the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 19 Feb 2024
HERA: a high-resolution pan-European hydrological reanalysis (1950–2020)
Abstract. Since 1950, European rivers have been put under increasing pressure by anthropogenic activities, resulting in changes in climate, land cover, soil properties and channel morphologies. These evolving environmental conditions can translate into changes in hydrological conditions. The availability of consistent estimates of river flow at global and continental level is a necessity to assess and attribute changes in the hydrological cycle. To overcome limitations posed by observations (incomplete records, inhomogeneous spatial coverage), we simulate river discharge for Europe for the period 1950–2020 using a state-of-the-art hydrological modelling approach. We use the new European set up of the LISFLOOD model, running at 1 arcminute (≈1.8 km) with six-hourly time steps. The hydrological model is forced by climate reanalysis data (ERA5-land) bias-corrected and downscaled to the model resolution with weather observations. The model also ingests 72 surface fields maps representing catchment morphology, vegetation, soil properties, land use, water demand, lakes and reservoirs. Inputs related to human activities are evolving through time to emulate changes in society. The resulting Hydrological European ReAnalysis (HERA), provides six-hourly river discharge for 282 521 river pixels with upstream area > 100 km2. We assess its skill using 2901 river gauging stations distributed across Europe. Overall, HERA delivers satisfying results, with a general weak underestimation of observed mean discharge and flow variability. We find that the performance of HERA increases through time between 1950 and 2020. The fine spatial and temporal resolution result in an enhanced performance compared to other reanalysis for small-to-medium-scale catchments (100–10 000 km2), with degraded performance remaining for small catchments. HERA is the first long-term, high-resolution hydrological reanalysis for Europe. Despite its limitations, it enables the analysis of hydrological dynamics related to extremes, human influences, and climate change at a continental scale while keeping local relevance. It also creates the opportunity to study these dynamics in ungauged catchments across Europe.
- Preprint
(3138 KB) - Metadata XML
-
Supplement
(720 KB) - BibTeX
- EndNote
Status: open (until 23 May 2024)
-
RC1: 'Comment on essd-2024-41', Anonymous Referee #1, 23 Apr 2024
reply
Short summary:
The authors present a high-resolution model-based runoff reanalysis for the period 1950-2020 which is based on bias-adjusted and downscaled ERA5-land meteorological inputs. In the paper the general framework and incorporated datasets have been described, as well as a model evaluation of mean and extreme statistics performed. Within this framework anthropogenic influences (e.g., water use, land use change) have been considered for the period 1950-2020 and many reservoirs have been included in the hydrological reanalysis.
General statements:
While the development of high-quality hydrological reanalysis products is relevant and I acknowledge the large effort in creating this dataset, I have some general comments on the paper as well as some detailed comments below which I believe will help future users of the dataset. Firstly, I have noticed that several in text citations are not included in the reference list of the manuscript. Please check all your references for correct referencing in the text and the reference list. I have not checked all references but noticed numerous inconsistencies, some examples highlighted in the comments below. Further, the manuscript needs another careful reading to eliminate many typos and grammatical errors. I have initially highlighted a few errors, but stopped because there are too many. Also, the figure referencing throughout the manuscript can be improved. In some parts (highlighted below) your methodological descriptions could benefit from a few more details.
From a methodological perspective I have a few comments related to your model evaluation. I didn`t fully understand your grid cell matching approach for the comparison of LISTFLOOD with gauging stations. Maybe you could elaborate on this more and see my detailed comment on this below. Further, for your evaluation it is not clear on which time period your comparison between observations and LISTFLOOD runoff is based on. It sounds like each catchment is based on any period from 1 to 71 years of data. This needs to be clarified to better interpret your entire model evaluation. With a variable evaluation length your runoff quantiles will not be comparable across catchments. Lastly, your final data product has a temporal resolution of 6-hours, however, your entire analysis is based on daily data. Hence, we can’t assess whether the performance of the reanalysis remains the same between daily and 6-hourly resolution. Some of your evaluation metrics might change. Therefore, at some point in your manuscript or in the Supplementary material you should show some evaluation of the 6-hourly data in comparison to the daily scale to showcase that your analysis is valid also for the higher resolution final product.
Detailed comments:
P3-L86: “environmental” -> “environment”
P4-L107: “fist” -> “first”
P4-L116: If “Here” stands at the beginning of the sentence a comma “,” is needed after “Here,”.
P4-L118: Dottori et al. 2022 is not in the reference list.
P5-L145: “1`” Are you using a regular grid?
P5-L147: Maybe clarify here that your yearly chunking is based on the calendar year. The reader can assume this from line 149 but it would be good to clarify this here already.
P5-L148: “71-year pre-run”. What is your 71-year pre-run? Please briefly explain.
P6-L149: Please briefly explain how you initialize the start of the new calendar year. I assume by the storage components of the previous simulation year. Also, how do you initialize when you for example introduce new reservoirs.
P6-L159: “data discharge” -> “discharge data”
P6-7: Model Calibration section. Can you elaborate a bit more on the calibration of the model. For example, on (a) whether the model was calibrated towards floods or more the general water balance, (b) whether the model was calibrated on daily or sub-daily timescales. It is difficult to find the relevant information in the online documentation that you cite.
P7-L179: What does "modified version" mean in this context? Does it mean, bias-adjusted and downscaled or is it related to something else?
P7-L188-89: Can you please clarify how you aggregated the hourly to 6-hourly/daily data. Aggregated is ambiguous and can mean taking the average, sum or instantaneous value.
P8-L198-99: Can you provide a reference for your statement of "robust bias adjustment of extremes". In the Lange 2019 citation that I believe you are referring to, I can`t find sufficient evidence to back your statement. Is this the Lange 2019 publication (https://gmd.copernicus.org/articles/12/3055/2019/gmd-12-3055-2019.html) you are citing?
P8-L220-22: Can you please elaborate on 6-hour bias adjustment modification more. It is not clear how you bias correct the 6-hour data. Do you define correction factors for the four 6-hour timesteps per month?
P8-L225: How was the EMO1-data aggregated?
P8-L225-227: Can you elaborate a bit more on the statistical downscaling. “Statistical relationship” is a bit vague. This description seems to deviate from the ISIMIP method.
P12-L323: How did you extrapolate the water withdrawal? Linearly?
P14-L358: Your criterion of station selection is not based on the record length. In the previous sentence you however mention that you have records ranging from 1 to 71 years, so does this mean, that for some stations your performance metric is based on 1 year only and others on 71 years of data? Can you please clarify this in the text. If this is the case, you might need to flag this in a way or check your performance results for subsets of stations. All stations vs. at least 30-year of stations vs. only a few years of data. This is otherwise an uneven performance quantification.
P13-14 L359-363: Spatial matching. How is the matching of river pixels and gauging stations done in the calibration? Is your approach different from this?
“LISTFLOOD coordinates”, does this mean that you use the same river pixels from the EFAS calibration? Also, not all of your 9 grid cells will be river pixels, right? If I am not mistaken LISTFLOOD has a routing scheme so for your comparison with a gauging station, you should likely use data from the routed river pixel. Wouldn`t it be more straightforward to find the closest river pixel and base the matching on the 2-3 upstream and 2-3 downstream river pixels?P14-L379: “Performance at daily scale”. Why daily scales and not 6-hourly scales? Would you expect that there is a noticeable difference in the performance metric when evaluated on a daily or 6-hourly scale?
P14-L388-89: KGE of -0.41. This is not clear. Clarify that you mean with this that you consider KGE values between -0.41 < kge < 1 as reasonable values and that your simulation is better than simply taking the mean values. You can use your sentence from P15-L403: “[…] a KGE>-0.41, meaning the reanalysis is skillful for these stations.”
Figure 5: Potentially the median is more meaningful than the mean here. In the figure caption the description of the dashed line is missing. Further, the ECDF is not explained. In the text you are also not commenting on the ECDF, therefore I would suggest removing the ECDF from the panel (a).
P16-L420: “Baltic countries” What is your hypothesis for the Baltic countries? Also, snowmelt processes? I briefly checked your Figure S2 and noticed that for the Baltic countries you have only a short record length. Therefore, I wonder if the poor performance in the Baltic regions is not related to the short record length. This directly corresponds to my comment from above on the inconsistencies with regard of the record length.
P16-L421: “southern Europe”. Do other studies, based on LISTFLOOD, also show this poor performance in southern Europe? Could this be a general issue of LISTFLOOD for dry catchments? Within the Alps many rivers are heavily influenced by reservoirs, and we don't see the degradation in performance. It could be worth checking other LISTFLOOD studies for similar results and potential explanations.
P16-L424: What is “negative 1.2es”?
P16-L426: “lower variability”. Is this a general issue of LISTFLOOD or is this due to lower variability in ERA5-land?
P18-L443: Sometimes you write HERA and other times HER. Please check for consistency.
P19-L473: Grammar. Here, we analyse how well ...
P19-L474: How long is your observational record for this analysis? In a previous section, and mentioned in a previous comment, you state that your records are between 1 and 71 years long. However, I have never seen a time constraint for the record length. Therefore, I am wondering here whether for some stations your quantiles are based on only 2 years of data while others represent 60+ years of data. Please clarify this. If this is the case, then your flow quantiles among the different stations mean totally different things.
P19-L474: “Person” -> “Pearson”
P21-L505: I think elaborating on the spatial patterns of the differences in timing would be worth including. For the mins 50% of your catchments show biases larger than 25 days. Figure 9 could be extended with two Maps, (eg. a) Map of differences in Annual-mins, b) Map of differences in Annual-max, c) violin plots.
P21-L506: The annual maxima/minima are not necessarily a flood or a drought event. I would rephrase this to "annual high and low flows".
Figure 10: Labels are quite small, especially the subplot titles. Please add (a)-(f) and include the rivers in the figure caption (e.g., (a) Ardeche, …). The ordering of the regime examples is also random. It would be easier to order them by regime type. Further, please extend the figure caption. What is the shading? Is the solid line the 30-year average?
P24-L540f: Please include appropriate figure referencing in the text (e.g., Figure 10b, c). Is connected to the above comment on Figure 10.
P24-L540-542: “For instance, …” I would suggest rephrasing this sentence. For example: “For the two pluvial rivers, the Schelde and the Ebro (Figure reference), we see opposite patterns of change. The Schelde shows an increase […], while the Ebro […].”
P24-L542: In the same sentence as above. Is this really a water deficit? Meaning it is not enough water? I would rather suggest to phrase this as "the entire regime of the Ebro shifts downward throughout the entire year.“
P24-L567-68: Can you briefly comment on how sensitive the LISTFLOOD model is to land use and water use changes? In most hydrological models very large disturbances or changes in the land use must be present to even see any influence on the hydrological response. Land use changes are very small as you have shown in your descriptions and these changes are based on the entire domain, so I would not assume any major differences arising from static and dynamic land use changes. Accounting for water use and its changes could locally certainly have some implications.
P25-L592-94: Here, maybe include a sentence that for long-term trends this needs to be considered and has to be included in any interpretation of results.
P25-L601: car -> can
P26-L614f: Do you think that one month spin-up is enough?
Citation: https://doi.org/10.5194/essd-2024-41-RC1
Data sets
HERA: a high-resolution pan-European hydrological reanalysis (1950-2020) A. Tilloy et al. http://data.europa.eu/89h/a605a675-9444-4017-8b34-d66be5b18c95
Model code and software
HERA Alois Tilloy https://github.com/Alowis/HERA
LISFLOOD code Joint Research Centre https://github.com/ec-jrc/lisflood-code
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 369 | 79 | 22 | 470 | 17 | 20 | 18 |
- HTML: 369
- PDF: 79
- XML: 22
- Total: 470
- Supplement: 17
- BibTeX: 20
- EndNote: 18
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1