the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Jan 2025
Dynamic assessment of rainfall erosivity in Europe: evaluation of EURADCLIM ground-radar data
Abstract. Heavy rainfall is the main driver of water-induced soil erosion, necessitating accurate spatial and temporal predictions of rainfall erosivity to predict the soil erosion response. This study evaluates the ground radar-based EUropean RADar CLIMatology (EURADCLIM) precipitation grids to quantify rainfall erosivity across European countries. Compared to Global Rainfall Erosivity Database (GloREDa) gauge-based interpolations, EURADCLIM overpredicts rainfall erosivity, principally due to residual artefacts in some regions which inflate the instantaneous rainfall rates. Overprediction is most pronounced in European regions with lower radar antenna coverage and complex topography, whereas flatter regions with lower erosivity and better radar coverage are better predicted. Disagreement attributes to the input radar quality in EURADCLIM (derived from OPERA) and to a lesser extent the uncertainty in GloREDa due to its limited gauge records in some regions. Event (EI30) time series analysis showed reasonably good performance (KGE > 0.4) in 50 % of the evaluated gauge locations, although significant overprediction by EURADCLIM was evident in the upper quantiles in some countries. Accounting for the propagation of these remaining time-slice artefacts, which have a large impact on the temporally-aggregated R-factor, applying a 80 mm/h threshold to limit the maximum I30 value (i.e., less than 0.1 % of GloREDa events exceed this threshold) during the calculation of rainfall erosivity significantly improves the performance of the EURADCLIM dataset at annual, monthly and event time scale. Following adjustment, EURADCLIM best agrees with GloREDa across Europe in July and August, while bigger differences were observed in June and winter in general. Annually, the spatially aggregated rainfall erosivity per country had a percent bias below 10 %. While applying simple I30 thresholds is promising, radar artefacts remain significant in areas with lower quality rainfall retrievals. Notably, regions in Europe with lower quality or absent data furthermore coincide with established high soil erosion rates. In the absence of spatiotemporally continuous, high-quality ground-radar retrievals across Europe, we show the value of ensemble R-factor layers of EURADCLIM with three other rainfall erosivity grids (e.g., satellite retrievals) and discuss the possibility of ground radar to offer unique spatial detail in such ensembles.
- Preprint
(1710 KB) - Metadata XML
-
Supplement
(2252 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on hess-2024-402', Shuiqing Yin, 05 Feb 2025
Radar-based datasets offer the advantage of high temporal and spatial resolution, making them crucial for capturing the high variability of rainfall erosivity, particularly at the event scale. This study aims to evaluate the ground radar-based EURADCLIM precipitation dataset, which has an hourly temporal resolution and a 2 km spatial resolution, to quantify rainfall erosivity at annual, monthly, and event scales across European countries—a significant endeavor.
However, several concerns need to be addressed:
EURADCLIM is a ground radar-based dataset that has been bias-corrected using station observations. However, there is no clear information on whether the stations used in the development of EURADCLIM are the same as those in GloREDa. If they are, a separate discussion is needed, as data accuracy is generally higher in areas with stations and lower in regions without them.
GloREDa’s data is interpolated from station observations. Due to the limited density of stations, it provides a reasonable representation of long-term average rainfall erosivity. However, for event-scale and daily-scale rainfall erosivity, GloREDa is likely to have significant errors in areas without stations due to the high spatial variability of precipitation, particularly from heavy convective storms, at these finer scales. Therefore, using GloREDa as the ground truth to assess the accuracy of EURADCLIM should be approached with great caution.
Additionally, some studies have used radar-based datasets to quantify rainfall erosivity, but they do not appear to be included in the review section. For example, the daily erosion model hosted by Iowa State University could be relevant to this study.
Other suggested improvements include:
• Adding the definition of the Kling-Gupta Efficiency (KGE).
• Including a sample event illustrating the dynamic evolution of rainfall erosivity.Citation: https://doi.org/10.5194/hess-2024-402-CC1 -
RC1: 'Comment on hess-2024-402', Anonymous Referee #1, 06 Feb 2025
I reviewed the manuscript "Dynamic assessment of rainfall erosivity in Europe: evaluation of EURADCLIM ground-radar data", and read it with interest, revealing a new potential use case for European gauge-adjusted radar data. The manuscript is well written and concise and considers the use of a new European gauge-adjusted radar precipitation dataset for erosivity modelling. I do have a couple of questions and comments that will hopefully help to authors to improve their manuscript, especially regarding the results and discussion.
Introduction:
- L. 40-43: Not being an expert in erosivity: the rainfall erosivity index is, through high spatial and temporal variability a product of the characteristics of rainstorm kinetic energy. Does the rainfall erosivity index not depend on soil type, land use and orography?
- The introduction clearly describes the aim of this manuscript. As far as I am aware, this is the first time that a pan-European radar dataset is used for erosivity assessment. You mention one other study that uses ground-based weather radar data for erosivity assessement. Are there other studies that use radar data for this purpose? If yes, add references. If no, the uniqueness of this manuscript in using radar data may be emphasized more in the introduction.
- L. 45/46: This needs rephrasing: "rain gauges are fundamentally represent point scale measurements with a limited cover limits"
- L. 70: What is meant by "sub-timestep rainstorm intensity"? Being smaller than the time step of the available precipitation dataset?Data and methods:
- L. 90-91: Are the "GloREDa 1.2 average monthly and annual predictions" based on observations, or are these true predictions, i.e., forecasts? My interpretation is that these are observations and that these observations are employed to compute variables such as rainfall erosivity. Perhaps another term than "precitions" is more appropriate then.
- EURADCLIM dataset: Pitfall 1) is actually caused by factors of which "radar beam attenuation, changes in the reflectivity profiles with distance from the antenna" are important ones. These should be mentioned here and not under 3), because these do not generate noise.
- EURADCLIM dataset: In my view, pitfall 3) is already covered by pitfall 2), because non-meteorological echoes and other artifacts can be considered noise. Such artifacts may also be due to "hardware related issues, such as calibration errors". So, I suggest to remove pitfall 3), where a phrase such as "hardware related issues, such as calibration errors" could be added to pitfall 2).
- L. 114: One has to realize that only a limited number of these rain gauges are actually used in the merged product. Partly because not all gauges are available over the entire period, but especially because of the applied 0.25 mm thresholding on 1-h gauge accumulations.
- L. 133-144: It seems that the 1-h radar accumulations are disaggregated to 30-min accumulations, where different disaggregation schemes were tested. This should be clarified in the text.
- Not begin familiar with rainfall erosivity computations: would it be possible to add an appendix with the most important equations and description of the computations?
- Would it be possible to derive EI60 instead of EI30 based on the GloREDa dataset and EURADCLIM dataset? This would avoid the disaggregation of radar data to 30 min, and hence avoid introducing uncertainty on the intensity over 30-min intervals. I guess that the 30-min interval is relevant because higher rainfall intensities causing stronger erosion are better captured, whereas these may be averaged out too much at the 60-min interval.Results and discussion:
- L. 158-159: Note that at least Italy has some coverage, so perhaps write "limited to countries with (almost) full coverage".
- L. 164: "reduced" is used twice in this "reduced percent bias reduced". And replace "THis" by "This".
- L. 163-165: Why would the different radar antennas and the lower rain gauge network density lead to overestimation of erosivity? This could be discussed in more detail, for instance, by referring to results shown for EURADCLIM version 1 (Overeem et al., 2023). There, Fig. 7h show large underestimations and overestimations for regions with low gauge network density (leave-one-out-statistics), which could be a proxy for underestimations in specific areas far away from gauges (note that this underestimation largely disappears for the dependent verification in Fig. 7i). Of course, this concerns all 1-h accumulations and not only the extreme ones that are most relevant for erosion.
- "artifact" and "artefact" are used interchangeable, choose one of them.
- Figure 4: Given the fact that underestimations are found in EURADCLIM version 1 (Overeem et al., 2023; and this will also be the case in version 2) for some regions, such as parts of Norway, Sweden and Austria, it is apparent that few areas with too low erosivity are found when compared to erosivity computed with GloREDa. Of course, these are based on all 1-h accumulations, and the focus in Overeem et al. (2023) is less on 1-h extremes. In addition, extreme rainfall tends to be stronger at the point scale compared to the radar pixel scale (think of areal reduction factors), implying that radar data could underestimate erosivity for the highest rainfall intensities for short duration rainfall, such as 30 min (this radar precipitation estimation could be in the order of 10%). Can you comment on this?
- Figure 5: Suggest to make this a square plot, because axes have the same scale. This would increase readability. In addition, this could be made more quantitative by incorporating a metric such as the Pearson correlation coefficient.
- Although it is to be expected and confirmed that remaining non-meteorological echoes in EURADCLIM can give rise to erosivity overestimation in several regions, being consistent with Figure 9g,h in Overeem et al. (2023) for EURADCLIM version 1, the red areas with overestimation in Figure 4 are quite abundant. Could this also be related to underestimation by the gridded GloREDa dataset? Rain gauges only sample a limited number of locations. Hence, much more extreme events may occur between rain gauge locations, that may be captured by radars though.
- Caption Figure 6: Since four and five datasets are mentioned, respectively, it seems a 4 and 5 "data source ensemble".
- Figure 6: Why the mean annual erosivity becomes generally lower in case the EURADCLIM dataset is added? Given the overestimation found for EURADCLIM, I would expect higher values? Or does this imply that the EURADCLIM dataset has relatively modest erosivity compared to the satellite-based datasets?Conclusions:
- Why would overprediction occur in European regions with lower radar antenna coverage? A longer distance to a radar would make precipitation estimates less reliable and can especially make it harder to correct for range dependent sources of error, such as the vertical profile of reflectivity and rain-induce attenuation along the radar beam.Figure S3:
- Note that although Austria is covered by weather radars, Austrian radars did not contribute yet to the OPERA data used in EURADCLIM.
Reference list:Overeem, A., van den Besselaar, E., van der Schrier, G., Meirink, J. F., van der Plas, E., and Leijnse, H.: EURADCLIM: the European climatological high-resolution gauge-adjusted radar precipitation dataset, Earth Syst. Sci. Data, 15, 1441–1464, https://doi.org/10.5194/essd-15-1441-2023, 2023.
Citation: https://doi.org/10.5194/hess-2024-402-RC1 -
RC2: 'Comment on hess-2024-402', Anonymous Referee #2, 25 Jun 2025
I was very pleased to read the manuscript "Dynamic assessment of rainfall erosivity in Europe: evaluation of EURADCLIM ground-radar data." The author aimed to evaluate the ground radar-based European RADar CLIMatology (EURADCLIM) precipitation grids to quantify rainfall erosivity across European countries. The manuscript was clear, well-structured, and easy to follow. The multi-scale comparison of erosivity values between EURADCLIM and GloREDa demonstrates the potential application of ground-radar data in soil erosion studies at large time and spatial scales.
I have some comments that could improve the interpretability of the methodology and results. The comments are principally focused on providing more interpretable and understandable results to the readers.
Abstract
Line 18: Please define KGE because it is the first time that it has been named.
Lines 26-27. This statement is not explicitly discussed in the manuscript.
Introduction
Line 45: Please rephrase the sentence: "… with a limited cover limits predictions…"
Line 78, 111, and 348. Please be consistent with the terms. In the document, it is read as "artefacts"
Methodology
Line 105: Please check the spelling of "kilometres"
Line 135 and 136: "Secondly, the measured 30-min rainfall data was aggregated to the hourly time step matching the EURADCLIM resolution, and four rule-based rainfall disaggregation schemes were tested:" This sentence is contradictory; it is not clear if GloREDa time series were aggregated to hourly time steps or EURADCLIM time series were disaggregated. I think it's the second option, but the sentence is unclear.
Line 99-100: The annual average erosivity map was produced in Panagos et al. (2017) 10.1038/s41598-017-04282-8. In the public repository of GloREDa (Panagos et al., 2013), maps are available only for monthly erosivity. Please clarify which is the source of the annual average erosivity.
It would be helpful to include a figure in the supplementary material showing the distribution of GloREDa gauge stations around the European countries included in this study. It will familiarize non-European readers with the dataset used in this study.
Results and discussion
Figure S2: The authors have related this figure to spatial analysis, for example, in line 16. It will be helpful if at least for the worst predictions (Croatia, Bosnia 160, Herzegovina, Serbia, and
Estonia), the points in the figure would have the name of the country. Additionally, the reference to Figure S2 in line 163 suggests spatial variability that is not depicted in Figure S2.
Line 164. Please rewrite the sentence "reduced percent bias reduced."
Line 164-168. It would be helpful for readers to add a map showing the spatial distribution of the OPERA antenna across Europe. Additionally, the map could include a DEM in the background to show the complexity of the topographic conditions in certain regions; it will be helpful to support the statement of lines 166-168.
Figure 1. It would be helpful for non-European readers to include country names in the figure. Maybe adding all names may result in a crowded map; however, it is necessary to include at least those country names referenced in the text (e.g., Finland, Slovenia, Poland, among others).
Figure S4: The figure's caption is wrong. I guess that GloREDa prediction is on the upper panel, and EURADCLIM is on the lower panel.
Line 213-216. The sentence refers to the spatial distribution of seasonal erosivity; however, Figure S6 does not display spatial characteristics.
Lines 245-247. The sentence is confusing. It is clear that 50% of the points with more than 10 comparable events obtained KGE greater than 0.4. However, the second part of the sentence looks like a contradiction of the first statement. Additionally, it would be beneficial to include the definition and equation for calculating the Kling-Gupta Efficiency in the methodology section.
Figure S8. Points and raster should have the same color scale. It will be helpful to compare.
Figure S12. There are no units in the graph or the caption.
Lines 352 -369. Why did the authors not include the comparison of the mean values of the R-factor (ensemble platform) with the GloREDa, which is used as ground truth in this study? Beyond showing a more detailed resolution of the R-factor, it is important to display the agreement with the reference database. It will be valuable to include this comparison.
Citation: https://doi.org/10.5194/hess-2024-402-RC2
Data sets
EURADCLIM: The European climatological gauge-adjusted radar precipitation dataset (1-h accumulations) KNMI Koninklijk Nederlands Meteorologisch Instituut https://doi.org/10.21944/ymrk-mr24
Global Rainfall Erosivity Database (GloREDa) European Soil Data Centre (ESDAC) https://esdac.jrc.ec.europa.eu/content/gloreda
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 238 | 48 | 19 | 305 | 32 | 13 | 19 |
- HTML: 238
- PDF: 48
- XML: 19
- Total: 305
- Supplement: 32
- BibTeX: 13
- EndNote: 19
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1