The Retrieval of Ground NDVI (Normalized Difference Vegetation Index) Data Consistent with Remote-Sensing Observations
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn this paper, titled “Retrieval of Ground NDVI Consistent with Remote Sensing Observation”, the authors proposed a methodology to obtain ground NDVI with camera. The consistency issue among observations of different sensors has been discussed in other literatures, mainly on multispectral sensors onboard satellite. In this manuscript, the authors focused mainly on the consistency issue between ground-based camera and satellite remote sensing, especially on spectral indices (i.e. NDVI and GCC). However, there are still several problems need to be addressed or presented clearly.
(1) In the comparison with satellite observation, PlanetScope was considered. The consistency between ground NDVI camera and PlanetScope was observed. How about sensors onboard other satellite? Actually, there are differences among sensors, such as the sensors mentioned in this manuscript (e.g., Landsat 8 OLI, QuickBird, GF-1 PMS, Sentinel-2A). If it is necessary or possible to reset (or redesign) the filter of the camera, when we compare it with other remote sensing sensors.
(2) In comparison study, spatial match between ground based measurement and satellite observation is critically important. As the authors mentioned (line 236), “an area of 4×4 m2” was used, covering as one pixel of imagery acquired by PlanetScope. How to ensure the accuracy in geolocation of them. Uncertainty in geolocation within each observation (ground measurement) should be considered.
(3) The authors proposed a simple method to conduct radiometric calibration, while a coefficient K was used. Please show readers clearly how to determine K.
(4) Advices to readers on applying this methodology for practices should be presented.
Thanks.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe study introduces a new NDVI camera that uses narrowband dual-pass filters to capture vegetation reflection, aligning with remote sensing sensors.
The Research paper provides respectable findings and is well-written, but before it is accepted, it needs to be strengthened in the following ways:
In Introduction section, Author should describe the NDVI.
On page-3, For a better understanding of the study, the author should provide the methodology flow work.
On page-4, Table-2, Please verify that the spatial resolution unit is in metres (m) rather than nanometers (nm).
Author should discuss the Power Requirements and the environmental conditions in which the NDVI camera should be sustained.
of the NDVI camera.
The author should elaborate on how distinguishes the proposed NDVI camera from conventional ground digital cameras.
The author should address the study's limitations and challenges, specifically regarding the camera's ability to ensure the accuracy of NDVI values.
The study focuses on the Hulunbuir Grassland. Please talk about whether the camera can be used with different kinds of vegetation.
The level of agreement between the NDVI readings acquired from the camera and those from the hyperspectral spectrometer should be explained by the author.
Comments on the Quality of English Language
Quality of english is good.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis manuscript describes a study that developed a new ground-based camera for estimating vegetation indices and compared results with a ground-based, collocated spectrometer and with satellite observations. The topic is of interest to readers of Remote Sensing.
I am a user of NDVI and familiar with the underlying physics, but I am not an expert in instrumentation, so was not able to evaluate those parts of the study or the novelty of the study.
I found the study interesting and potentially publishable. However, there are major issues with the manuscript that need addressing. The most important ones include the lack of details in many places; the justification for needing a new camera in terms of measuring biological data; the mismatch between results (as represented by the figures) and the text, including in the Results, Discussion, and Conclusions (and Abstract) (for instance, I think the authors try too hard to show that their results are accurate, and I’m not sure I agree); and, sometimes, the structure of the manuscript in terms of what text is where and in what order. As a result, I recommend that the manuscript needs a major revision.
The manuscript’s English is pretty good, but there are multiple errors (I only noted a few below), so the authors should reread their text carefully and/or have an English language editor provide suggestions.
Comments
It seems that an important result from this study is the reduced utility of GCC compared with NDVI, and an explanation why. Please add this result to the abstract and conclusions as it will be helpful for readers to learn about this.
There are a lot of figures in the manuscript. I’m not sure the box plots show different information compared with the scatter plots (e.g., Figures 14 and 15), so I think the box plot figure could be deleted…?
L 22: What does “real radiance” mean? As opposed to what?
L 25-27: What is the “traditional color-based index”?
L 29: The values of the RMSE and MD don’t mean much to me. What is the range of NDVI? What is the percentage error?
L 59: I’m not clear about the need to evaluate NDVI from satellites. NDVI is an observation; how can it be wrong? Perhaps the issue is the interpretation of NDVI in terms of some biological phenomenon, like NPP or biomass. The authors cite Brown et al., but I would appreciate reading more here about the need for ground-based cameras. This manuscript describes a study that evaluates NDVI from a camera compared with NDVI from other instruments, and it’s not clear to me that a “perfect” calculation of NDVI from the camera (as evaluated with the hyperspectral instrument) would solve the problems of interpreting NDVI as a measure of NPP (for instance). Please provide a paragraph describing this in more detail.
L 71, 72: What does “regular” mean?
L 82: First name, middle initial not typically used in citations…?
L 89-94: I’m not clear why a scaled Camera NDVI is unusable for studying NPP, for instance. Please provide more explanation/justification.
L 91: What is “quantitative consistency analysis”?
L 108-110: Cameras are remote sensing sensors; can the authors clarify what they mean?
L 110-111: I don’t understand the ISP processing comment. Why does it not allow the camera to record NDVI? Please provide more description.
L 114: Handling what? Incomplete sentence.
L 115-116: Again, I’m not clear about the need and justification for this study. I would like to read a description that of a study that reported that the NPP or phenology or… derived from a camera was incorrect. It’s possible that the authors cited such a study, but I would like to read a brief description of it to learn more about why the study described by this manuscript is important.
L 144: And wideband information for the red band too, according to Figure 2a. Seems important to add.
Figure 2: Because NDVI is supposed to capture vegetation, it would be helpful (and easy, I think) to add a typical vegetation reflectance spectrum to these panels to give the reader an idea of, especially, where the NIR band is placed relative to the vegetation reflectance.
L 170-172: JPEG is, of course, a lossy compression format. I wonder if a particular feature of interest would be affected. Could another lossless format, like PNG, be used instead?
L 209-210: I assume the camera records RGB and R+NIR images consecutively (if so, what is the time difference?). Can the authors describe this explicitly in an above section?
L 214: What causes abnormal filter switching? What fraction of images is this?
L 227: For the study setup, what is the size of the hyperspectral image, and how do the authors match the area with the area sensed by the camera?
L 228: Please add a description of the radiometric calibration of the satellite data, including variability among the sensors/satellites. Was any radiometric normalization among these sensors done? My understanding is that there might be substantial variability…? Perhaps any variability might be minimized by calculating NDVI, but the authors should address this by showing the NDVI data they obtained from different PlanetScope satellites.
Similarly, are there differences in NDVI for various times of day and viewing angles (how much do these vary among PlanetScope satellites?)?
Was any atmospheric correction performed on the satellite data? Were these top-of-atmosphere or surface reflectances. Please describe.
L 245: Thus far, this study has been all about NDVI. Why then is GCC analyzed? Please provide a description and perhaps introduce the need for GCC earlier in the manuscript.
L 260-283: It seems there are multiple normalization steps…? If so, add “in X steps” to sentence on L 260. L 262, “Firstly”: good signal to readers. Add “Second”, etc. to beginning of text describing subsequent steps to help readers.
L 268-269: What are B1-B3? The text just described PlanetScope bands as B?, but I don’t think these are them. I expect these are the bands from the camera (?). Please add a reminder.
L 280: Spell out ROI.
L 282-283: What is the source of the temporal noise? Describe.
L 284: How was K determined? Did the authors estimate K themselves? If so, how often was the calibration performed?
Section 3.1.1: Why do this analysis before radiometric calibration? I’d like to see the analysis after calibration, unless there’s a good reason not to (in which case, please explain why). LATER: After reading more, perhaps the reason is that the camera was calibrated using the spectrometer. This wasn’t clear in the Methods; please expand the description there and mention that reason here.
Why don’t the authors show comparisons of the individual bands (R, G, B, NIR)? Presumably they can convolve the hyperspectral data to the camera bands…? Or maybe that is not even necessary (pick the best single hyperspectral band?).
Figure 7: Please change “Spectrum” in legend to “Spectrometer”.
L 325: I wouldn’t say there is a “robust” agreement between the camera and spectrometer. For GCC, the initial increase in the spectrometer is not matched in the camera; the pattern is opposite in NDVI (greater initial increase in camera).
Also, reporting R2 will be another evaluation tool. (Change Section 2.5 to use R2, not R).
L 330-331: Can the authors explain more about how the limited dynamic range of the camera affects NDVI? If true, wouldn’t that difference decrease as NDVI decreased (later in season)?
L 333: What are these external influences that affect the camera but not the spectrometer?
L 340 and Figure 8: Please report R2.
Figure 8: Shouldn’t we expect the slope to be 1 and the offset to be 0? The different slopes, in particular, indicate different responses, I think. Please add a discussion.
Also, please add the 1:1 line to all appropriate figures (not just Figure 8).
L 248-350: Incomplete sentence.
Section 3.1.2: Why do the K values stabilize over time? What changes?
Also, why don’t the authors use the dynamically changing K values plotted in Figure 9 instead of using the average value of K? Early in the season there is substantial variability in K.
L 368: What is a decrease in calibration? Clarify.
Given my comment above about using the average K instead of a varying K, I’m not surprised that the R values decrease after calibration.
Again, can the authors go back to individual bands to evaluate their speculation about the cause of differences (especially relative to NDVI)?
Figure 10: The orange and red colors are too close together. Please pick another color, like purple (if that results in clear distinction among the three variables).
Is there something about the GCC formula, which has one value in the numerator and three in the denominator, that makes GCC more susceptible to K issues than NDVI (two values in each of numerator and denominator, perhaps canceling errors)? Perhaps the authors can play with these equations and evaluate impacts of different K values for the same RGBNIR values.
L 393: Why did the authors apply and show the fit using a hyperbolic tangent function? More description needs to be added to Methods. The fit is pretty poor in places (e.g., early in season for spectrometer GCC); a spline fit would result in a better fit, depending on the purpose of this filtering/fitting.
Section 3.2: I think a useful addition would be a time series figure showing GCC and NDVI from the satellite, the spectrometer, and the camera. LATER: I see this is Figure 14, almost. To be consistent with previous analyses, please move time series earlier.
Also, add spectrometer data to Figure 14.
Also, use consistent colors with previous figures, and avoid red and green to help readers with red-green color blindness.
Again, the slopes are not 1…please discuss.
There appears to be an important bias at low GCC values…please discuss (will become more evident when 1:1 line is added).
L 452-460: This is an interesting analysis. Can the authors explicitly state that they expect NDVI not to change from midday to morning or evening during these dates (right?)?
Also, what are the spectrometer results for the same analysis? I’m wondering if the issues result from something about the camera or if the issues result regardless of sensor. Seems important to communicate this to the reader.
L 460-465: I’m less excited about this given my comments about this fit above. The fit is poor; why would such a filter be applied? Doing so will eliminate potentially important biological information.
L 468: Not for GCC. Please discuss GCC here. What about the RGB bands is important? One reason mentioned above is that these bands are wide…please explain more how this translates to poorer agreement.
L 475: I think by “collinearity” the authors mean that there is substantial overlap in the wide red and NIR bands of built-in filters. Please add a bit more explanation to help readers understand, perhaps referring to Figure 2.
L 478-488: This information seems to belong in the Introduction, and if already mentioned there, the length can be reduced to avoid duplication. The authors can then be briefer here in their comparison of their results with other studies.
L 491-495: There are at least two places where explanations are given for the poorer performance of GCC, here and in the Results around L 369, and the explanations are not the same. I think this interpretation may be better only in the Discussion. Please decide where to put this discussion and have it only in one place.
L 499: Replace “the latter” with a specific study name; I’m confusing if “latter” means the authors’ study or the Nigland study.
L 501: Strictly speaking, the authors did not measure canopy status, so please change this to something about the spectrometer results instead. You can then say there are implications of this result for monitoring canopy status.
L 510-511: That’s not what Figure 9 shows to me…?
L 532-533: Comparing Figures 8b and 11b, and I see some improvement in evaluation metrics. However, I think saying that calibration is essential and necessary is too strong. I expect one can still get excellent results with the uncalibrated data, and that is useful information for readers. Please modify this text.
L 557-558: The authors are thinking too narrowly. Some research relies on long-term field observations, but other research does not. And maybe those observations should span multiple years, but maybe not. Please alter the language to be less definitive.
L 568: Please replace “can infer” with “suggest”.
Also, why might the NDVI camera be uncertain for evergreen vegetation? Is it NDVI regardless of sensor? More description, please.
L 572-574: What results is this statement based on? What was the evaluation? More information is needed in Methods and Results.
Also, do the authors recommend using GCC? I thought that they were not, based on their findings. GCC results need to be summarized in a separate paragraph earlier in the Discussion.
L 580: Doesn’t resolution depend on distance to object being measured and sensor properties, and isn’t resolution sometimes coupled with image size (extent)? More description, please.
This paragraph seems to be about the advantages of this camera versus other sensors and other cameras. However, the authors did not compare their results with other NDVI cameras (only with their spectrometer), so it seems that much of their text can apply to other NDVI cameras. Please clarify the text, and if you are talking about NDVI cameras in general, state that. If you think there are advantages of your camera over other NDVI cameras, please stick to specific comparisons. For example, is automated image capture a feature of your NDVI camera alone and not other NDVI cameras?
L 602: I’ve asked above for band-specific comparisons, and until those are presented, I think the authors should stick to reporting the NDVI results, not individual bands.
L 604: Please be clearer that “band selection” means placement of band within the spectrum and band width.
L 610-617: The increase in accuracy is compared with the spectrometer results, not with any representation of biology. That is, other calculations or measurements of NDVI might represent phenology equally well. Please modify text accordingly.
Comments on the Quality of English Language
The manuscript’s English is pretty good, but there are multiple errors (I only noted a few below), so the authors should reread their text carefully and/or have an English language editor provide suggestions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsNo more comments.
