Cloud-Free Global Maps of Essential Vegetation Traits Processed from the TOA Sentinel-3 Catalogue in Google Earth Engine

Round 1

Reviewer 1 Report

The study employed Google Earth Engine (GEE) to develop global essential vegetation traits from Sentinel-3 data. Four vegetation maps including fraction of photosynthetically active radiation (FAPAR), leaf area index (LAI), fractional vegetation cover (FVC), and leaf chlorophyll content (LCC) were produced in the year 2019. The result looks interesting, and the data is valuable for global vegetation related study.

However, the manuscript contains several unclear parts and need to be clarified.

1.       It is not clear how the data was developed as a cloud-free product. Remote sensing imagery can always be contaminated by clouds. For the global effort, choosing cloud free images is vital. The manuscript did not explain these details.

2.       In both abstract and conclusion, the authors claimed that vegetated land covers with pronounced phenology fluctuations led to high correlations between the different products. I can not see the connection of high correlation and fluctuations. More explanations are needed. Also, the reason for ever sparsely vegetated fields as well as areas near the Equator linked to smaller seasonality led to lower correlations are not clear. Why in these high-latitude regions where vegetation has large phenology variations the product still has high correlation?

3.       Pearson correlation coefficients were used in the paper to illustrate the connections between products. However, the level of significance or P-value was not given.

4.       Comparison over glass land was missing. Since glass land counts for a large portion of the earth surface and is not a spare vegetation land, it is important to look at the performance of the developed data over grass land.

5.       What are spatial resolutions of maps displayed in Figure 7 & 8?

6.       The high resolution maps displayed in Figure 11 should be compared with the global map in 5-km to highlight the difference.

Author Response

Dear Reviewer :

I am writing in reference to the review you provided for our recently submitted manuscript in Remote Sensing. We want to extend our appreciation for the time and dedication you devoted to reviewing our work. On behalf of all the co-authors, we would like to address the queries you raised regarding the submitted manuscript.

1. It is not clear how the data was developed as a cloud-free product. Remote sensing imagery can always be contaminated by clouds. For the global effort, choosing cloud free images is vital. The manuscript did not explain these details.

Thanks for the question. Indeed, the initial products contained gaps that were induced by clouds, as illustrated in Figure 3.  We addressed this issue by employing the Whittaker Smoother (WS) to reconstruct and interpolate missing data. See also our abstract (for instance) : “We used the Whittaker smoother (WS) for the temporal reconstruction of the four EVTs, which led to continuous data streams, here applied to the year 2019. Cloud-free maps were produced at 5 km spatial resolution at 10-day time intervals.” In the introduction and methods we describe the filter in more detail, e.g.: “The WS function, which is based on penalized least squares, fits a discrete series to discrete data and balances the reliability of the observations with the roughness of the smoothed curve. In particular, the smoother allows continuous control over smoothness, works fast and is capable of automatic interpolation…”.

We hope that these and further clarifications sufficiently address your concern.

2. In both abstract and conclusion, the authors claimed that vegetated land covers with pronounced phenology fluctuations led to high correlations between the different products. I can not see the connection of high correlation and fluctuations. More explanations are needed. Also, the reason for ever sparsely vegetated fields as well as areas near the Equator linked to smaller seasonality led to lower correlations are not clear. Why in these high-latitude regions where vegetation has large phenology variations the product still has high correlation?

We apologize for having chosen this unsuitable wording. We replaced “fluctuations” with “more pronounced phenological dynamics” throughout the manuscript. These changes were made in the 

Abstract: “As a general trend, vegetated land covers with pronounced phenological dynamics led to high correlations between the different products.”

Materials and Methods Section 2.6:  “...sparsely vegetated surfaces with very low vegetation cover and less pronounced yearly phenological dynamics over Western Australia.”

Results Section 3.4 : “The models' correlation consistency agreed closely over these land covers with R>0.6. Lower correlation was found over sparsely vegetated areas where pronounced phenological dynamics are absent.”

Section 3.5 : “Surfaces with considerable phenological dynamics, particularly deciduous broadleaf forests and agricultural areas,...”

Conclusions: “For all EVTs, the S3-TOA-GPR-1.0-WS produced generally consistent values over regions with pronounced yearly phenological dynamics,...”

3. Pearson correlation coefficients were used in the paper to illustrate the connections between products. However, the level of significance or P-value was not given.

Thank you for indicating this discrepancy in the manuscript. Pearson correlation associated global distribution of p-values have been mapped and subsequently added to the updated version of the manuscript. These maps are  presented in Figure A2 in Appendix A in the manuscript. 

The manuscript has been duly updated with information on p-values in Materials and Methods Section 2.7: “The P-value serves as an index measuring the strength of evidence against the null hypothesis, that no significance exists on the given observations. P-values lower than 0.05 indicate evidence against the null hypothesis being tested.”

Furthermore, in the Results Section 3.4, explanation was provided on the spatial distribution of the p-values.

4. Comparison over glass land was missing. Since glass land counts for a large portion of the earth surface and is not a spare vegetation land, it is important to look at the performance of the developed data over grass land.

Although grasslands were not explicitly investigated as a separate land cover in the study,  most of the Sparsely vegetated areas in Australia, that are subject to thorough analysis in the study, contain grasslands as well. 

The global land cover database used throughout the study was the Copernicus Global Land Cover: CGLS-LC100 Collection 3 on Google Earth Engine. This dataset does not specifically include a discrete classification for Grasslands. Furthermore, after investigating other land cover datasets, such as the ESA WorldCover 100, that features a discrete Grassland classification, it could be observed that most of the area of analysis for Sparse land covers in Australia are classified as “Grassland” by this dataset. 

5. What are spatial resolutions of maps displayed in Figure 7 & 8?

Thank you for pointing out this missing information. The spatial resolution (5km) of both figures are now added to the caption of Figures 7 and 8: “Pearson correlation coefficients for FAPAR (top) and LAI (bottom) on Figure 7 and FVC (top) and LCC (bottom) on Figure 8,  for the year 2019 between S3-TOA-GPR-1.0-WS models and datasets from MODIS and CGLS at 5km spatial resolution. The global mean and standard deviation of the R correlation are also given. Examples of three distinct correlation values are highlighted with their corresponding temporal profiles. Both WS reconstructed (blue line) and original S3-TOA-GPR-1.0 values (light blue dots) are plotted. The vertical lines indicate the start and end of the year 2019.”

6. The high resolution maps displayed in Figure 11 should be compared with the global map in 5-km to highlight the difference.

The 5 km spatial resolution maps over the Iberian peninsula have been added for all four EVTs in Figure A3 in Appendix A. Please see the attached Figure in the attached pdf document.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper titled "Global mapping of essential vegetation traits (EVTs) through data acquired by Earth-observing satellites" presents an important contribution to the field of vegetation analysis. The authors address the need for global and consistently derived multi-temporal trait maps that are cloud-free, and they propose a scalable processing approach to achieve this goal. The study focuses on four essential vegetation traits: fraction of photosynthetically active radiation (FAPAR), leaf area index (LAI), fractional vegetation cover (FVC), and leaf chlorophyll content (LCC). The authors introduce the processing chain and workflow for the production of the four EVTs at a global scale. They utilize hybrid retrieval models, named S3-TOA-GPR-1.0-WS, implemented into Google Earth Engine (GEE) using Sentinel-3 Ocean and Land Colour Instrument (OLCI) Level-1B data. The proposed approach allows for cloud-free mapping of EVTs with associated uncertainty estimates. The Whittaker smoother (WS) is used for temporal reconstruction, resulting in continuous data streams for the year 2019. The cloud-free maps are generated at a spatial resolution of 5 km and at 10-day time intervals. To evaluate the consistency and plausibility of the EVT estimates, the authors compare them with corresponding vegetation products from MODIS and Copernicus Global Land Service (CGLS). The results show that the EVT estimates, particularly for LAI, exhibit high intra-annual correlations with an average Pearson correlation coefficient (R) of 0.57 against the CGLS LAI product. The EVT products demonstrate consistent results globally, with higher correlations (R > 0.5) in the Northern Hemisphere between 30-60° latitude. Additionally, the authors calculate intra-annual goodness-of-fit statistics locally against reference products over four distinct vegetated land covers. They observe that areas with pronounced phenology fluctuations show higher correlations between the different products, while sparsely vegetated fields and equatorial regions with smaller seasonality display lower correlations. Overall, the study concludes that the global mapping of the four EVTs using the proposed methodology is consistent and reliable. The authors highlight the efficiency and scalability of Google Earth Engine, which enables the processing of the entire OLCI L1B catalogue for EVT production on a global scale while ensuring cloud-free maps through the WS temporal reconstruction method. Furthermore, they emphasize the operational applicability and accessibility of the workflow, making it available to the broader scientific community.

This research paper contributes significantly to the field of global vegetation analysis and mapping. The methodology presented offers valuable insights into the current vegetation states and dynamics of our planet. The study's findings and the availability of the processed EVT products will undoubtedly aid researchers, policymakers, and other stakeholders in understanding and managing global vegetation systems.

But I have specific questions need to answers:-

What is the purpose of global mapping of essential vegetation traits (EVTs)?

What are the four EVT variables that were mapped in this study?

How were the EVT maps produced and what satellite data was used?

What is the significance of having cloud-free EVT maps?

What is the Whittaker smoother (WS) and how was it used in this study?

Which year was the study focused on, and at what temporal resolution were the EVT maps produced?

How were the EVT estimates evaluated for consistency and plausibility?

Which vegetation product showed the most consistent results and what was the average Pearson correlation coefficient?

In which latitude range in the Northern Hemisphere did the EVT products show higher correlations with reference products?

How did the correlations between EVT products and reference products vary across different vegetated land covers?

Author Response

Dear Reviewer,

I am writing in response to the review you provided for our recently submitted manuscript in Remote Sensing. On behalf of my co-authors, we would like to express our gratitude for the time and effort you dedicated to providing a review. We would like to provide a response to Your questions regarding the submitted manuscript.

1. What is the purpose of global mapping of essential vegetation traits (EVTs)?

It is imperative that the essential vegetation traits are monitored globally, because the EVTs are quantifiable variables as opposed to indices that provide relative measures.

As opposed to vegetation indices, which are relative measures, EVTs are quantifiable variables with physical units and can be obtained during field campaigns for validation purposes, however, these field campaigns are difficult, expensive and tedious processes and only infer local data, global spatio-temporally continuous retrieval of EVTs via satellite observations is crucial. 

Information about the purpose of global mapping of EVTs can be found in the Introduction.

2. What are the four EVT variables that were mapped in this study?

The description and explanation of the four targeted EVTs mapped in this study can be found in the Introduction. Also, we would like to provide a succinct description here:

• Fraction of Absorbed Photosynthetically Active Radiation (FAPAR), which refers to the amount of incoming solar radiation absorbed by living green vegetation in the spectral range from 400–700 nm, divided by the total amount of radiation absorbed at the surface.
• Leaf Area Index (LAI), which  is defined as half of the total intercepting leaf area per unit ground surface area.  The variable is strongly related to canopy photosynthesis and evapotranspiration and plays an essential role in the exchange of energy and water between the biosphere and atmosphere.
• Fractional Vegetation Cover (FVC), which corresponds to the fraction of green vegetation as seen from nadir, and reflects the spatial extent of photosynthetic leaf areas. Generally, FVC is an important biophysical indicator required for modelling land surface processes, climate change, and numerical weather prediction.
• Leaf Chlorophyll Content (LCC). Leaf chlorophyll is a driver in the exchange of carbon, water and energy between the biosphere and the atmosphere The chlorophylls, Chl a and Chl b, are virtually essential pigments for the conversion of light energy to stored chemical energy.

3. How were the EVT maps produced and what satellite data was used?

Thank you for the question. We believe that we sufficiently described this in the manuscript. See for instance a summary in the abstract: 

“Hybrid retrieval models, named S3-TOA-GPR-1.0-WS, were implemented into Google Earth Engine (GEE) using Sentinel-3 Ocean and Land Colour Instrument (OLCI) Level-1B for the mapping of the four EVTs along with associated uncertainty estimates. We used the Whittaker smoother (WS) for the temporal reconstruction of the four EVTs, which led to continuous data streams, here applied to the year 2019. Cloud-free maps were produced at 5 km spatial resolution at 10-day time intervals.”

More details can be found in material and methods.

4. What is the significance of having cloud-free EVT maps?

Cloud-free EVT maps are of high importance, because they provide spatio-temporally continuous data streams over the whole planet. There is still a lack of global and consistently derived multi-temporal trait maps that are cloud-free, therefore the presented workflow shows novelties in this area.

This significance was highlighted in the Abstract and described throughout the manuscript.

5. What is the Whittaker smoother (WS) and how was it used in this study?

The WS function, which is based on penalized least squares, fits a discrete series to discrete data and balances the reliability of the observations with the roughness of the smoothed curve. In particular, the smoother allows continuous control over smoothness,works fast and is capable of automatic interpolation.  The GEE platform allows the implementation of the WS, so it is scalable to any time series data stream. The workflow for programming the WS presented by Khanal et al. [39] was taken as a basis and slightly modified to allow the function to run smoothly with S3-TOA-GPR-1.0 models in GEE. We chose to generate output maps at a 10-day temporally-composited resolution. 

The usage and the description of the Whittaker smoother was presented in the Materials and Methods Section 2.4.

6. Which year was the study focused on, and at what temporal resolution were the EVT maps produced?

2019. 10-day temporal resolution. This information is presented in the Abstract and throughout the manuscript.

7. How were the EVT estimates evaluated for consistency and plausibility?

We cross-compared our estimates of the S3-TOA-GPR-1.0-WS models against established reference products. A correlation analysis was run per pixel on a global scale along the temporal domain. We compared our results against Copernicus and MODIS products. Namely: MCD15A3H (LAI/FAPAR), MOD09A1v006 (LCC), Copernicus Global Land Service (LAI/FAPAR/FVC).

The description of the inter-comparison data sets are described in Materials and Methods Section 2.5.

8. Which vegetation product showed the most consistent results and what was the average Pearson correlation coefficient?

The most consistent results were obtained for LAI, which showed intra-annual

correlations with an average Pearson correlation coefficient (R) of 0.57 against the CGLS LAI product.

The information about the most consistently retrieved EVT can be found in the Abstract and throughout the manuscript.

9. In which latitude range in the Northern Hemisphere did the EVT products show higher correlations with reference products?

The EVT products showed consistent results, specifically obtaining higher correlation than R >0.5 with reference products between 30-60° latitude in the Northern Hemisphere.

The information about the latitudinal distribution of R correlation can be found in the Abstract and in the Results Section 3.4, where the latitudinal and longitudinal distribution of R correlation is depicted on the maps, see Figure 7 and 8.

10. How did the correlations between EVT products and reference products vary across different vegetated land covers?

As a general trend, vegetated land covers with pronounced phenology fluctuations led to high correlations between the different products. However, sparsely vegetated fields as well as areas near the Equator linked to smaller seasonality led to lower correlations.

Information about the correlation between different EVT products is mentioned in the abstract and more exhaustive analysis is performed in Section 3.5, where S3-TOA-GPR-1.0-WS products are intra-annually compared against reference products over 4 radically different land covers.

Author Response File: Author Response.pdf