the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 09 Sep 2024
Evaluation of remote sensing and reanalysis based precipitation products for agro-hydrological studies in semi-arid tropics of Tamil Nadu
Abstract. This study provides a comprehensive evaluation of eight high spatial resolution gridded precipitation products in semi-arid regions of Tamil Nadu, India, focusing specifically on Coimbatore, Madurai, Tiruchirappalli, and Tuticorin, where both irrigated and rainfed agriculture is prevalent. The study regions lack sufficiently long-term and spatially representative observed precipitation data, essential for agro-hydrological studies and better understanding and managing the nexus between food production and water and soil management. Hence, the present study evaluates the accuracy of five remote sensing-based precipitation products, viz. Tropical Rainfall Measuring Mission (TRMM), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks – Climate Data Records (PERSIANN CDR), CPC MORPHing technique (CMORPH), Global Precipitation Measurement (GPM) and Multi-Source Weighted-Ensemble Precipitation (MSWEP) and three reanalysis-based precipitation products viz. National Center for Environmental Prediction Reanalysis 2 (NCEP2), and European Centre for Medium-Range Weather Forecast (ECMWF) Reanalysis version 5 Land (ERA5-Land), Modern-Era Retrospective analysis for Research and Application version 2 (MERRA 2) against the station data. Linearly interpolated precipitation products were statistically evaluated at two spatial (grid and district-wise) and three temporal (daily, monthly, and yearly) resolutions for 2003–2014. Based on overall statistical metrics, ERA 5 Land was the best-performing precipitation product in Coimbatore, Madurai, and Tiruchirappalli, with MSWEP closely behind. In Tuticorin, however, MSWEP outperformed the others. On the other hand, MERRA2 and NCEP2 performed the worst in all the study regions, as indicated by their higher Root Mean Square Error (RMSE) and lower correlation values. Except in Coimbatore, most precipitation products underestimated the monthly monsoon precipitation, which highlights the need for a better algorithm for capturing the convective precipitation events. Also, the Percent Mean Absolute Error (%MAE) was higher in non-monsoon months, indicating that these product-based agro-hydrological modeling, like irrigation scheduling for water-scarce periods, may be less reliable. The ability of precipitation products to capture the extreme precipitation intensity differed from the overall statistical metrics, where MSWEP performed the best in Coimbatore and Madurai, PERSIANN CDR in Tiruchirappalli, and ERA5-Land in Tuticorin. This study offers crucial guidance for managing water resources in agricultural areas, especially in precipitation data-scarce regions, by helping to select suitable precipitation products and bias correction methods for agro-hydrological research.
- Preprint
(2238 KB) - Metadata XML
-
Supplement
(803 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on egusphere-2024-2369', Suraj Shah, 10 Sep 2024
It has been almost a decade; gauge-free evaluation methods for satellite precipitation or soil moisture are available, like the TC method. In such a scenario, I suppose researchers use them as a linear interpolation of the gauge data and comparing them again only introduces a new layer of uncertainty. What would the author like to say about this matter?
Citation: https://doi.org/10.5194/egusphere-2024-2369-CC1 -
AC1: 'Reply on CC1', Aatralarasi Saravanan, 13 Sep 2024
Dear reviewer,
Please find attached our detailed response to your comment. We greatly appreciate the time and effort you took to share your perspectives with us.
-
CC2: 'Reply on AC1', Suraj Shah, 13 Sep 2024
Thanks for the early reply.
Yes, the central assumption of the TC is independent of the inputs for the correct calculation of the error covariance matrix. However, the previous question is about the efficacy of using linear interpolated gauge data for validity. The main reason why I stressed that question is that at the end, the author is tallying the relative performance of each product; in such cases, TC or a similar method, SNR_opt method (which can handle non-zero error cross-correlation), can be more effective in comparison for non gauged area as it offers direct pixel by pixel comparison of the data. Please read (Duan et al. 2021, and Lu et al. 2021) for a detailed understanding of what the previous question was indicating.
The paragraph "Previous studies evaluating TC ......" makes no sense.
In addition, the author talks about (zero-inflated data) mainly effects while using a multiplicative error model, but we can use an additive error model.
Lastly, why did the author not use Aphrodite? Is it not available? If available, it should be more correct.
I see the author's strong disagreement with the proposed gauge free. However, these stuff should be discussed in the Discussion section rather than repeating results. I hope the editor or reviewer has some focus on the above topic.
Citation: https://doi.org/10.5194/egusphere-2024-2369-CC2 -
AC5: 'Reply on CC2', Aatralarasi Saravanan, 03 Dec 2024
We thank you for your response. The multiplicative error model is an improvement over the additive error model. Hence, its assumptions were considered which limits choosing multiple precipitation products.
The performance of linear interpolation was tested with the LOOCV method, and results are attached in Reviewer 1 and Reviewer 2’s response.
In accordance with your suggestion, the ‘Discussion’ section will be revised to emphasize the reason behind choosing this methodology over TC.
Citation: https://doi.org/10.5194/egusphere-2024-2369-AC5
-
AC5: 'Reply on CC2', Aatralarasi Saravanan, 03 Dec 2024
-
CC2: 'Reply on AC1', Suraj Shah, 13 Sep 2024
-
AC1: 'Reply on CC1', Aatralarasi Saravanan, 13 Sep 2024
-
RC1: 'Comment on egusphere-2024-2369', Anonymous Referee #1, 06 Oct 2024
Thank you for giving me the opportunity to review this article.
General comments:
The paper: "Evaluation of remote sensing and reanalysis-based precipitation products for agro-hydrological studies in semi-arid tropics of Tamil Nadu" addresses a very innovative topic, especially in the era of remote sensing where satellite products must be rigorously validated before any hydrological application. The study focuses on the reliability of these products in the semi-arid environment of Tamil Nadu.
"Hydrology and Earth System Sciences" is indeed an appropriate journal for the article.
Specific comments :
Line 87. When you mention GPM, are these IMERG data? GPM-IMERG is generally used.
Line 140. Are the rain gauges (69 rain gauges) used for validation included in the GPCC? If not, please specify.
In Table 2, IMERG-GPM can go up to hourly.
Why is there redundancy on the study regions in the methodology and study area in part 2?
Figure 1. It's a bit strange that the ground stations are very aligned. How is this possible? There are rain gauges in the mountains even though the area is supposed to be difficult to access.
Line 253. Doesn't this already introduce a large bias for the data used for validation? In other studies, "point-gridded" is used. Because the location of rain gauges most often does not coincide with gridded precipitation products (GPP) grid centroids, a second strategy was implemented: the point-gridded approach. In practice, a cell is delineated around each rain gauge (cell size of 0.04, 0.05, or 0.1°… depending on the GPP; Table. 2). Then, the rainfall value in those new cells was estimated as the area-weighted mean (max. 4) of the GPP grid cells overlapping with the new cell.
Results.
Grid scale/district scale method are not really mentioned in the methodology part.
Monsoon/Non-Monsoon not clear in the methodology.
Discussion
The first paragraph seems out of place and confuses the reader. Start directly by discussing the results.
Line 665. Why are ERA-5 Land and MSWEP performing better than others? Algorithms used? Reanalyzed products? Is this the case for other studies?
Line 687. From what rainfall intensity does ERA-5 Land struggle to detect?
Line 694 – 704. The entire paragraph should be dispatched into other paragraphs to properly explain the reasons for the performance of precipitation products compared to others."
Technical comments :
Line 266. What do all the terms mean? *Pi, …*
Line 283. Equation 7 not mentioned in the text.
Same for Eq. 8, 9, 10
Line 290. Equation 11 instead of Equation 7.
Equation 12 not mentioned in the text.
Line 302. Equation 13 instead of Equation 9.
Review all Equation numbers.
Citation: https://doi.org/10.5194/egusphere-2024-2369-RC1 -
AC2: 'Reply on RC1', Aatralarasi Saravanan, 02 Dec 2024
We express our gratitude to the referees and the handling editor for their valuable time in helping improve this manuscript. In our opinion, the referees’ input will considerably improve the revised manuscript. Here will outline all changes we will add to the initial manuscript. This is our initial response. Planned activities are also included in the document and will all be implemented.
- Restructuring subsections “Study region and Ground station”
- The ‘Datasets’ section will be revised only highlighting important information
- Inclusion of methodology – Grid and District scale evaluation, Monsoon and Non-monsoon month evaluation
- Inclusion of all the equation numbers and abbreviations in the methodology
- Consistent use of Acronyms, e.g. ‘GPM-IMERG’
- Tables are revised to include only Correlation Coefficient (CC), Root Mean Square Error (RMSE) and RB (Relative Bias)
- Revised Figure. 2 to include only IDF for 5-year Return Period
- Restructuring the ‘Discussion’ section by describing the results with related studies followed by the limitations of the present study.
- Conclusion will be revised and include an outlook of the present study’s impact in other data-scarce regions in India.
We added a response to the reviewer’s individual comments in the attached document.
-
AC2: 'Reply on RC1', Aatralarasi Saravanan, 02 Dec 2024
-
RC2: 'Comment on egusphere-2024-2369', Anonymous Referee #2, 14 Oct 2024
The authors present a validation exercise for different precipitation products in several regions of India.
The topic is suitable for HESS, but I have some methodological concerns that need to be addressed before publications
General comments:
Station to grid comparison: The authors assert that station to grid comparisons are difficult and conduct 'linear interpolation' from station to grid and from product grid to a common 0.1x0.1 degree comparison. I find the overall description of this vague and am also not entirely convinced that this solves the problem given that a lot of the variability within each grid-cell is due to topography and localized patterns that don't change linearly. I am also concerned/ confused that the authors ony consider stations within the district for the regional/ district comparison. Given the irregular shape additional stations outside the region should also be considered.
Choice of regions: I might have missed that, but why are only some regions compared and not India as a whole?
Reliance on Tables and many figures to compare: Apart from the Taylor diagrams, the authors have many tables and many figures with subplots that compare values between the products. It is very difficult to keep track of all of these comparisons. It would be good to think about a better way to integrate and present results.
Specific comments:
Section 2.2. This section could probably be shortenened to focus on the most important information here.
L153: Climate Data Guide, 2024 is not an appropriate citation for the datasets since the CDG is not the primary source of the data, but a guide for data users by UCAR.Figure 1: The regular gridded station distribution seems to be an error in the figure?
L240: " For quality reasons, the years 2005 and 2010 were excluded from the present study" > Please explain
L253: "The interpolated 0.1degree station dataset was used as ground truth to evaluate all the other precipitation products" > See my general comment. Also, it would be really good if the authors could come up with a way to provide any kind of quality measure for this. For example, the authors could have reserved some stations for verification of that methodolody. Or conduct a leave-one-out cross-validation to assess how well the interpolated data reflects actual precipation in that location.
L320: It would be good to also provide MAE as a percentage value of mean precipitation.
Figure 2: I am a bit confused with this figure because alls of these lines seem to be perfectly straigt on a log-log plot and that is something that I would not have expected. It is also not possible to always see all lines.
Figures 3-6: I was initailly confused by these figures. I guess the key message here would be, how the errors compare between monosson and non-monsoon season, but for that, the reader has to do their own math.
Figure 8-11: These should contain the station locations to get a better sense of the interpolation.
L490: "ERA5-Land produced the closest approximation to the station data (Fig. 8). " > It would be good to back this up with some quantittative quality measure rather than a qualitative comparison.
Section 5: This should start with a discussion of results including general patterns and limitations of the study. Then followed by
Citation: https://doi.org/10.5194/egusphere-2024-2369-RC2 -
AC3: 'Reply on RC2', Aatralarasi Saravanan, 02 Dec 2024
We express our gratitude to the referees and the handling editor for their valuable time in helping improve this manuscript. In our opinion, the referees’ input will considerably improve the revised manuscript. Here will outline all changes we will add to the initial manuscript. This is our initial response. Planned activities are also included in the document and will all be implemented.
- Restructuring subsections “Study region and Ground station”
- The ‘Datasets’ section will be revised only highlighting important information
- Inclusion of methodology – Grid and District scale evaluation, Monsoon and Non-monsoon month evaluation
- Inclusion of all the equation numbers and abbreviations in the methodology
- Consistent use of Acronyms, e.g. ‘GPM-IMERG’
- Tables are revised to include only Correlation Coefficient (CC), Root Mean Square Error (RMSE) and RB (Relative Bias)
- Revised Figure. 2 to include only IDF for 5-year Return Period
- Restructuring the ‘Discussion’ section by describing the results with related studies followed by the limitations of the present study.
- Conclusion will be revised and include an outlook of the present study’s impact in other data-scarce regions in India.
We added a response to the reviewer’s individual comments in the attached document.
-
AC3: 'Reply on RC2', Aatralarasi Saravanan, 02 Dec 2024
-
RC3: 'Comment on egusphere-2024-2369', Anonymous Referee #3, 21 Oct 2024
The manuscript is highly relevant, as data scarcity is a major constraint in both hydrological and agricultural modeling. Providing reliable information about alternative data sources can aid modelers and support decision-makers. However, some sections of the manuscript need restructuring. Detailed comments are provided in the attached PDF. The manuscript can be accepted after minor revision.
-
AC4: 'Reply on RC3', Aatralarasi Saravanan, 03 Dec 2024
We express our gratitude to the referees and the handling editor for their valuable time in helping improve this manuscript. In our opinion, the referees’ input will considerably improve the revised manuscript. Here will outline all changes we will add to the initial manuscript. This is our initial response. Planned activities are also included in the document and will all be implemented.
- Restructuring subsections “Study region and Ground station”
- The ‘Datasets’ section will be revised only highlighting important information
- Inclusion of methodology – Grid and District scale evaluation, Monsoon and Non-monsoon month evaluation
- Inclusion of all the equation numbers and abbreviations in the methodology
- Consistent use of Acronyms, e.g. ‘GPM-IMERG’
- Tables are revised to include only Correlation Coefficient (CC), Root Mean Square Error (RMSE) and RB (Relative Bias)
- Revised Figure. 2 to include only IDF for 5-year Return Period
- Restructuring the ‘Discussion’ section by describing the results with related studies followed by the limitations of the present study.
- Conclusion will be revised and include an outlook of the present study’s impact in other data-scarce regions in India.
We added a response to the reviewer’s individual comments in the attached document.
-
AC4: 'Reply on RC3', Aatralarasi Saravanan, 03 Dec 2024
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 249 | 78 | 106 | 433 | 23 | 2 | 2 |
- HTML: 249
- PDF: 78
- XML: 106
- Total: 433
- Supplement: 23
- BibTeX: 2
- EndNote: 2
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1