WHU-SGCC: a novel approach for blending daily satellite (CHIRP) and precipitation observations over the Jinsha River basin

Shen, Gaoyun; Chen, Nengcheng; Wang, Wei; Chen, Zeqiang

doi:https://doi.org/10.5194/essd-11-1711-2019

Articles | Volume 11, issue 4

https://doi.org/10.5194/essd-11-1711-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

https://doi.org/10.5194/essd-11-1711-2019

© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Articles | Volume 11, issue 4

21 Nov 2019

| 21 Nov 2019

WHU-SGCC: a novel approach for blending daily satellite (CHIRP) and precipitation observations over the Jinsha River basin

Gaoyun Shen, Nengcheng Chen, Wei Wang, and Zeqiang Chen

Download

Final revised paper (published on 21 Nov 2019)
Preprint (discussion started on 02 Jan 2019)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'essd-2018-150', Anonymous Referee #1, 20 Feb 2019
- AC1: 'essd-2018-150-reply to Review 1', Nengchen Chen, 02 Apr 2019
RC2: 'Manuscript revision comments', Anonymous Referee #2, 21 Feb 2019
- AC2: 'essd-2018-150-reply to Review 2', Nengchen Chen, 02 Apr 2019
RC3: 'Review of WHU-SGCC: A novel approach for blending daily satellite (CHIRP) and precipitation observations over the Jinsha River Basin', Anonymous Referee #3, 21 Feb 2019
- AC3: 'essd-2018-150-reply to Review 3', Nengchen Chen, 02 Apr 2019
EC1: 'additional comments received by one of the referees', Giulio G.R. Iovine, 31 Jul 2019

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Nengchen Chen on behalf of the Authors (25 Apr 2019) Author's response

ED: Referee Nomination & Report Request started (02 May 2019) by Giulio G.R. Iovine

RR by Anonymous Referee #3 (16 May 2019)

RR by Anonymous Referee #4 (17 Jun 2019)

Suggestions for revision or reasons for rejection

The manuscript “WHU-SGCC:A novel approach for blending daily satellite (CHIRP) and precipitation observations over Jinsha River Basin” by Shen et al. describes an approach to blend and homogenize daily satellite (CHIRP) and precipitation surface observations over Jinsha River Basin. The approach is named WHU-SGCC.

The manuscript “WHU-SGCC: A novel approach for blending daily satellite (CHIRP) and precipitation observations over Jinsha River Basin” by Shen et al. describes an approach to blend and homogenize daily satellite (CHIRP) and precipitation surface observations over Jinsha River Basin. The approach is named WHU-SGCC.
The manuscript is quite well written though sometimes it is too schematic and should be instead, to my opinion, more discursive. The manuscript is in the scope of ESSD. However, I suggest below a few potential major revisions of the manuscript along with specific line-by-line suggestions or corrections.
General comments
1. The introduction of the dataset underlines the central role of the observations used as the “reference dataset” to build up the homogenized dataset with the needed bias adjustment. Nevertheless, surface rain gauge observations at present have a very low level of metrologic and uncertainty characterization, therefore the manuscript can benefit from a more accurate description of the real and potential limitations due to the use of rain gauge observations as the reference for the homogenization of other rain gauge data.
2. Table 6 and 7 allows the reader to quantify the better performances WHU-SGCC dataset compared to CHIRP and CHIRPS for rain event included in the class 0-10 mm, the moderate improvement for the in 10-25 mm class, while, as the authors themselves discussed in the manuscript, performances are negative for rain events larger than 25 mm. Though this class of rain events contains a minor number of events, the authors should discuss in the manuscript if this might indicate that extreme rain events in the Jinsha River Basin are attenuated in their intensity by the adopted homogenization approach.
3. The authors should comment more on the effect of and the reason why the pixel spatial autocorrelation (and the temporal autocorrelation only within the same season of the same year) is not considered and how its study can help to improve the quality of their results.
4. Table 6 shows that most of the data belongs to C4 pixel category (55-60% of the total dataset), given that these data are homogenized using the IDW technique based on the neighboring observations, the authors should clarify if the application of this technique on the majority of the dataset can smooth out extreme events (very low or very amount) due to the smoothing introduced at the C4 pixels. Moreover, it could be helpful to the highlight the position of C1-C3 pixels in Figure 4 to allow the reader to check the position of that part of the dataset where data homogenization is likely more robust.
5. In Figure 4, the spatial distribution of those statistical indicators used to study the WHU-SGCC performances is shown. It would be interesting to compare the results in Figure 4 with a similar figure showing the statistics for CHIRP. Though it is true that several table already summarizes the results of the statistics for WHU-SGCC, CHIRP and CHIRPS, a comparison of the reported statistics at the spatial level is missing.
6. The statistical analysis shows that CHIRPS is worse than CHIRP in terms of variability but similar in terms of bias: the authors should clarify what is the role of the representativeness uncertainty in the merging of gridded and in-situ point observations as happens in CHIRPS.
Specific comments
Line 11: please here and throughout the manuscript you cannot state that random errors are “removed”. I also encourage to have a consistent use of the terms “error” and “uncertainty”.
Line 21: replace “the” with “a”.
Line 62: it looks clear that the Jinsha River Basin is a region of interest for monitoring precipitation, not only for the authors; however, in order to justify the relevance of the dataset presented in the manuscript, it could be appropriate for the authors to say something more general about Jinsha River Basin and the interest to focus on this region as a special place to monitor monsoon precipitations.
Line 71: gauge-satellite fusing means “combined gauge-satellite products”?
Line 76: please check the style to report citation in the text.
Line 102: please add “weather and climate variations” instead of only climate.
Line 117-118: Discuss in more detailed the limitation of your “reference” dataset, here in the text.
Line 131: what does it indicates the word “strict” here? Please be more specific because this is critical to learn more about the data quality used as “reference” in your approach.
Line 159: “longer length” instead of “longer-term”
Line 195: This sentence should be supported by metadata, a reference, maps, or photos.
Line 196: Are you assuming that there are no outliers? Please explain more.
Line 219-220: these seems to be a repetition of previous text, please check.
Line 246: can the authors validate their cluster analysis with a GIS map, for example, showing the terrain features?
Line 295-296: “Excluding the C1, C2 and C3 pixels, the number of remaining pixels, called C4 pixels which are adjusted by Inverse Distance Weighted (IDW)” there is something in this sentence.
Line 321-322: the word strength here is incorrect. SCC measures if a process is linear or not, you cannot infer more than this.
Line 350: To perform the accuracy evaluation, did you use the 30% of in-situ observation mentioned at the beginning? From this section, this is not clear and how this fraction of observation was selected, is it nor clear as well.

Hide

RR by Anonymous Referee #5 (20 Jun 2019)

RR by Anonymous Referee #6 (01 Jul 2019)

RR by Anonymous Referee #7 (03 Jul 2019)

Suggestions for revision or reasons for rejection

L11: The authors mention that the existing blending algorithms are very bad at removing random errors at the daily scale. However, in the introduction, the authors only mention three articles related to blending techniques. The authors must present a better literature review. I do not agree that all the algorithms are very bad at removing random errors at the daily scale; see MSWEP as an example. MSWEP is a merged precipitation product (using rain gauges, reanalysis, and satellite-based precipitation products) which has good performance in many areas.

L70: The authors mention that the articles presented between L67-69 evaluated the products mainly at monthly and seasonal scales lacking the evaluation of daily precipitation. In the case of Chile, Zambrano-Bigiarini et al., (2017) include a daily evaluation and Yang et al., (2016) is a study of bias correction and its also at daily scale.

Table 1: I think that MSWEP should be included in the study because is a state-of-the-art merged product, it can be compared to the product obtained applying this method.

L122: Here the authors mention that: 'because of the Jinsha River Basin has more precipitation during the summer season the blending of satellite estimations with gauged observations during JJA is the main focus of the research'. I completely disagree with this... If the authors are presenting a novel approach for blending ground-based measurements with a satellite product it has to be evaluated throughout the yea and not only for one season.

L196: Assumption number three states that the correlation coefficient can indicate the statistical similarity of rainfall characteristics among pixels. What would happen if two different areas show similar correlation values but different precipitation amounts?

The authors present four classes of pixels: C1 (pixels with at least 1 station); C2 (pixels statistically similar to C1); C3 (similar to C2); and C4 (the rest of the pixels): How does the total precipitation amount is considered when the authors apply the RF models to the C2 pixels?

The authors mention that the development of methods for high-accuracy precipitation estimates over complex terrain is of vital importance, how is this considered in the method?

The amount of C4 pixels is around 60%, and for these pixels the authors apply an IDW interpolation, how do the authors justify that some pixels are blended with machine learning and others are simply interpolated? Which are the expected errors of this methodology for each type of pixels (C1-4)? What happen with the small scale convective events (which are present in summer) neover the C4 pixels?

The authors should present a map with the mean annual precipitation for the study area using CHIRP and the blended product. This would be interesting because it would help the reader to see if the transition of precipitation between the different classes of pixels is smooth.

Table 5: which values are represented here? mean, median?

L360: The authors mention an increase of NSE of 93% which sounds very good, however, this value is still negative. The authors should present the values and not the percentage of improvement as it is relative to the performance of the product and it can be misleading.

Table 5: There is some improvement in POD, however, the PCC and the BIAS are almost similar. The categorical indices (POD, FAR, and CSI) can be decomposed to different precipitation intensities in order to see if the method improves the detection of medium and high intensities or only the no-rain events which are higher in number.

L381: The authors mention that the presented approach is still efective in adjusting the satellite biases. However, in L357 it is mentioned that the bias showed no significant improvement (which is in agreement with Table 5).

Figure 4a and 4d: The legend should be improved. The correlation (PCC) goes from 0.1-0.2; 0.2-0.3 and then just >0.3. If 1.0 is the maximum and 0.3 is still low correlation why not showing the complete ranges?

L393: If PCCs of WHU-SGCC, CHIRP and CHIRPS areroughly the same, then there was no improvement in the daily correction of events. Can the authors discuss this?

L403: The authors mention that the WHU-SGCC method seems to be more promising in detecting precipitation compared to CHIRP and CHIRPS altough it performs poorly when evaluated with the FAR. However the objective is to provide high quality precipitation estimates.

Figure 5: There are not labels in each boxplot.

Table 7: Please include POD, FAR, and CSI to see the categorical performance over different rainfall intensities.

Hide

ED: Reconsider after major revisions (16 Jul 2019) by Giulio G.R. Iovine

AR by Nengchen Chen on behalf of the Authors (27 Aug 2019) Author's response Manuscript

ED: Referee Nomination & Report Request started (27 Aug 2019) by Giulio G.R. Iovine

RR by Anonymous Referee #7 (04 Sep 2019)

Suggestions for revision or reasons for rejection

The authors have substantially improved this version of the manuscript. They have taken all my suggestions into account and therefore, I recommend the publication of this manuscript after addressing the following points.

Figure 5: Here the authors show Pearson's correlation coefficients. I believe that the legend is a bit misleading. Generally, the green is used to represent well performance but the authors use it in the range between 0.2 and 0.5. It would be better to use a two-color palette (e.g., from red to blue).

There is still not clear the influence in the number of gauge stations in the approach. I assume that the performance of this method will decrease with a reduction in the number of C1 pixels. The authors need to include an evaluation of the performance of the method with a varying number of rain gauge stations.

One of my main concerns is that the main objective of this approach is to improve the characterization of precipitation over mountainous regions. However, ~50-60% of the pixels are classified as C4 and interpolated using IDW which does not account for the influence of elevation in the interpolation of precipitation. The authors may apply another interpolation method that accounts for the precipitation gradient related to elevation.

Minor comments:

Line 10: What do the authors mean with "existing fusion precipitation estimates"?

Line 27: might be due to the ...

Line 28: events.

Please add a space between the references.

Line 55: Please replace "spacing" with "spatial resolution".

Line 89: evaluated and indicated.

Table 1: Some of the temporal resolutions of the products start with uppercase and some of them with lowercase. Also, the period of CMORPH is not complete.

Line 112: probably present fits better than demonstrate.

Line 121: replace the comma with "is".

Lines 135-136: I think that now that the research is not focused in summer, it is better to remove this sentence.

Line 178: Recommendation, replace "in situ" with ground-based.

Line 194: remove "is to".

Recommendation: change "multi-year" for "multi-annual" throughout the manuscript

Hide

RR by Anonymous Referee #6 (06 Sep 2019)

RR by Anonymous Referee #4 (15 Sep 2019)

Suggestions for revision or reasons for rejection

After the last review stage, the authors largely revised the manuscript to improve its quality and to show quantitatively what are the qualities of WHU-SGCC vs CHIRP and CHIRPS datasets. I think the manuscript has reached a fair quality and it can be subject to minor revisions, though a few of the comments reported below, if not seriously address, can definitely affect the consistency of the manuscript and contains serious mistakes.

First of all, concepts of errors and uncertainties are still not correctly used in the manuscript and I have the impression the use of the word “error” is sometimes abused by the authors which also mention in their reply the words “more errors”, which definitely makes no sense.
Error is the difference between the true value of the measurand and the measured value. Accuracy is an expression of the lack of error. Uncertainty characterizes the range of values within which the true value is asserted to lie with some level of confidence.
Looking at the abstract I am amazed to see at line 17 the term “error adjustment”. Error is error and cannot be adjusted. Values of climate variables can be adjusted if the presence of systematic effects is detected. At line 23-24, the authors state that “precipitation errors” can be reduced. May be the authors are talking about the RMSE which is not the error. Going through, the text I can provide other examples.
I recommend again to define all the quantities discussed in the manuscript in a proper way. This is mandatory to my opinion and the authors can also check the GUM (https://www.bipm.org/en/publications/guides/gum.html) to stay consistent.

It is also not clear to me why a few information reported in the reply to the reviewers are not provided also in the manuscript. For example, the authors could describe the way CMA imposes strict quality control on rain gauge observations.
In their comments the authors say:
“The process of quality control is as follows:
(1) Climate threshold or allowable value check;
(2) Extreme values at gauge stations check;
(3) Internal consistency check between fixed value, daily average value and daily extreme value;
(4) Time consistency check;
(5) Manual verification and correction;
this quality control approach is provided by the official document from CMA. So, the daily rain gauge observations were used as the “Reference dataset”.

I think the information above must be provided when the CMA dataset is introduced in the manuscript to increase confidence in this dataset.

The authors spent an appropriate effort to improve the quality of the analysis used and to smooth the tone of the writing which was often too enthusiastic in the previous version. Issues which may affect the quality of the dataset are now better highlighted though this is done still in a qualitative way. Generally, this may be sufficient to allow the reader to properly use the WHU-SGCC dataset; nevertheless, wherever possible, I recommend the authors to be more quantitative.
For example, when the authors say: “In addition, the spatial distribution of C2 and C3 pixels also significantly impact the overall accuracy in different seasons that the most uniform in the fall, while the sparsest in the winter.”, it would be desirable to report a quantification of the impact which the authors are referring to. This happens also in other parts of the manuscript.

Other comments:
1. I am not sure why the authors speak about "attenuating the simulation” of extreme rain events at line 29 and other times thereinafter; I’d simply write “attenuating extreme rain events”.
2. Check if all the acronyms are reported, for example NSE is not introduced in the abstract
3. At line 235-236 the sentence could be clearer, adding the term “separately” at the end of the sentence.
4. At line 260, there is the assumption of “no long-held outliers”, whose impact is anyhow not assessed in the manuscript. A few words more to demonstrate the robustness of this assumption would be useful.
5. Table 3 shows interesting information which if reported in a plot style would be much clearer and useful for the reader, otherwise is quite confusing
6. Line 627-632: this is one of the points where the limitations affecting the WHU-SGCC dataset are discussed. I think that, at least for this case, the paragraph should be reported also in the conclusions to give the right balance to the final section of the manuscript.
7. Figure 8: this is the core of the presented analysis and it is not clear why the WHU-SGCC dataset has almost always a negative bias, while CHIRP and CHIRPS has a positive bias. The authors should comment more on this important aspect of their homogenization approach.
8. In addition, Figure 8 clearly states that CHIRP and WHU-SGCC have very similar performance, expect for the bias (which for WHU-SGCC is slightly better in absolute value). This should be clearly stated in the analysis and in the conclusions.

Finally, I recommend an appropriate use of decimal places throughout the presented analysis. The manuscript can also benefit from a final English review (I recommend).

Hide

ED: Publish subject to minor revisions (review by editor) (16 Sep 2019) by Giulio G.R. Iovine

AR by Nengchen Chen on behalf of the Authors (25 Sep 2019) Author's response Manuscript

ED: Publish as is (26 Sep 2019) by Giulio G.R. Iovine

AR by Nengchen Chen on behalf of the Authors (01 Oct 2019) Manuscript

Post-review adjustments

AA: Author's adjustment | EA: Editor approval

AA by Gaoyun Shen on behalf of the Authors (07 Nov 2019) Author's adjustment Manuscript

EA: Adjustments approved (18 Nov 2019) by Giulio G.R. Iovine

Short summary

The development of effective methods for high-accuracy precipitation estimates over complex terrain and on a daily scale is important for mountainous hydrological applications. This study offers a novel approach called WHU-SGCC by blending rain gauge and satellite data to estimate daily precipitation at 0.05° resolution over the Jinsha River basin, the complicated mountainous terrain with sparse rain gauge data, considering the spatial correlation and historical precipitation characteristics.