the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 14 Oct 2024
Very high spatial and temporal resolution rainfall data for accurate flood inundation modelling
Abstract. High-quality rainfall data are crucial for various climatological and hydrological applications, especially in detailed modelling. However, obtaining precipitation data with fine spatiotemporal resolution is often challenging due to the limited availability of sub-daily point measurements and the sparse distribution of rainfall stations in many regions. This paper presents and demonstrates a method to generate the Commonwealth Scientific and Industrial Research Organization Hourly Rainfall (CHRain) dataset, which provides hourly and 1 km gridded rainfall surfaces for hydrological/hydrodynamic modelling. The method applies thin-plate spline interpolation to generate rainfall surfaces using hourly input time series obtained from hourly rainfall stations, and from daily data disaggregated into hourly intervals based on patterns observed in nearby hourly rainfall stations, and also guided by continuous radar images. The method is used to represent rainfall patterns and amounts from 2007 to 2022 in the Richmond River catchment in New South Wales, Australia. The CHRain dataset is compared with hourly measurements and other gridded datasets currently available in Australia. The correlation coefficient of 0.948 shows that the CHRain dataset can adequately reproduce the patterns of hourly rainfall measurements. The spatial and temporal analyses also indicate that the CHRain dataset outperforms other gridded datasets in representing the sub-grid distribution as well as the daily and hourly variation of rainfall across the study area. These are all essential for capturing the spatiotemporal characteristics of flood inundation in the study area which is frequented by disastrous flood events. The proposed method opens an opportunity to develop high resolution spatiotemporal rainfall datasets for other regions.
- Preprint
(13860 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 11 Nov 2024
The need for better spatial and temporal resolutions of hydrological data has been an ongoing demand by researchers. To meet the demand, for example, with the advancement of technology and resources, the joint collaboration of NASA and USGS has decided to improve the spatial and temporal resolutions of its Landsat products (e.g., Landsat 8, 9) by 2030 with the impending introduction of Landsat NEXT, which is expected to have a spatial resolution of 10 m for some of the bands. In this manuscript, the authors propose a methodology to generate gridded rainfall data for a river basin in Australia. The spatial and temporal resolutions of the generated data are 1 h and 1 km, respectively. The proposed methodology is based on a thin-plate spline interpolation model that was introduced in 1990. The authors claim that the generated gridded rainfall data has outperformed the other sources of rainfall data available in Australia.
The current version of the manuscript has many flaws. I suggest the authors go through the video on how to write a research paper (https://www.egu.eu/webinars/229/how-to-write-a-research-paper/). Although the presentation is not the best on the topic, I believe that it has something to grasp. Thanks to the UK (University of Leeds), even after Brexit, it is playing the role of gladiator to safeguard the education system of the EU.
1) The title of the manuscript needs to be examined by the journal office. I presume that the journal office has appointed a handling editor and referees after going through their academic transcripts and their PhD works.
There is a need to understand the way titles have been coined in manuscripts in this journal office. For example, recently, in this journal office, the handling editor published a manuscript authored by Geogle Research. The title of the manuscript is Never Train a Long Short-Term Memory (LSTM) Network on a Single Basin. Is there a verb in the title? The title of this manuscript is "VERY HIGH spatial and temporal resolution rainfall data for accurate flood inundation modeling." Is there a verb in the title?
To evaluate the quality of the review process of the journal office, it would be more appropriate to conduct research to examine the titles of the manuscripts that have been published by the journal office since the establishment of the journal.
2) See the attached file (Figure 2_3.pdf). Are the figures correct? Have you pasted incorrectly? I am expecting the investors of the journal office to spend a few euros to develop an AI/ML-generated model to do a preliminary check on the manuscripts that are submitted to this journal office.
3) Refer to PART II
- AC3: 'Reply on CC1', Chi Nguyen, 14 Apr 2025
-
CC2: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 11 Nov 2024
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2024-228/hess-2024-228-CC2-supplement.pdf
-
CC3: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 12 Nov 2024
1-3) Refer to PART I
4) Australia is a developed country. If I have misunderstood, do correct me. Therefore, I am expecting the authors to produce their own datasets to map Figure 1. There are too many sources/references cited to produce the map shown in Figure 1. This gives a bad impression about the state of water resources management in Australia. Since the spatial locations of raingages are available from the Bureau of Meteorology, it should be a matter of a few minutes to have the map of gages. Moreover, Australia has a fine resolution DEM. Therefore, producing the catchments and stream networks for the nation is not a pain-taking task. Are the authors lacking knowledge and experience?
5) In the current version of the manuscript, the thresholds that are used to remove unrealistically high hourly and daily rainfall data are not justified. What was the reason to set the daily threshold to 1500 mm/d? What was the reason to set the hourly threshold to 300 mm/h? Considering the authors’ statements [LN 171-172, LN 106-107, LN 104] in the current version of the manuscript, these threshold values are not justified.
6) LN 170-Some unusually high values of hourly rainfall, mostly occurring at midnight, were detected. LN 168- The suspicious data were removed, including negative, ...
The authors have found some interesting points in the Australian datasets. It would be more appropriate to include a few lines to elaborate more on this for the readers to learn more on what has been found by the authors. What was the reason to have negative values in the dataset? What was the reason to have very high values of hourly rainfall, specifically at midnights? Are these observational errors, or are these due to some interesting unexplored researchable areas?
7) Refer to PART III
-
CC4: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 13 Nov 2024
1-6) Refer to PART I & II
7) As per the authors, the spatial extent of the analysis is 30,389 km2[LN 111]. Moreover, there are 139 hourly stations for generating hourly rainfall surfaces using the proposed method. This includes the disaggregated stations (i.e., daily to hourly) as well. Considering these details, on average, a gauge accounts for 218 km2(= 30,389/139). Based on this information, you have fitted your 1 km gridded surface. Is this what you want to achieve in this manuscript? If I have misunderstood, do correct me.
8) Table 1 is a good piece of work in this manuscript. It would be more useful for the readers of this manuscript if the authors added a few more sections to outline more about the datasets (e.g., ANUClimate and AGCD) that the authors have listed in Table 1. Moreover, I suggest the authors work with the handling editor to find a better title for Table 1.
9) The spatial and temporal resolutions of ANUClimate are 1 km and 24 h, respectively. What if we disaggregate the daily grid of ANUClimate to an hourly grid using the existing hourly rain gauges? Wouldn't it yield the grid surface of your interest (i.e., 1 km in spatial and 1 h in temporal)? How would this differ from your methodology?
10) Refer to PART IV
-
CC5: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 13 Nov 2024
1-9) Refer to PART I, II, and III
10) As per the authors, the radar rainfall shows the rainfall in the atmosphere instead of the rainfall reaching the ground. Moreover, as per the authors, the Bureau of Meteorology has found that there are errors in the radar-derived rainfall data, showing unreasonable high rainfall values in some areas. Due to this reason, the authors acknowledge that the hourly rainfall accumulating from radar data was not used in the analysis [LN 146-151].
With the advancement of technology and theoretical development, researchers have been employing ML and AI algorithms to learn from data. Considering this trend, wouldn't ML/AI algorithms improve the development of gridded rainfall surfaces of your interest (i.e., 1 km in spatial and 1 h in temporal)? With the availability of daily rainfall gauges, hourly rainfall gauges, and hourly radar rainfall data, wouldn't it be feasible to develop an ML algorithm to relate the radar data and the existing hourly and daily gauged data to predict the correct values of hourly radar values (i.e., 1 km in spatial and 1 h in temporal)? How would this differ from your methodology?
11) Refer to PART V
-
CC6: 'Comment on hess-2024-228', Sivarajah Mylevaganam, 14 Nov 2024
1-10) Refer to PART I-IV
11) A few instances of ambiguous citing have been found in the current version of the manuscript. For example,
Statement #1-LN 18-21
High resolution temporal and spatial representations of precipitation data are required in many hydrological applications, such as modelling flood inundation (Jhong et al., 2017; Pappenberger et al., 2005), analysing catchment responses in rainfall-runoff models (Xu et al., 2022; Pappenberger et al., 2005; Acharya et al., 2019), and forecasting extreme events and natural hazards (Ficchi et al., 2016; Mukherjee et al., 2018).
Statement #2-LN 365-366
Hourly rainfall data are essential for many hydrological, ecological, and meteorological applications (Lewis et al., 2018; Hatono et al., 2022).
Why would Lewis et al. (2018) and Hatono et al. (2022) not fit into statement #1 in the current version of the manuscript? I suggest the authors work with the handling editor to resolve these issues.
12) As per the authors, the independent variable transformation for the DEM is x/a, where “x” is the DEM value, and “a” is the transformation parameter. In this study, “a” was set as 10,000 to reduce the impact of the DEM on the hourly rainfall surfaces. The usual value recommended for interpolating monthly and daily data is 1000 (LN 202-203).
The latitude and longitude values are in decimal degrees. The DEM is in liner unit (e.g., m). The rainfall data is in mm. Therefore, there exists a non-normality issue with the data that have been used to fit the surface. I am not sure if the authors are aware of this. Does this lead to setting a random number to the value of “a” (e.g., 9999 instead of 10000)? The current version of the manuscript doesn’t justify the magnitude of “a”. Why would it jump to 10,000 from 1000 just because the work is hourly instead of daily or monthly?
13) Refer to PART VI
-
RC1: 'Comment on hess-2024-228', Anonymous Referee #1, 05 Feb 2025
The manuscript presents an hourly rainfall product at 1 km resolution, derived by disaggregating daily station data (where needed) to an hourly scale and interpolating using a thin-spline method. The authors compare their dataset to existing rainfall products and demonstrate improvements. The manuscript is generally well-written and structured. However, I find the novelty of the study lacking. The thin-plate spline method has been previously used for spatial interpolation of climatic variables (by Hutchinson, one of the authors), and the authors do not propose any methodological advancements (unless I overlooked them) from what has been published in the past. Similarly, the disaggregation approach appears overly simplistic compared to more sophisticated state-of-the-art methods. Given the limited methodological innovation, I question the suitability of this paper for HESS. A more appropriate venue might be a journal focusing on dataset development.
Below, I provide more specific comments:
1. I suggest revising the title. A 1 km resolution is not necessarily "very high," and the precipitation dataset (why focus only on rainfall?) has applications beyond flood inundation modeling.
2. Lines 9-10: The CHRain dataset is compared to other gridded datasets available in Australia, but does it outperform them? This should be clarified.
3. Line 10: A correlation of 0.948 is quite high, but was it calculated using all the data involved in interpolation and merging? A "leave-one-out" validation approach would provide a more reliable assessment.
4. Line 20: The same citation is used for two different statements on lines 19 and 20. Please verify.
5. Lines 21-23: This section discusses temporal variability, but what about spatial variability? Given its impact on flood modeling, I recommend considering the following references: https://doi.org/10.5194/hess-17-2195-2013 and https://doi.org/10.5194/hess-21-1559-2017
6. Lines 26-27: This sentence seems disconnected from the surrounding discussion.
7. Lines 29-30: The phrase "1 km to 12 km" is unclear. Since stations provide point-scale data and radar typically operates at 1 km resolution (but may not cover all of Australia), this should be clarified.
8. Lines 72-75: GCMs are not designed to simulate observed rainfall. The current wording is misleading, and I suggest removing references to GCMs in this context.
9. Line 99: "Changes significantly" should be quantified. What is the elevation difference?
10. Sections 2.1 and 2.5. Do you simply apply the thin-plate spline as suggested by Hutchinson? If so, what is the novelty of your approach?
11. Section 2.4: If the closest station is 10 km away (just giving an example), the correlation may be too low for reliable disaggregation... A sensitivity analysis using stations at varying distances could provide insights into the method’s limitations.
12. Line 203: Why is alpha not treated as a calibration parameter?
13. Lines 231-232: The manuscript reports too many goodness-of-fit measures. Why include both NSE and KGE, for instance? I suggest focusing on two distinct indices that provide complementary information.
Citation: https://doi.org/10.5194/hess-2024-228-RC1 - AC2: 'Reply on RC1', Chi Nguyen, 14 Apr 2025
-
RC2: 'Comment on hess-2024-228', Anonymous Referee #2, 20 Feb 2025
This is a well-structured and straightforward paper. I have no doubt that the authors’ results are likely to be very useful to flood-risk and flood-disaster managers. If I understand correctly, the authors derive a new 15-year high-res spatiotemporal precip. dataset from existing rain gauge and reanalysis data in an Australian location which is prone to short timescale flooding and hence where good hydrological precipitation/flood modelling is highly desirable. The authors compare the resulting product with existing alternatives and find that it is superior when their specific metrics are used. I do not disagree in principle with the authors conclusions, nor do I find fault with the methodology used to produce the CHRain dataset or used to compare the CHRain dataset with BARRA-SY, ANUClimate, and AGCD.
My sole reservation is with this study’s contribution to the current state-of-the-art. The datasets used are all well-established. The interpolation is done via an off-the-shelf software tool. The temporal downscaling is done “using the hourly distribution pattern from the nearest hourly station” and applying some reasonable quality control which is not an innovation.
The authors are correct in that: “The proposed method opens an opportunity to develop high resolution spatiotemporal rainfall datasets for other regions” which are essential for developing “detailed flood modelling”. However, I find this paper to be more a successful, and very useful, application of an established methodology that a progression beyond the state of the art.
Minor comments and typos:
Line 31) “Observation” should be “observations”
33) “… more than 20 years” should be “… more than 20 years long”
65) “showed to improve” should be “appeared to improve”
84) “An accurate high resolution spatial and temporal resolution rainfall” should be “An accurate high spatial and temporal resolution rainfall”
104) spurious comma after “especially”.
132) in “an” area…
176) “Disaggregate daily rainfall data to hourly” should be “Disaggregation of daily rainfall to hourly” or something similar…
189) “After cleaning, disaggregating, and detailed quality control of the data” should be “After cleaning, disaggregating, and completing a detailed quality control of the data” I think…
Figure 7) The first and last row could be removed without loss of clarity…
Citation: https://doi.org/10.5194/hess-2024-228-RC2 - AC1: 'Reply on RC2', Chi Nguyen, 14 Apr 2025
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 341 | 88 | 113 | 542 | 31 | 19 |
- HTML: 341
- PDF: 88
- XML: 113
- Total: 542
- BibTeX: 31
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1