CLWC-375: A national-scale lake water clarity dataset for China derived from VIIRS 375-m observations (2012&ndash;2025)

Chen, Tianzi; Cao, Zhigang; Wang, Menghua; Jiang, Lide; Shen, Ming; Liu, Dong; Duan, Hongtao

doi:10.5194/essd-2026-314

Preprints

https://doi.org/10.5194/essd-2026-314

Preprints

22 Jun 2026

| 22 Jun 2026

Status: this preprint is currently under review for the journal ESSD.

CLWC-375: A national-scale lake water clarity dataset for China derived from VIIRS 375-m observations (2012–2025)

Tianzi Chen, Zhigang Cao, Menghua Wang, Lide Jiang, Ming Shen, Dong Liu, and Hongtao Duan

Abstract. Lake water clarity, quantified as Secchi disk depth (SDD), serves as a useful indicator of aquatic ecosystem health and sensitive proxy for environmental shifts. As the MODIS mission nearly ends, maintaining long-term, high-frequency monitoring of inland waters becomes an urgent Earth observation challenge. Current alternatives face trade-offs between spatial resolution and temporal frequency. Addressing this critical observational gap, this study presents a lake water clarity dataset for China derived from VIIRS 375-m observations (CLWC-375), the first national-scale inland water clarity dataset (2012–2025) for 767 lakes larger than 10 km² across China, derived from the high-resolution VIIRS 375-m imagery band (I-band).

Our methodological framework utilizes a robust semi-analytical empirical model optimized for the remote sensing reflectance (R_rs(λ)) of the VIIRS Image-band (I1). Extensive independent validation against 954 in situ measurements from 74 lakes across a broad optical range (0.1 m to 16.0 m) demonstrates good retrieval accuracy. The customized regional algorithm achieves a high coefficient of determination (R² = 0.85) and uncertainty of 29.5 %, effectively eliminates systematic overestimation in existing global operational VIIRS products. The resulting 14-year, high-frequency dataset captures dynamic aquatic processes and reveals spatiotemporal trajectories. Morphology is related to optical properties: deep plateau lakes consistently have high clarity, whereas shallow lowland lakes have considerable turbidity. Chinese lakes had a significant overall clearing trend (0.28 m decade^-1) since 2012, driven predominantly by rapid improvements in the Tibetan and Yunnan-Guizhou plateaus. By successfully overcoming the spatial-temporal resolution trade-off, the CLWC-375 dataset provides a sustainable observational baseline for decoupling the complex optical responses of inland waters to intense anthropogenic activities and global climate change in the post-MODIS era. The dataset is publicly available at the ScienceDB repository (https://doi.org/10.57760/sciencedb.31441).

Received: 26 Apr 2026 – Discussion started: 22 Jun 2026

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 paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Tianzi Chen, Zhigang Cao, Menghua Wang, Lide Jiang, Ming Shen, Dong Liu, and Hongtao Duan

Status: final response (author comments only)

RC1:
'Comment on essd-2026-314', Anonymous Referee #1, 23 Jul 2026
This study provides an alternative approach for monitoring inland lakes with higher spatial and temporal resolution compared with VIIRS-M and Sentinel-3 OLCI, especially considering the recent degradation of MODIS observations. Overall, the methodology is reasonable, the results are reliable, and the current organization of the dataset is appropriate. The data format is well designed, with the products provided in NetCDF format following common practices in the community, which facilitates data accessibility, interoperability, and further applications. Providing the original source data and related information is also helpful for users to understand, evaluate, and reuse the dataset. However, several aspects, particularly regarding model accuracy and limitations, need further clarification. Below are my comments and suggestions.
Major comments:
I understand that the VIIRS-I band reflectance was generated through interpolation from VIIRS-M bands following Wang et al. (2018). Although the validation results appear satisfactory, more details regarding the interpolation procedure would be helpful. For example, how is the 375 m VIIRS-I reflectance derived using the 750 nm information? Are there any assumptions involved in this process? In addition, providing some examples of lake images comparing the 638 nm VIIRS-I band and the 671 nm VIIRS-M band would be helpful to evaluate the spatial patterns and reflectance characteristics.

The Secchi Disk Depth (SDD) model uses only one spectral band, and the suitability of the I1 band for this purpose is both interesting and somewhat fortunate. From a physical perspective, the red band has a strong relationship with suspended matter concentration, particularly in medium- to highly turbid waters (Neil et al., 2011, Remote Sensing of Environment, 115(12), 3719–3730). However, some Chinese lakes are relatively clear, where phytoplankton-related optical properties are generally considered dominant and blue-green bands are commonly used. Beyond the current validation results, it would be helpful to provide a stratified evaluation of model performance for high, medium, and low SDD levels.

Many clear-water lakes are large lakes, particularly those on the Tibetan Plateau. The blue-green bands from VIIRS-M may potentially provide higher accuracy for SDD retrieval in these waters. Although the manuscript mentioned that the performance may not be satisfactory (Line 506), I wonder whether a combined strategy using both VIIRS-M and VIIRS-I bands could improve SDD estimation for clear lakes. Since clear lakes usually have relatively homogeneous spatial characteristics, slightly reduced spatial resolution with improved spectral information may be a reasonable strategy. This approach does not necessarily need to be included in the current dataset, but it could be discussed as a potential direction for improving future products.

The current organization of the dataset is appropriate. Regarding data uncertainty, it would be helpful to include additional information in the distributed NetCDF files. Although empirical models do not provide formal uncertainty estimates, an additional layer describing the number of valid observations would be useful for users to evaluate data quality and determine whether the dataset is suitable for subsequent analyses.

Section 6.2 presents long-term SDD variations, and the results appear generally reasonable. It would be helpful to compare these trends with previously published studies to demonstrate whether the observed patterns are consistent or different. In addition, briefly discussing the potential drivers of long-term SDD changes would improve the interpretation of these results.

There are several spelling errors and inconsistencies throughout the manuscript. In addition, some references appear repeatedly in the text (e.g., Lines 65–67 and Lines 494–498), which may be related to EndNote formatting issues. Please carefully check and correct these problems. Specific suggestions are listed below.

Minor comments:
Line 34, the reported performance should be revised to “determination (R² = 0.87) and uncertainty of 27.34%” according to the main text.

Line 55, suggest correcting the typographical error “occu” to “occur”.

Lines 121–122, suggest revising the lake-zone names and abbreviations to “the Eastern Plain Lake Zone (EPL), the Northeast Plain and Mountain Lake Zone (NPML), the Yunnan-Guizhou Plateau Lake Zone (YGPL), the Inner Mongolia-Xinjiang Lake Zone (IMXL), and the Tibetan Plateau Lake Zone (TPL)” to ensure consistency with Fig. 1 and the terminology used elsewhere in the manuscript.

Line 124, suggest changing “The EPL and NPL” to “The EPL and NPML”.

Line 127, suggest changing “the MXPL is dominated” to “the IMXL is dominated” to use the revised abbreviation consistently.

Line 132, suggest deleting the stray backslash before the caption of Fig. 1 and changing “TPGL” to “YGPL”.

Line 164, suggest changing “Fig. 1” to “Fig. 3”, because the distributions of Chl-a, SPM, and SDD described here are presented in Fig. 3.

Line 172, suggest revising the caption to “Table 1. Statistical results of in situ water transparency (SDD) in different lake zones in China...”.

Line 242, suggest changing “Fig. 2” to “Fig. 4”.

Line 318, suggest changing the variable name “Zsd” to “SDD” and revising “Secchi Disc Depth” to “Secchi Disk Depth” in Table 2.

Line 352, suggest changing “Zhang et al. (2025)” to “Zhang et al. (2021)”.

Line 363, the font size of “a” in Fig. 7 is inconsistent with other plots.

Line 380, suggest changing “in the east China Plain (a and b)” to “in the east China Plain (a)”; changing “Tibetan Plateau (c and d) (August 8, 2015)” to “Tibetan Plateau (b) (August 8, 2015)”.

Line 397, “Changan” and “Bositeng” should be corrected to “Chagan” and “Bosten”.

Line 397, “January SDD” is unclear.

Line 424, suggest changing the study period from “2012 to 2021” to “2012 to 2025”.

Line 448, suggest adding the missing unit and revising “0.36 decade⁻¹” to “0.36 m decade⁻¹”.

Line 469, “Lui et al. (2020)” should be corrected to “Liu et al. (2020)”; “(Shen et al. 2020)” should be revised to “Shen et al. (2020)”.

Line 505, suggest correcting “the Tibetan Platea lakes” to “the Tibetan Plateau lakes”.
Citation: https://doi.org/10.5194/essd-2026-314-RC1
RC2: 'Comment on essd-2026-314', Anonymous Referee #2, 21 Sep 2026

This paper examines a VIIRS high resolution band for determination of water turbidity, based on Secchi depth for a number of lakes in China, showing patterns. This is an interesting result for these lakes, and shows potentially useful results with this satellite for turbidity in this region. The method suggests broader applicability of the VIIRS or the relationship to Secchi depth.

There are editorial and statistical issues. The paper needs a close read. Figures 3 and 4 are mislabeled. Figure captions do not explain what is in the figures.

Comments.

Figure “1” (which isn’t Figure 1) line 162, STD is meaningless for a log distributed data set. Report the IQR or another appropriate metric of dispersion.

Line 90 Wei et al, Wei et al, Wei et al, Wei et al. ????

3.5 Better statistics. statistical significance is not meaningful, and is not scientific or ecological significance (Wasserstein et al., 2019, American Statistician)
Use Seegers et al., 2018 for metrics, e.g., multiplicative error, and multiplicative bias. These are particularly necessary given the large range of data. MAPE is ok instead of multiplicative error, as long as it is small (e.g., < 50%) and is not used for algorithm comparison (MAPE biases towards the algorithms that underestimate). However, as Figure 7d has a MAPE > 100%, which is physically meaningless (implies negative values), use multiplicative error and bias throughout.

Lines 235-236 and Figure 5. P is not an error metric. P has no value, given the 906 samples, p<0.001 is achieved once Rsquared > 0.012, which is not “significant” by any standard.

Figure 6 and after. Rsquared scales the RMSE to the range of data, which is ok if you are building a model. Otherwise, R is useful for saying there is a general relationship (e.g., Figure 5, Figure 7a), but not to validate a model. Report the multiplicative error, or better yet, explain what the pink zone is. It’s probably an error metric, and may be a suitable one, like one that encompasses X% of the data.

For validation, an excellent model can have a low Rsquared, if the range is small. (And it is trivial to demonstrate a model with Rsquared =0.99 that is completely wrong). Because Rsquared scales to range, Rsquares can often not be compared. There is no reason to show a regression line for validation, unless it is explained what value it provides. Yes, people do it frequently, but that doesn’t make it valid or useful. A model that performs well except at the lowest (or highest) values may have a regression line that significantly deviates from 1, while a model that is severely biased in the middle of the range can have a slope of 1.

295-298. Kendall’s tau is not defined. That is an uncertainty metric. P is not. There is no explanation of what is done. Given the strong seasonality, was deseasonalized version used? (figure 13, and related).

Figures 5 and 6, the pink zone is not explained.
Line 265. Reference figure 7 here (The relationship of SDD to Rrs(638)).

As far as R2= 0.87, p<0.001. Report the error/uncertainty around the regression. Again p is meaningless. As a note, no metric should be reported that is not explained. The purported probability of a null hypothesis is not relevant to your hypothesis.

Figures 7 c and d. Use multiplicative error. A percentile error of > 100% is meaningless, this implies negative SDD. Explain the pink region.

Figure 8, what is black? Are the SDD statistics by pixel or by lake? Is this the mean of all pixels? STD is meaningless. The distribution is clearly not Gaussian, so STD is not only incorrect, but an STD of 3.06 with a mean of 1.17, means that negative SDD occur. The IQR would be a good choice.

Line 405. Change to “highlights substantive differences associated with lake morphology”.

Sen’s slope is appropriate for the trends. However, this should be deseasonalized. Was it? No information in the methods.
Figure 12. Remove the “statistical significance” arrows. What is of interest is how much the trend compares to the variability, or to the typical lake turbidity. There are lakes with “significant” trends that have a rate of < 0.1 m per decade (given there is only one decade of data). I’m skeptical that these are ecologically significant changes (the only significance that matters). What is the ratio of the change to the mean condition of the lake over the time period? A ratio of < 10-15% over the decade is not particularly significant. This gets close to Cohen’s d differences. The other question is how the trend compares to the variability, Kendalls’ tau threshold as a marker, e.g. tau > 0.3, or better the “trend to variability ratio”. No arrow required. This requires deseasonalized.

Abstract and Lines 443-445, Figure 13. First, is this mean trend based on the trend on each lake, vs all the pixels? Next a bulk trend is not informative. The mean is driven by a specific region, as noted. The trends should be presented as a distribution, what is the median in each region, and the IQR. Histograms might be useful. An issue, for example with YGPL, is that the summer is not getting better, but the winter, in bulk appears to have higher SDD, clearer water.

Citation: https://doi.org/10.5194/essd-2026-314-RC2

Tianzi Chen, Zhigang Cao, Menghua Wang, Lide Jiang, Ming Shen, Dong Liu, and Hongtao Duan

Data sets

CLWD-375: China's lake water clarity dataset from VIIRS 375-m observations (2012–2025) Zhigang Cao https://www.scidb.cn/s/mIz6ji

Tianzi Chen, Zhigang Cao, Menghua Wang, Lide Jiang, Ming Shen, Dong Liu, and Hongtao Duan

Viewed

Total article views: 372 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
199	86	87	372	121	80

HTML: 199
PDF: 86
XML: 87
Total: 372
BibTeX: 121
EndNote: 80

Views and downloads (calculated since 22 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	56	29	4	89
Aug 2026	34	13	2	49
Sep 2026	73	21	58	152
Oct 2026	36	23	23	82

Cumulative views and downloads (calculated since 22 Jun 2026)

Month	HTML	PDF	XML	Total
Jul 2026	56	29	4	89
Aug 2026	34	13	2	49
Sep 2026	73	21	58	152
Oct 2026	36	23	23	82

Viewed (geographical distribution)

Total article views: 346 (including HTML, PDF, and XML) Thereof 346 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 14 Oct 2026

Short summary

This study provides a new, long-term record of lake clarity across China from 2012 to 2025, covering 767 lakes. We carried out this work because existing satellite records are ending or lack the detail or frequency needed for continuous monitoring. Using daily observations from VIIRS (375 m), we developed a method to improve tracking. Results show many lakes have become clearer, while shallow lakes remain more turbid, providing a consistent basis to track lake change and inform water management.


Total:	0
HTML:	0
PDF:	0
XML:	0