the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Jan 2025
CNSIF: A reconstructed monthly 500-m spatial resolution solar-induced chlorophyll fluorescence dataset in China
Abstract. Satellite-derived solar-induced chlorophyll fluorescence (SIF) offers valuable opportunities for monitoring large-scale ecosystem functions. However, the inherent trade-off between satellite scan range and spatial resolution, along with incomplete spatial coverage and irregular temporal sampling, limits its broader application. In this study, we developed a 500-m spatial resolution monthly SIF dataset for the China region (CNSIF) from 2003 to 2022, using a data-driven deep learning approach based on high-resolution apparent reflectance and thermal infrared data. The results indicate that CNSIF effectively captures the spatial patterns of vegetation photosynthetic activity and exhibits a positive annual growth trend of 0.054. Comparisons with tower-based observations validated the ability of CNSIF to track changes in photosynthetic intensity over time across different ecosystems. Furthermore, the strong correlation (R2_2016 = 0.768, R2_2020 = 0.743; P<0.001) between CNSIF and the MODIS monthly Gross Primary Production (GPP) product demonstrates its potential for estimating carbon flux. CNSIF's higher-resolution estimation of photosynthetic activity offers a promising tool for monitoring vegetation dynamics across China and estimating fragmented agricultural production. It enables the incorporation of ecosystem fragmentation effects into earth observation and carbon cycle systems. The CNSIF dataset is available at https://doi.org/10.6084/m9.figshare.27075145 (Du et al., 2024).
Competing interests: At least one of the (co-)authors is a member of the editorial board of Earth System Science Data.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.- Preprint
(2266 KB) - Metadata XML
- BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on essd-2024-432', Anonymous Referee #1, 18 Feb 2025
This study developed a downscaled CNSIF product for China using deep-learning methods and multi-source datasets. This dataset at lower spatial resolution is helpful for smaller scale application of SIF. However, this CNSIF product is developed based on intermediate SIF data product (GOSIF), which could weaken the precision and application of this dataset. In addition, the evaluation method of the model performance is not professional too. Furthermore, some key information in Method is lacked. Some other comments/suggestions are shown below:
- The English writing need an extensive improvement.
- Line 16: “the China region” should be revised to “China”. Check similar expressions throughout the MS and make revisions.
- The CNSIF is based on the downscaled GOSIF product. During the development GOSIF, some errors have been brought, so based on GOSIF to generate the CNSIF could greatly expand the errors (e.g., uncertainties from MODIS EVI data, extrapolation method and climate data). Suggest using the original data for GOSIF product to develop this CNSIF dataset.
- P7 Line 1: why the “median” values are used for both SR and LST?
- How to generate the continuous monthly images during 2003-2022 based on the Landsat images? Descriptions are needed.
- P7 Line 3: what is the resample method?
- Section 3.1: The data from GOSIF can not be used to evaluate your dataset accuracy since your product is based on the GOSIF (statistically not independent). You can discuss the consistency though. Instead, the evaluations should be based on the site-level observational SIF dataset (Section 3.4).
- Sections 3.3 and 3.4 have the same titles.
- Section 3.3: As I commented above, not necessary to make comparisons between GOSIF and CNSIF. You should focus on analyzing the spatial and temporal change patterns in CNSIF.
- Section 3.4: This section actually belongs to model evaluation (move to Section 3.1 and rewrite). In addition, the pixel-level GOSIF and CNSIF can put together to compare with site level SIF.
- 6: what the meanings for the background colors? An explanation should be added in the caption. In addition only 7 sites show here. Why does not also show other 2 sites?
- Section 3.6: It is not necessary to compare CNSIF with the MODIS GPP product in this manuscript. In addition, the correlation coefficient between site GPP and CNSIF is very low, but why the correlation coefficient is significantly higher as compared with the MODIS GPP?
- P12 Line 2-3: this sentence should be moved to the Method.
- Section 4.1: This is actually not a discussion. The determination of the optimal window should be put in the Method. Also, I am curious how do you prove that your dataset has reduced uncertainty?
- A comment: you can discuss similarity/dissimilarity of the spatiotemporal patterns among different SIF products (except GOSIF).
- Section 4.2: A deeper discussion is needed to discuss the effective downscaling methods (e.g., statistical, machine-learning & deep-learning methods, variable selections).
Citation: https://doi.org/10.5194/essd-2024-432-RC1 -
RC2: 'Comment on essd-2024-432', Anonymous Referee #2, 24 Feb 2025
The authors developed a monthly 500-meter SIF dataset in China which is helpful to identify plant stress and for evaluating crop yield. However, the validation process is not solid, which prevent a comprehensive uncertainty analysis.
As the data is developed from GOSIF, it's not reasonable to use GOSIF as a validation dataset. Instead, the author might want to evaluate the improvement of their data upon the GOSIF data in capturing the tower SIF dynamics.GOSIF is a reconstructed SIF, which contains its own uncertainty and weakens the precision and application of the authors' dataset built upon it. Why not use the original observations from OCO?
I also don't understand the part of using MODIS GPP to evaluate their SIF dataset. It's natural that the GPP would be correlated with your SIF data as the input variables should be correlated with GPP. To me, the only relevant observations for validation is the tower SIF (or probably the tower GPP can be a qualitative reference).
The spatial patterns of their SIF product is important, but the authors mostly focused just on the large-scale patterns (which GOSIF can also capture), but the spatial details (what is beyond GOSIF's ability to show), which are their improvement, should be discussed and validated more.
Fig. 3. What is the reference SIF? GOSIF? As I mentioned, GOSIF should not be treated as a reference.
Any explanations on the large deviations between your SIF and GOSIF at high-latitude?
The surface reflectance and thermal infrared data before and after 2017 are from two satellite. How would this affect the inter-annual and trending analysis?
The r-square between site SIF and their SIF is not high
The author may also want to compare their data with other reconstructed SIF product at those SIF towers. Also, there are existing high-resolution SIF dataset for China but were not recognized in the paper. e.g., [Hu J, Jia J, Ma Z, Yuan K, Yu H, Liu L. A spatially downscaled TROPOMI SIF product at 0.005 degree resolution with bias correction. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2024 Jul 25.]; another one for globe but also with high resolution: [Chen S, Liu L, Sui L, Liu X, Ma Y. An improved spatially downscaled solar-induced chlorophyll fluorescence dataset from the TROPOMI product. Scientific Data. 2025 Jan 22;12(1):135.]Citation: https://doi.org/10.5194/essd-2024-432-RC2
Status: closed
-
RC1: 'Comment on essd-2024-432', Anonymous Referee #1, 18 Feb 2025
This study developed a downscaled CNSIF product for China using deep-learning methods and multi-source datasets. This dataset at lower spatial resolution is helpful for smaller scale application of SIF. However, this CNSIF product is developed based on intermediate SIF data product (GOSIF), which could weaken the precision and application of this dataset. In addition, the evaluation method of the model performance is not professional too. Furthermore, some key information in Method is lacked. Some other comments/suggestions are shown below:
- The English writing need an extensive improvement.
- Line 16: “the China region” should be revised to “China”. Check similar expressions throughout the MS and make revisions.
- The CNSIF is based on the downscaled GOSIF product. During the development GOSIF, some errors have been brought, so based on GOSIF to generate the CNSIF could greatly expand the errors (e.g., uncertainties from MODIS EVI data, extrapolation method and climate data). Suggest using the original data for GOSIF product to develop this CNSIF dataset.
- P7 Line 1: why the “median” values are used for both SR and LST?
- How to generate the continuous monthly images during 2003-2022 based on the Landsat images? Descriptions are needed.
- P7 Line 3: what is the resample method?
- Section 3.1: The data from GOSIF can not be used to evaluate your dataset accuracy since your product is based on the GOSIF (statistically not independent). You can discuss the consistency though. Instead, the evaluations should be based on the site-level observational SIF dataset (Section 3.4).
- Sections 3.3 and 3.4 have the same titles.
- Section 3.3: As I commented above, not necessary to make comparisons between GOSIF and CNSIF. You should focus on analyzing the spatial and temporal change patterns in CNSIF.
- Section 3.4: This section actually belongs to model evaluation (move to Section 3.1 and rewrite). In addition, the pixel-level GOSIF and CNSIF can put together to compare with site level SIF.
- 6: what the meanings for the background colors? An explanation should be added in the caption. In addition only 7 sites show here. Why does not also show other 2 sites?
- Section 3.6: It is not necessary to compare CNSIF with the MODIS GPP product in this manuscript. In addition, the correlation coefficient between site GPP and CNSIF is very low, but why the correlation coefficient is significantly higher as compared with the MODIS GPP?
- P12 Line 2-3: this sentence should be moved to the Method.
- Section 4.1: This is actually not a discussion. The determination of the optimal window should be put in the Method. Also, I am curious how do you prove that your dataset has reduced uncertainty?
- A comment: you can discuss similarity/dissimilarity of the spatiotemporal patterns among different SIF products (except GOSIF).
- Section 4.2: A deeper discussion is needed to discuss the effective downscaling methods (e.g., statistical, machine-learning & deep-learning methods, variable selections).
Citation: https://doi.org/10.5194/essd-2024-432-RC1 -
RC2: 'Comment on essd-2024-432', Anonymous Referee #2, 24 Feb 2025
The authors developed a monthly 500-meter SIF dataset in China which is helpful to identify plant stress and for evaluating crop yield. However, the validation process is not solid, which prevent a comprehensive uncertainty analysis.
As the data is developed from GOSIF, it's not reasonable to use GOSIF as a validation dataset. Instead, the author might want to evaluate the improvement of their data upon the GOSIF data in capturing the tower SIF dynamics.GOSIF is a reconstructed SIF, which contains its own uncertainty and weakens the precision and application of the authors' dataset built upon it. Why not use the original observations from OCO?
I also don't understand the part of using MODIS GPP to evaluate their SIF dataset. It's natural that the GPP would be correlated with your SIF data as the input variables should be correlated with GPP. To me, the only relevant observations for validation is the tower SIF (or probably the tower GPP can be a qualitative reference).
The spatial patterns of their SIF product is important, but the authors mostly focused just on the large-scale patterns (which GOSIF can also capture), but the spatial details (what is beyond GOSIF's ability to show), which are their improvement, should be discussed and validated more.
Fig. 3. What is the reference SIF? GOSIF? As I mentioned, GOSIF should not be treated as a reference.
Any explanations on the large deviations between your SIF and GOSIF at high-latitude?
The surface reflectance and thermal infrared data before and after 2017 are from two satellite. How would this affect the inter-annual and trending analysis?
The r-square between site SIF and their SIF is not high
The author may also want to compare their data with other reconstructed SIF product at those SIF towers. Also, there are existing high-resolution SIF dataset for China but were not recognized in the paper. e.g., [Hu J, Jia J, Ma Z, Yuan K, Yu H, Liu L. A spatially downscaled TROPOMI SIF product at 0.005 degree resolution with bias correction. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2024 Jul 25.]; another one for globe but also with high resolution: [Chen S, Liu L, Sui L, Liu X, Ma Y. An improved spatially downscaled solar-induced chlorophyll fluorescence dataset from the TROPOMI product. Scientific Data. 2025 Jan 22;12(1):135.]Citation: https://doi.org/10.5194/essd-2024-432-RC2
Data sets
CNSIF_Monthly_2003-2022 Kaiqi Du https://doi.org/10.6084/m9.figshare.27075145
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 379 | 49 | 5 | 433 | 7 | 5 |
- HTML: 379
- PDF: 49
- XML: 5
- Total: 433
- BibTeX: 7
- EndNote: 5
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1