Seasonal Comparisons of Himawari-8 AHI and MODIS Vegetation Indices over Latitudinal Australian Grassland Sites

Round 1

Reviewer 1 Report

The manuscript presents a relatively simple and straightforward evaluation of cross sensor compatibility between a geostationary imaging sensor, and a polar-orbiting sensor with roughly similar spatial and spectral resolution.  The main finding of the study is that if appropriate corrections for illumination and viewing geometry are made, values for three commonly used vegetation indices are strongly correlated and comparable.  These results could be of great use for remote assessment of phenology and other time-dependent canopy properties, where the increased frequency of coverage from the GSO platform would be applicable.

 

The methods used for the analysis are appropriate, as far as I could tell, with a few minor questions listed below.  The manuscript is clear and well-written.  The research questions are answered, and the results are relevant to the journal.  I really don't have many comments, apart from those below, which are more intended as clarifications or additional useful information for the reader.

1. The time series are based on pixel averages, and it would be useful for the reader to know what the variation associated with those averages was.  The number of averaged pixels is (obviously) much larger for the MODIS data,still knowing if the overall variance of the MODIS versus HSI data would make it easier to assess the comparisons of their corrected values that are the main thrust of the analysis.
2. In the figures where linear comparisons are made between the two sets of vegetation indices, could an error metric such as RMSE be used along with R2?  R2 is a useful measure of association, but it doesn't really fully quantify how similar two measurements are in absolute value.

One additional comment: the numbering the figures is off.  The numbers go from 7 to 12, and the body of the text therefore makes reference to several figure numbers which don't exists (e.g. 9, 11).  It was possible to determine which figure was the appropriate one based on the text description, but obviously this needs to be fixed before publication.

Author Response

We appreciated the positive comments and feedback from the reviewer. The reviewer had a few minor comments and clarifications which we address below.

1. With respect to pixel averaging, this was done merely to match the footprints of AHI with MODIS due to their widely different projections, as depicted in Fig. 5. Our goal was to minimize any potential differences caused by variations in footprint between MODIS and AHI data. Due to such differing pixel projections, we were not able to simply subsample an AHI 1km pixel into 16 MODIS 250m pixels; and considering the pixel misregistration issues, we adopted a larger 3x3. AHI pixel approach.  As we averaged to identical footprints and we did not subsample within the footprint, the spatial variance, or surface heterogeneity, information was not be needed. 
2. We added RMSE to the figure in question (Fig. 11) as requested by the reviewer. RMSE values were in the order of 0.02 to 0.03.

We also corrected the Figure numbering.

Reviewer 2 Report

The potential utility of geostationary observations for understanding vegetation phenology makes the analysis described in the manuscript of broad interest to a wide audience.  Comparison of grassland sites over 2 years should be very informative.  The comparison of NDVI and EVI2 is interesting, but the omission of EVI from the analysis detracts from the study.  Particularly because the results show a considerable difference between NDVI and EVI2 while other studies have shown that EVI is more strongly correlated with other vegetation indices than NDVI. This study should include EVI also.

There appears to be an inconsistency in the results as they are presented.

The figures labeled Figure 6 and Figure 7 (which are actually the 10th & 11th figures in the manuscript) seem to be inconsistent with each other.

Figure 6. ("Comparison of smoothed, daily composite AHI...") shows AHI NDVI consistently underestimating MODIS NDVI, with geometry adjustments having little effect, while  AHI EVI matches the sun & view angle adjusted MODIS EVI more closely on all sites with the geometry adjustments having strong effect.

In contrast, Figure 7. ("Global cross-site grassland relationships between AHI and MODIS...") shows NDVI much closer to the 1:1 line than it appears on the previous figure.

The AHI NDVI is conspicuously lower than all three MODIS NDVI time series for 3 of 4 sites in Figure 6, yet all three MODIS NDVI are closer to the 1:1 line in Figure 7.  How is this possible?

 

Other issues to be addressed:

Fig. 2 Solar elevation angle mislabeled as SZA.

l.140 - Why 9 days?  Minimum to attain 100% cloud-free?  If not, what % cloud-free?

l.190 What is the size of the filter kernel, and how was it chosen?

l.204  Why are both median & SG filters used?  

All figures after the 4th appear to be mislabeled.

 

 

 

Author Response

We thank reviewer 2 for the constructive suggestions and comments. 

1. We have now added EVI to all the relevant figures (Figs. 6-12) for proper comparisons of the standard MODIS EVI product. Although EVI and EVI2 were quite similar and comparable to one another, the EVI2 performed better, hence some discussion was added on AHI blue band issues, as reported by Li, S. et al. (2019).    Li, S. et al. First provisional land surface reflectance product from geostationary satellite Himawari-8 AHI. Remote Sens. 2019, 11, 1–12
2. Regarding potential inconsistencies between Figs. 6 and 7 (or technically Figs 10 and 11), we checked our data and they are consistent. AHI NDVI time-series appears much lower than MODIS NDVI time series for certain times of the year (e.g., April-October for Redesdale). As Figure 7 (corrected to Figure 11), shows, MODIS NDVI is higher than AHI NDVI, and always above the 1:1 line.  Redesdale and Richmond show this most dramatically with maximum MODIS NDVI at 0.8 and AHI NDVI at 0.75, as seen in both figures.  This is similarly found at Richmond with MODIS at 0.7 and AHI at 0.65.

Other issues:

3. We thank reviewer 2 for pointing out the mislabelled SZA in figure 2 and have corrected this.
4. Why 9 days: Finding a single-day with 100% cloud-free data was difficult and choosing 9-days of data enabled us to show a complete diurnal profile for all seasons, especially the rainy periods. This also represented a measure of consistency in diurnal variations across consecutive days.
5. Size of filter kernel: We chose a size of 17 for the SG filter because we wanted to compare AHI daily VI with 16-day MODIS VI (MOD13Q1 and MYD13Q1).
6. Median and SG filters: We used a MAIAC version 1 prototype of AHI surface reflectance data. While it was generally good data, it did contain certain noise and data spikes which SG couldn’t totally remove. Therefore, we found that using both median and SG filters was necessary.
7. The mislabelled figure numbers were fixed.

Reviewer 3 Report

This is a valuable cross-sensor comparison illustrating the possible combined use of geostationary and polar orbiting sensors data for grassland monitoring. Fensholt et al. (2010) have shown that the degree of anisotropy in red and NIR reflectances depends on the amount of vegetation present. MODIS View Zenith Angle VZA and Relative Azimuth Angle RAA effects on NDVI are highest for medium dense vegetation (NDVI ≈ 0.5–0.6, a range of NDVI observed for Australian grasslands, cf. Richmond site). The need of BRDF corrections is clearly demonstrated in this paper and interesting perspectives are suggested.

Reference :

Rasmus Fensholt, Inge Sandholt, Simon R. Proud, Simon Stisen & Mads Olander Rasmussen (2010): Assessment of MODIS sun-sensor geometry variations effect on observed NDVI using MSG SEVIRI geostationary data, International Journal of Remote Sensing, 31:23, 6163-6187

Author Response

We thank the reviewer for the positive and constructive comments. We’ve added the recommended reference by the reviewer into our manuscript and integrated it into our discussion (l.387 – l.390).

Reviewer 4 Report

This study conducts seasonal comparisons of Himawari-8 AHI and MODIS vegetation indices over four Australian grassland sites. It has four objectives: (1) assess the potential to harmonize MODIS and AHI VIs; (2) investigate the diurnal variations in AHI reflectances and VIs in relation to sun-angle variations; (3) develop a daily compositing methodology; and (4) compare seasonal AHI VIs with MODIS standard VI products and BRDF-corrected VIs.

However, the manuscript has many limitations:

(1) This paper does not seem to contribute much to the remote sensing community. The authors only describe the common sense that every remote sensing researcher knows. The observed seasonal variation of grassland should be the same no matter what the satellite sensors are. The atmospheric correction and BRDF correction should be conducted before the time series analysis to minimize their impacts. Develop a daily compositing methodology by only selecting mid-day, near solar noon observations within a four-hours sampling period (10–14h) is not a contribution, and it is also a common pre-process that every remote sensing researcher knows.

(2) The authors only used four points to depict the phenomenon, where the size of the samples is too small to prove the relationship between MODIS and AHI VIs. The authors should add more samples, and the significance test is also required.

(3) The manuscript should not emphasize the objective that they do not (or cannot) achieve. For example, the first objective: assess the potential to harmonize MODIS and AHI VIs. The manuscript does not contain any attempts or examples of harmonizing MODIS and AHI VIs.

(4) The manuscript has many obvious mistakes, such as errors in citations, confusing numbers of figures and tables, and many grammar errors. The authors are too careless, and these low-level errors should not appear in the manuscript that is officially submitted to any journals.

Therefore, I cannot recommend publishing the manuscript in the current form.

Author Response

We thank the reviewer for his/her critical review of this paper and pointing out limitations.  We address the 4 major concerns posed by the reviewer below: 

1. (1^st) “The observed seasonal variation of grassland should be the same no matter what the satellite sensors are” - We agree with this basic premise and the temporal profiles of vegetation indices from different sensors are generally similar, as was found in this study.  However, more detailed analyses can reveal important differences, which can result in systematic differences when further processed to extract certain information (such as phenological metrics).  Variations in cross-sensor VI profiles may include differences in spectral bandwidth and observation geometries.  We have now better defined the objectives to include comparisons of VIs from the AHI and MODIS sensors (as in the title of the paper).

 

“Atmospheric and BRDF corrections should be conducted before…” – We also agree with this reviewer statement, however, MODIS VI products, GIMMS NDVI, and the current VIIRS VI products from NOAA and NASA are not normalized to a standard geometry with a BRDF model.  Therefore, investigating and understanding how VI temporal profiles would be different empirically with the actual datasets is an important step, particularly given that with this new-generation of geostationary satellites, we are experiencing observation geometries that we never experienced with MODIS, VIIRS or AVHRR VI time series data.  Figure 9 readily shows how using the most robust phenological metric of the half-amplitude (peak minus minimum), will result in a different start of season between AHI and MODIS. This study also empirically verifies a key argument made by Adachi et al. 2019, about aligning relative azimuth angles.

 

Regarding the daily compositing methodology of using mid-day values around solar noon – we have now removed this from the study objectives.  We do think the compositing method is different in that over these dry grassland ecosystems, it was the minimum VI values that had to be selected to represent daily composited values, rather than the common maximum value-based approaches. As diurnal observations are relatively new in VI studies, it is worthwhile to show such observations along with compositing challenges.

 

2. (2^nd and 3^rd) We have processed more grassland areas, but only presented 4 sites to represent and sample across the functional range of grasslands, from subtropical C4 grasslands to temperate C3 grasslands, not only to cover the major grassland types but also explore a latitudinal range of AHI view angle variations. Additional sites added nothing new to this study and followed the same patterns already shown, including before and after any BRDF adjustments to the MODIS data.  We ran a significance test for linear regression model with p-values below 0.0001, and have included this in the Figure 11 caption. As we never intended to do a harmonisation study, we did not attempt to prove any relationship, especially given our very limited scope of dry grassland sites.  We have removed the objective to “assess the potential to harmonize MODIS and AHI VIs” to avoid confusion. 

 

3. (4^th) We apologize and have now corrected the figure numbering. We also double checked all our citations. We additionally searched and corrected for grammar issues and believe that the manuscript is now fully understandable and free of English errors.

Round 2

Reviewer 2 Report

The inclusion of EVI in the analysis increases the value of the work considerably.  However the discussion of the results seems to reveal a bias in favor of NDVI over EVI and EVI2.  This seems odd because NDVI's shortcomings are now well established.

Two observations that should be discussed are 1) the strong sensitivity of NDVI and insensitivity of EVI to phase angle in Fig.7 and 2) the apparent saturation of the H8 NDVI (for all sites except Mullunggari) in Fig. 10.  Together, these observations seem to contradict the suggestion that NDVI is more robust to seasonal BRDF effects.  If conclusions are to be drawn from the small differences among the very similar correlations and slopes in Fig. 11, the authors should include a through analysis of the statistical significance of each comparison for each site separately as well as in aggregate.

Author Response

We thank the reviewer for his/her extra comments on this paper. We address the concerns posed by the reviewer below.

1. We agree with reviewer 2’s assessment regarding NDVI and EVI robustness to BRDF effects and have revised these sections to clarify and be more explicit about VI robustness. Our revisions reinforce our findings that NDVI show  strong sensitivity to diurnal changes of sun-view angles in mentioned the Discussion section (Iines 397- 403).  
2. We entirely rewrote the interpretation of the seasonal comparisons of  AHI and MODIS data to show the apparent inconsistencies of NDVI sensitivity to sun angle, but not to view angle which made it appear the NDVI was robust to seasonal BRDF effects. We now highlight the same negative NDVI hotspot effect, present in the diurnal data, to be responsible for an NDVI negative backscatter effect in the seasonal results, such that NDVI backscatter effects offset the view angle effects.  We also insert the possible role of NDVI saturation effects in 3 of the 4 sites.
3. We agree that the correlations and slopes of Fig 11 (now Fig. 12) are essentially the same following BRDF correction of the MODIS data.  The datasets before BRDF corrections reflect the proper need for BRDF corrections, but falsely show NDVI to not require corrections due to the offsetting effects of negative backscatter (phase angle) responses with what should be positive view angle responses. 

Reviewer 4 Report

This study conducts seasonal comparisons of Himawari-8 AHI and MODIS vegetation indices over four Australian grassland sites. It has four objectives: (1) compare actual cross-sensor datasets from MODIS and AHI VIs over a diverse functional range of dry grasslands; (2) investigate the diurnal variations in AHI reflectances and VIs in relation to sun-angle variations; (3) construct seasonal VI profiles from daily composites, and (4) compare the seasonal AHI VIs with MODIS standard VI products and BRDF-corrected VIs. This version has improved a lot compared with the previous version. However, the manuscript still needs to be improved:

 

(1) The abstract was poorly written. It does not clearly state the contributions of this study.

 

(2) The introduction of the manuscript was poorly organized and written. It does not clearly introduce the corresponding backgrounds of the contributions (e.g. why they did these, or what is the purpose).

 

(3) Many of the explanations in the response should be included in the manuscript because readers may have the same questions as reviewers.

 

(4) There should be a table showing the details of data (Sensors, spatial resolution, temporal resolution, bands used, and so on) used in this study.

 

(5) There should be a figure showing the entire comparison flowchart of this study, which can make the readers easier to understand this study.

 

Therefore, I think the manuscript needs further revisions and improvements.

Author Response

We thank the reviewer for his/her critical review of this paper and pointing out limitations.  We address the 5 major concerns posed by the reviewer below: 

1. We revised and re-wrote the abstract to better represent the findings and purpose for this study.
2. The introduction was also better organised, with purpose and aims delineated, as requested by the reviewer.
3. We incorporated the explanations of first round review responses into the discussion section, and have re-written major parts of the discussion section.
4. We added a table representing resolution, spectral bandwidth of MODIS and H8-AHI bands used in our study (line 127).
5. We also inserted a new Figure 3, that is a flowchart of our methodology used in the paper. (Iine 144). 

Round 3

Reviewer 2 Report

I applaud the authors for their efforts to address all the points raised, and look forward to seeing the work in print.

Reviewer 4 Report

This version is much better.