarifications. I also have some short suggestions for trying to improve the paper. All of them are explained next.

-L89: PAZ (identical to TSX and TDX) and the Radarsat Constellation Mission should be also cited here as current spaceborne SAR’s.

We highly appreciate your comment and of course the PAZ and Radarsat Constellation are recent space borne SAR missions. We changed the manuscript accordingly. (Line 89; 92)

-L100-104: AgriSAR2009 was not a ‘flight’ (airborne) campaign and was not aimed at ‘supporting the RADARSAT-2 mission’. It was a satellite campaign in which Radarsat-2 images were used to test the potential and performance of the future Sentinel-1.

Thank you very much for this clarification. As you stated, the purpose of the AgriSAR2009 campaign was confused in the manuscript. As the paragraph intends to list airborne campaigns, we deleted the AgriSAR2009 campaign in the manuscript. (Line 103 – 107)

-L164: -scale -> multi-scale.

Thank you for noting the wrong wording. We edited the manuscript according to your suggestion. (Line 172)

-Section 3.1. There is no mention to the polarimetric calibration. Was it carried out in order to get the right phase differences between channels? This aspect is completely skipped in the text, but the authors state that the data are “fully polarimetric” (e.g. in line 285), so the data must have been calibrated from the polarimetric point of view. Please add the description of this calibration. Otherwise, if the data are not calibrated, then it should be pointed out that only the backscattering coefficients of the 4 channels (HH, HV, VH, VV) are available.

We highly appreciate your comment on the polarimetric calibration. In fact, the SAR data was not polarimetrically calibrated. In this regard, the term “fully-polarimetric” may lead to misunderstandings about this issue, even though here only the technical parameter of the C- and L-band sensor is described. As the SAR data contains only the backscattering amplitude / intensity of the four polarization channels, we edited the wording within the manuscript (Line 229; 294; 302). Furthermore, we added a small paragraph about this topic in the data chapter (3.1), stating that the airborne SAR data is not polarimetrically calibrated and in this regard without further calibration not suitable for eigen-based and model-based decompositions. (Line 288 – 292).

-Section 3. Which speckle filter was used for each data type? Please indicate the filter and kernel size.

Thank you very much for indicating this lack of information in the data chapter. As stated in the method chapter, the airborne data was speckle filtered using a Lee filter with a kernel size of 10 x 10 (line 546 – 548). The multi-looked ALOS-2 data was initially filtered using a Lee sigma filter with a kernel size of 7 x 7 pixel, while the Senitnel-1 data was speckle filtered using a focal median filter with a kernel size of 3 x 3 pixel (line .334 – 335). In order to have a more consistent pre-processing of both the Sentinel-1 and ALOS-2 data for the correlation analysis, we re-processed the ALOS-2 data, using also a focal median filter with a kernel size of 3 x 3 pixel (line 355). The recalculation of the ALOS-2 scenes did not lead to any major changes in the results or conclusions, besides some minor changes in the linear regression statistics. We changed the related manuscript including figures and tables accordingly. (chapter 5.2; figure 10, table 4, chapter 5.3; figure 11, table 5, table 6).

-Discussion on variability comparison between data (airborne and satellite) in section 5: in my view, a fair comparison between the two types of data can only be performed once they have the same equivalent number of looks (ENL). Otherwise, the data may have been estimated under different conditions, hence showing different variance. Please comment on that and clarify accordingly.  

Thank you very much for issuing this topic. Indeed you are right, that a true comparability of the variance between the airborne and space borne data is not given due to the different ENL. In this regard chapter 5.5 may be misleading, without addressing the limitations of the different ENL. As the main intention of this chapter is to analyse the correlation between airborne C- and L-band SAR data and NDRE, and not the comparison between space and airborne data, we agree, that the chapter needs to be edited. As different surface parameters (soil moisture/vegetation parameters vs. NDRE) were chosen, we intended to show the potential of both the air- and space borne data of the SARSense campaign for analysing C- and L-band backscatter behaviour. Furthermore, we wanted to show that despite the misaligned corner reflectors and resulting suboptimal calibration, a similar behaviour of C- and L-band can be observed in both air-and space borne data for the different crop types. We have revised the manuscript so that our intent is now clear and we were not seeking a direct comparison between air- and space borne analysis. (Line 933 – 935; 940 – 943; 948 – 951; 953 – 956; 959 – 966; 1037 - 1040)

-There is no mention to the potential use of interferometry and/or time series with the future L and C satellites for retrieving soil moisture and vegetation parameters. Please fill this gap.

Thank you very much for providing this input. It is true, that we do not cover the topic of interferometry and time series methods for soil and vegetation parameter retrieval, as we did not intend to focus on any retrieval method in this paper. Nevertheless, as the upcoming satellite missions, offering publicly available consecutive temporally and spatially high resolution C- and L-band SAR acquisitions, it will be especially valuable for change detection retrieval methods. In this regard, we issued the potential in the introduction chapter. (Line 72 – 75).

- L729: ‘significantly’ Is the right word (statistically speaking)?

We highly appreciate your comment on this wording. As this term is used in a qualitative rather than a quantitative way, we deleted it to avoid any misunderstandings. (Line 1064)

- The final report of the campaign (accessible at the website indicated for the dataset) should be also referred to in the manuscript. I found it and downloaded it, but other readers may overlook its existence.

Thank you very much for your comment. We edited the manuscript accordingly and mentioned it right next to the data set link. (Line 995)

Author Response File: Author Response.pdf