the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Enhanced Evaluation of Sub-daily and Daily Extreme Precipitation in Norway from Convection-Permitting Models at Regional and Local Scales
Abstract. Convection-permitting regional climate models (CPRCMs) have demonstrated enhanced capability in capturing extreme precipitation compared to regional climate models (RCMs) with convection-parameterization schemes. Despite this, a comprehensive understanding of their added values in daily or sub-daily extremes, especially at local scale, remains limited. In this study, we conduct a thorough comparison of daily and sub-daily extreme precipitation from HARMONIE-Climate model, cycle 38 at 3 km resolution (HCLIM3) and 12 km resolution (HCLIM12) across Norway’s diverse landscape, divided into eight regions, using both gridded and in-situ observations. Our main focus is to investigate the added value of CPRCMs (i.e., HCLIM3) compared to RCMs (i.e., HCLIM12) for extreme precipitation at daily and sub-daily scales from regional to local scales, and quantify to what extend CPRCM can reproduce the orographic effect on extreme precipitation at daily and sub-daily scale. We find that HCLIM3 better matches observations than HCLIM12 for daily and sub-daily precipitation extreme indices at regional scale in Norway. More specifically, HCLIM3 better captures the maximum 1-day precipitation (Rx1d) at most of the regions except south-western region in Norway. Notably, HCLIM12 shows underestimation in the complex orography for annual Rx1d. For the maximum 1-hour precipitation (Rx1h), the superiority from HCLIM3 have also been found on average, although with slightly higher wet-bias in the western, middle-inland and middle-coastal during summer. In addition, the reverse orography effect on seasonal Rx1h at regional scale can be better reproduced by HCLIM3 than HCLIM12 in most seasons except spring. At the local scale, HCLIM3 can better capture the temporal evolution of Rx1h than HCLIM12 when compared with observations between 1999–2018. However, we see that the benefit from HCLIM3 in capturing seasonal Rx1d within western region diminishes at local scale. Most interesting finding is that the added value from HCLIM3 is clearer in Rx1h than in Rx1d at both regional and local scale, especially in the extreme seasonality. In general, HCLIM3 performs better than HCLIM12 on Rx1d and Rx1h in Norway with the mean of bias distribution closer to zero, although it varies a bit among regions. Specifically, HCLIM3 performs slightly poorer in the south-western region. This study highlights the importance of more realistic convection-permitting regional climate simulations in providing reliable insights into the characteristics of precipitation extremes across Norway's eight regions. Such information is crucial for effective adaptation management to mitigate severe hydro-meteorological hazards, especially for the local extremes.
- Preprint
(5629 KB) - Metadata XML
-
Supplement
(1254 KB) - BibTeX
- EndNote
Status: open (extended)
-
RC1: 'Comment on hess-2024-68', Anonymous Referee #1, 24 May 2024
reply
Review for “Enhanced Evaluation of Sub-daily and Daily Extreme Precipitation in Norway from Convection-Permitting Models at Regional and Local Scales “ by Xie et al.
The paper presents an evaluation of extreme precipitation from two regional climate models; one is a regional model at 12km resolution, the other is a convection-permitting model at 3km resolution. The evaluation is based on different observational products: two grid products at 1day and 1hour temporal resolution, and a network of rain gauges (about 190 at daily resolution and 10 at hourly resolution). Seasonality and relation with elevation are also explored, at regional (by contrast with the gridded product) and local scale (by contrast with rain gages). The main findings indicate generally better performance for the HCLIM3 than the HCLIM12 model, more clearly at hourly resolution.
The study is of interest on the general topic of evaluation of extreme precipitation from convection-permitting models, which may help in better understanding how to use them in practical applications. In my opinion, it gives an incremental advancement in this field more than novelty, considering the regional scale (Norway) and the use of metrics commonly used in these kind of studies (Rx1d, Rx1h, return levels). It is well written, but I found difficult to get the main messages on the results because of the total length, number of figures and panels. The topic is of interest, and fitting the journal scopes, but I suggest a few major revisions and some minor before publication in HESS. My comments are listed below.
- Major 1. You used 8 regions. I wonder if less can be used, pooling together smaller ones. I say this based on two considerations: 1) extension is very different across the regions, and regional evaluation of models are then based on quite different number of grid points (for example, how many grid points for the two small regions in the south?) and number of daily rain gages (just 4-11-14 for three regions!); 2) it is difficult to follow the explanations and figures with comparisons on 8 regions, 2 models, 4 seasons, 2 durations … and get a message on the results; maybe having less could help.
- Major 2. Regional scale is here referred to the analysis using the gridded products; local scale is referred to the analysis based on rain gages. They show different results, but I wonder how much this is due to the use of different observation products. How is SeNorge-models comparison sampled on the same rain gage points? Or, how is seNorge compared to rain gages? Differences you highlight in your text (e.g lines 337, 352, 384, etc) could be due to the use of a different observational benchmark. Moreover, when comparing gridded products and climate model, you compare values from a same size grid (12x12km), when comparing model and rain gages, the comparison is made on a 12x12km grid and a point measurement. I suggest to add further analysis comparing seNorge with rain gages, or extracting the comparison between seNorge vs models on the same locations of rain gages, or to add some considerations in the discussion.
- Major 3. Biases (e.g. figure 2, 6) are shown in mm (absolute differences). Maybe relative differences (modelled-observed)/observed could be more meaningful considering that precipitation has a big range across the regions: a bias of 5 mm is different on a 30 mm or a 80 mm daily precipitation! My suggestion is to update maps and plots of biases with relative bias (%), and to revise description of results and comments on “magnitude” of bias based on this.
Minor comments
- title: I suggest hourly in place of sub-daily, considering that you just evaluate 1h
- Line 120: these results on orographic effect “were based on the annual maxima”… yes. But also yours are based on Rx1d (see line 222-223). So … why do you highlight these about other studies at line 120, if then you do the same? I suggest to remove.
- Line 121-122. “The dependence on seasonality … need the evaluation based on season”. Of course! Maybe you wanted to tell something different here and I didn’t get it. Please calrify/modify.
- Line 195. Not clear: you say here you averaged the indices in the region … then in caption of figure 2 you write the bias is calculated at each grid point (this makes sense to me). So, when do you use the averaged indices?
- Line 206. Specify somewhere in the section that this is done for daily data (both gridded and rain gages), while just on 10 rain gages at 1h duration. More importantly, later in the text (lines 367, 372,..) you mention uncertainty … and it is never explained before in the paper how it is evaluated. Add it in the methodology.
- Figure 2 (same for figures 6 and 7). The caption mentions absolute bias but describes a relative bias calculation. Please correct.
- Figure 2c (same for figures 6 and 7). I suggest to add another row with the regional bias for annual Rx1d, not just seasonal, in order to have a synthesis of what is shown in the maps in panel a.
- Figure 4. Same y-axis limits could help in comparing the bias… and maybe you can revise the description of results better considering the different magnitude of the bias (example at lines 290-291).
- Figure 4. I can’t understand why Northern-Coastal has so big bias for HCLIM3, considering that figure 3 shows a tendency of underestimation of the empirical distribution, similar to Northern-Inland, for which you find big underestimation of return levels. And for the Norther-Inland, why underestimation of return levels is bigger for HCLIM3, while the distribution in figure 3 show more underestimation for HCLIM12? … please check the correctness of the results.
- Line 370. For 50yr return-periods they seem identical, not larger bias for HCLIM3. I would remove it
- Line 439-440. This is based on just 10 points. This can’t be considered a general finding, I suggest to mention the limit of the analysis.
- Line 483, section 4.5. You show the slope of precipitation with elevation as absolute value, mm/km. I strongly suggest to calculate and show it as relative slope, for example with respect to the average value of Rx. Because 1mm/km has a different magnitude for Rx1d and Rx1h. Then I suggest to revise your discussion considering this … (e.g. I see very weak relation of Rx1d with elevation, so I’m sure you can really speak about reverse orographic effect …also at line 629)
- Line 494. “Significant”? Based on a specific test? Maybe “relevant”…
- Line 561-562. Not very informative consideration …. Could you elaborate more on this? Or delete …
- Line 580-582. I see here two contrasting points. 1) You mention underestimation for return levels, but for Rx1d in figure 8 I see bias around zero, while for Rx1h you have evaluation on just 10points. 2) Then you say this is in line with results in Malawi (!!!!) finding overestimation. I can’t really understand your reasoning here.
- Line 587. I can’t understand the meaning of “weakening the superior” …
- Line 635. Dallan et al. 2023 analyzed annual Rx1d: this can’t be related with the seasonal Rx1d. I suggest to rephrase in some way: ”An unclear relation of Rx1d with elevation at regional scale was also seen from the study of Dallan et al. (2023), in which, they analyzed annual Rx1d based on CPRCMs and in-situ observation over Alpine”
- Line 640. I suggest to add a few recent references on orographic enhancement at daily scale observed in different regions (e.g. Formetta et al. 2021 https://doi.org/10.1016/j.advwatres.2021.104085 and Amponsah et al. 2022 https://doi.org/10.1016/j.jhydrol.2022.128090); same at line 651 for the reverse orographic effect, adding also Formetta et al 2021, considering they explored durations from subhourly to daily.
- Please also revise your conclusions accordingly to the modifications you will do in the revised version of the manuscript
Citation: https://doi.org/10.5194/hess-2024-68-RC1 -
AC1: 'Reply on RC1', Lu Li, 25 Jun 2024
reply
The comment was uploaded in the form of a supplement: https://hess.copernicus.org/preprints/hess-2024-68/hess-2024-68-AC1-supplement.pdf
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
416 | 78 | 35 | 529 | 45 | 34 | 28 |
- HTML: 416
- PDF: 78
- XML: 35
- Total: 529
- Supplement: 45
- BibTeX: 34
- EndNote: 28
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1