the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 03 Aug 2023
Exploring patterns in precipitation intensity-duration-area-frequency relationships using weather radar data
Abstract. Accurate estimations of extreme precipitation return levels are critical for many hydrological applications. Extreme precipitation is highly variable in both space and time, therefore, to better understand and manage the related risks, knowledge of their probability at different spatial-temporal scales is crucial. We employ a novel non-asymptotic framework to estimate extreme return levels (up to 100 years) at multiple spatial-temporal scales from weather radar precipitation estimates. The approach reduces uncertainties and enables the use of relatively short archives typical of weather radar data (12 years in this case). We focus on the eastern Mediterranean, an area of high interest due to its sharp climatic gradient, containing Mediterranean, semi-arid and arid areas across a few tens of kilometres, and its susceptibility to flash-flood. At-site intensity-duration-area-frequency relations are derived from radar precipitation data at various scales (10 min–24 h, 0.25 km2–500 km2) across the study area, using ellipses of varying axes and orientations to account for the spatial component of storms.
We evaluate our analysis using daily rain gauge data over areas for which sufficiently dense gauge networks are available. We show that extreme return levels derived from radar precipitation data for 24 h and 100 km2 are generally comparable to those derived from averaging daily rain gauge data over a similar areal scale. We then analyse differences in multi-scale extreme precipitation over coastal, mountainous, and desert regions. Our study reveals that the power-law scaling relationship between precipitation and duration (simple scaling) weakens for increasing area sizes. This has implications for temporal downscaling. Additionally, precipitation intensity varies significantly for different area sizes at short durations, but becomes more similar at long durations, suggesting that, in the region, areal reduction factors may not be necessary for computing return levels over long durations. Furthermore, the reverse orographic effect, which causes decreased precipitation for hourly and sub-hourly durations, diminishes for larger areas. Finally, we discuss the effects of orography and coastline proximity on extreme precipitation intensity over different spatial-temporal scales.
- Preprint
(1955 KB) - Metadata XML
-
Supplement
(1070 KB) - BibTeX
- EndNote
Status: closed
-
RC1: 'Comment on egusphere-2023-1530', Anonymous Referee #1, 02 Oct 2023
The authors explore the patterns of rainfall intensity-duration-area-frequency (IDAF) curves derived from adjusted weather data on the eastern Mediterranean region. IDAF curves at durations from 10min to 1 day and area from 0.25km2 to 500km2 are derived by applying the simplified meta-statistical extreme value analysis (SMEV) on 18 years of available weather radar data. Their study concludes that the simple scaling of rainfall intensities with duration is only valid for point scale, that area reduction factors are mainly useful for short durations (<3h), and that the reverse orographic effect is weaken with larger areas. Overall, I find the study relevant, very well written and easy to read/follow. However, I have still some recommendations or points that I would like to discuss with the authors regarding the study:
- Figure 1 is a bit difficult to understand because it contains so much information. I would suggest that the background colour shows the land elevation and that the climate classification is given in semi-transparent polygons or lines. The size of the rain gauge-points can be a bit bigger so we can distinguish them better. Maybe the x and y axis for the right part of the figure can show the distance in meters from the weather radar location.
- I’m a little bit confused with the correction and adjustment of the radar data. So as far as I understood there are in total 3 adjustments performed to radar data based on the rain gauge information. So with the two first adjustments you are trying to adjust rainfall intensities (using daily stations), and then with the third one you are adjusting directly the SMEV parameters (using the 10min stations). I am wondering if all three steps are necessary and not redundant, since in the end you adjust the SMEV radar parameters according to the 10min station parameters. Could you please comment a bit more on the necessity of these three adjustments? Do you know how much the SMEV radar parameters are changing due to the correction based on daily stations (i.e. if you do only adjustment 3 vs adjustment 2 and 3, vs all adjustments together?
- Also regarding the parameter scaling of different areas based on rainfall stations, do you know how drastic the change in the parameters of the bigger areas is? It would be interesting to see how the mean value of the parameters of different duration and area are changing after the adjustment. So to have an idea how “wrong” the radar parameters are, and which duration and areas are mostly affected by it. Maybe this could explain also the convergence of the IDAF vcurves for longer durations? On the other hand, it would be also interesting to see what parameters are mainly differing with station based parameters (either shape, or scale or the number of ordinary events). This is probably outside the scope of your study, but maybe you can give your insight in the discussion based on your experience so far.
- Another thing that is not completely clear to me, is the identification of storms and ordinary events at the pixel scale. So you first determine the storm events based on daily average data (a total of 498 storm events). Then at each pixel for these storm events are you; a) either defining new “local” storm events, that can have completely different durations than the “regional” ones, or b) are you just checking which “regional” storms are manifested in this pixel and then decide whether to exclude them or not (but you keep the event duration same). I am asking because the events for each pixel are based on 10 min radar data, and it may be that the duration of such events is shorter than 24 hours (which would them compromise your fixed number of events over different durations).
- Following the explanation on line 210-211, is the number of ordinary events reduced according to the 55th percent, or just the input series for the CDF fitting is reduced to leave out the 55% of the events? Also, in Line 210 you mention than censoring between 55th to 80th quantile doesn’t influence much the results, but then still why did you choose to censor below the 55th quantile?
- Line 240, could you please describe shortly the bootstrapping from Overeem et al. 2008? Does it pool together all stations inside a region and samples from pooled storms, or is it just storm sampling with replacement from a single series?
- Lines 251-252, why do you validate the radar data based on station data of another time period? Wouldn’t this also punish more the radar data IDAF curves?
- At line 260, you mention the discrepancy between radar and daily station IDF curves due to different daily measurements. Since radar is at 10mins, wasn’t it possible to calculate the daily maximum intensity according to the daily measurement times (between 6 am to 6 am)?
- Figure 3 – c, I agree it might be the distance to the radar station that is causing such overestimation. However, this pattern is not consistent with Figure 4, as we see that in the region near to the radar station there is a clear underestimation (or very little rainfall). I was wondering if there is a specific parameter that is overestimated in this area that might be directly link with this IDF overestimation? Can it be that the adjustment to 10min station data parameters had something to do with the overestimation (like the density of 10min station data in the vicinity)?
- At section 4.3 you explain how the figures are derived, however you mention in line 313 a 5 by 10 km2 box, and then on line 319 a 10 by 10 km2 Is this a typo, or these are actually two different types of box-sizes used for the investigation?
- Figure 5 – I think it is also interesting to point out the duration when the areas converge for these three regions. In the desert the convergence happens at 1 hour, while at coast and mountains at 3 hours. Do you have any explanation for that? Maybe to explain why the IDAF curves are converging after a certain duration, it may be useful to have a look at the SMEV parameters and see how they are changing with duration and area, or even see the average characteristics of the ellipses for each duration and area; so for instance if for 24 h duration the axis ratio of the ordinary events is closer to 1 than those of 1h duration, or even the spatial variability of the rainfall within an ellipse for different durations and areas.
- Also the results from Figure 5 are a bit controversial, as I would expect that the ARF are dependent on duration and area (see for instance Overeem et al. 2010), and in my opinion these results should be discussed more. Line 375-397 – here you are discussing about other studies that have more or less contrary results to your investigation. The main reason for this contrast, you list the different study areas. However, might there be other factors like the methodology applied or the data used that might explain the difference in the results (i.e. use of ellipses instead of circles, use of SMEV instead of GEV and so on). Lastly is the same pattern as shown in Figure 5 also valid for other locations, i.e. the validation sites or other random sites?
- In Section 5.2 (more specifically starting from Line 410 and on) you mention that the power-law relation weakens as the area size increases. Do you know of any other study that might back you up in this conclusion?
- It seems that this works is largely based on the previous work of Marra et al. 2022. Maybe you can consider to join this paper with the previous one, so readers will go directly to the previous one if they have any questions.
References:
Overeem, A., T. A. Buishand, I. Holleman, and R. Uijlenhoet (2010), Extreme value modeling of areal rainfall from weather radar, Water Resour. Res.,46, W09514, doi:10.1029/2009WR008517
Citation: https://doi.org/10.5194/egusphere-2023-1530-RC1 - AC2: 'Reply on RC1', Talia Rosin, 04 Dec 2023
-
RC2: 'Comment on egusphere-2023-1530', Anonymous Referee #2, 17 Oct 2023
Review of Rosin et al. Exploring patterns in precipitation intensity-duration-area-frequency relationships using weather radar data
The paper explores precipitation patterns and rainfall statistics in space and time based on weather radar data and the application of the method SMEV. This method is described in detail in Marra et al (2022) for radar pixel values however here the method is extrapolated to also include spatial rainfall and thus dependence of area.
The paper is generally well-written and understandable. Several things are indeed interesting – especially the pixel-area relations depending on rainfall duration as well as the significant climatological differences in the study area. Below I have added a few clarification and additions that can help to clarify the manuscript:
- The novelty of the paper should be emphasized more in the introduction. I guess compared to former efforts, the novelty here is the application of SMEV also on the areal component rather than single pixels?
- In my view, it is not clear what the differences between MEV and SMEV are. The authors refer to past publications, but could be relevant to describe the SMEV in a bit more detail here in order to understand how it differs from other methods of extreme value statistics. For example line 157-158 could be detailed further.
- The equation in line 128: Shouldn’t it be Z=316R^1.5?
- Do you think the discrepancy between gauge and radar is due to differences in climatology, bias adjustment, or radar artifacts? Consulting figure 4 it seems that there are some radar issues close to the radar – maybe something related to scanning and CAPPI generation?
- Figure 5. I think it would be very interesting also to include the rain gauges statistics in this figure – and in principle also in figures 6-7. Typically, you would see an underestimation of the radar estimates at short durations. See e.g. Schleiss et al (2020) or Andersen et al 2021. The point scale is indeed interesting to study in addition to the study of the difference between pixel scale and areas of 10, 100 and 500 km2. In regards to point 1 in the conclusion the actual comparison in the manuscript is not made on point scale but on a pixel scale. Even going from point to pixel scale will result in scaling – which will be more dominant for shorter durations than larger ones.
- It would be relevant to present fitted parameter values of shape, scale, kappa, lambda in order to study how the vary over different durations and area. Maybe present selected values in a table. Potentially an empirical relation between model parameter estimates and duration and area could be sought.
Andersen, C.B.; Wright, D.B.; Thorndahl, S. Sub-Hourly to Daily Rainfall Intensity-Duration-Frequency Estimation Using Stochastic Storm Transposition and Discontinuous Radar Data. Water 2022, 14, 4013.
Schleiss, M., Olsson, J., Berg, P., Niemi, T., Kokkonen, T., Thorndahl, S., Nielsen, R., Nielsen, J.E., Bozhinova, D., Pulkkinen, S. (2020). The accuracy of weather radar in heavy rain: a comparative study for Denmark, the Netherlands, Finland and Sweden. Hydrology and Earth System Sciences, 24, 3157–3188,
Citation: https://doi.org/10.5194/egusphere-2023-1530-RC2 - AC1: 'Reply on RC2', Talia Rosin, 04 Dec 2023
Status: closed
-
RC1: 'Comment on egusphere-2023-1530', Anonymous Referee #1, 02 Oct 2023
The authors explore the patterns of rainfall intensity-duration-area-frequency (IDAF) curves derived from adjusted weather data on the eastern Mediterranean region. IDAF curves at durations from 10min to 1 day and area from 0.25km2 to 500km2 are derived by applying the simplified meta-statistical extreme value analysis (SMEV) on 18 years of available weather radar data. Their study concludes that the simple scaling of rainfall intensities with duration is only valid for point scale, that area reduction factors are mainly useful for short durations (<3h), and that the reverse orographic effect is weaken with larger areas. Overall, I find the study relevant, very well written and easy to read/follow. However, I have still some recommendations or points that I would like to discuss with the authors regarding the study:
- Figure 1 is a bit difficult to understand because it contains so much information. I would suggest that the background colour shows the land elevation and that the climate classification is given in semi-transparent polygons or lines. The size of the rain gauge-points can be a bit bigger so we can distinguish them better. Maybe the x and y axis for the right part of the figure can show the distance in meters from the weather radar location.
- I’m a little bit confused with the correction and adjustment of the radar data. So as far as I understood there are in total 3 adjustments performed to radar data based on the rain gauge information. So with the two first adjustments you are trying to adjust rainfall intensities (using daily stations), and then with the third one you are adjusting directly the SMEV parameters (using the 10min stations). I am wondering if all three steps are necessary and not redundant, since in the end you adjust the SMEV radar parameters according to the 10min station parameters. Could you please comment a bit more on the necessity of these three adjustments? Do you know how much the SMEV radar parameters are changing due to the correction based on daily stations (i.e. if you do only adjustment 3 vs adjustment 2 and 3, vs all adjustments together?
- Also regarding the parameter scaling of different areas based on rainfall stations, do you know how drastic the change in the parameters of the bigger areas is? It would be interesting to see how the mean value of the parameters of different duration and area are changing after the adjustment. So to have an idea how “wrong” the radar parameters are, and which duration and areas are mostly affected by it. Maybe this could explain also the convergence of the IDAF vcurves for longer durations? On the other hand, it would be also interesting to see what parameters are mainly differing with station based parameters (either shape, or scale or the number of ordinary events). This is probably outside the scope of your study, but maybe you can give your insight in the discussion based on your experience so far.
- Another thing that is not completely clear to me, is the identification of storms and ordinary events at the pixel scale. So you first determine the storm events based on daily average data (a total of 498 storm events). Then at each pixel for these storm events are you; a) either defining new “local” storm events, that can have completely different durations than the “regional” ones, or b) are you just checking which “regional” storms are manifested in this pixel and then decide whether to exclude them or not (but you keep the event duration same). I am asking because the events for each pixel are based on 10 min radar data, and it may be that the duration of such events is shorter than 24 hours (which would them compromise your fixed number of events over different durations).
- Following the explanation on line 210-211, is the number of ordinary events reduced according to the 55th percent, or just the input series for the CDF fitting is reduced to leave out the 55% of the events? Also, in Line 210 you mention than censoring between 55th to 80th quantile doesn’t influence much the results, but then still why did you choose to censor below the 55th quantile?
- Line 240, could you please describe shortly the bootstrapping from Overeem et al. 2008? Does it pool together all stations inside a region and samples from pooled storms, or is it just storm sampling with replacement from a single series?
- Lines 251-252, why do you validate the radar data based on station data of another time period? Wouldn’t this also punish more the radar data IDAF curves?
- At line 260, you mention the discrepancy between radar and daily station IDF curves due to different daily measurements. Since radar is at 10mins, wasn’t it possible to calculate the daily maximum intensity according to the daily measurement times (between 6 am to 6 am)?
- Figure 3 – c, I agree it might be the distance to the radar station that is causing such overestimation. However, this pattern is not consistent with Figure 4, as we see that in the region near to the radar station there is a clear underestimation (or very little rainfall). I was wondering if there is a specific parameter that is overestimated in this area that might be directly link with this IDF overestimation? Can it be that the adjustment to 10min station data parameters had something to do with the overestimation (like the density of 10min station data in the vicinity)?
- At section 4.3 you explain how the figures are derived, however you mention in line 313 a 5 by 10 km2 box, and then on line 319 a 10 by 10 km2 Is this a typo, or these are actually two different types of box-sizes used for the investigation?
- Figure 5 – I think it is also interesting to point out the duration when the areas converge for these three regions. In the desert the convergence happens at 1 hour, while at coast and mountains at 3 hours. Do you have any explanation for that? Maybe to explain why the IDAF curves are converging after a certain duration, it may be useful to have a look at the SMEV parameters and see how they are changing with duration and area, or even see the average characteristics of the ellipses for each duration and area; so for instance if for 24 h duration the axis ratio of the ordinary events is closer to 1 than those of 1h duration, or even the spatial variability of the rainfall within an ellipse for different durations and areas.
- Also the results from Figure 5 are a bit controversial, as I would expect that the ARF are dependent on duration and area (see for instance Overeem et al. 2010), and in my opinion these results should be discussed more. Line 375-397 – here you are discussing about other studies that have more or less contrary results to your investigation. The main reason for this contrast, you list the different study areas. However, might there be other factors like the methodology applied or the data used that might explain the difference in the results (i.e. use of ellipses instead of circles, use of SMEV instead of GEV and so on). Lastly is the same pattern as shown in Figure 5 also valid for other locations, i.e. the validation sites or other random sites?
- In Section 5.2 (more specifically starting from Line 410 and on) you mention that the power-law relation weakens as the area size increases. Do you know of any other study that might back you up in this conclusion?
- It seems that this works is largely based on the previous work of Marra et al. 2022. Maybe you can consider to join this paper with the previous one, so readers will go directly to the previous one if they have any questions.
References:
Overeem, A., T. A. Buishand, I. Holleman, and R. Uijlenhoet (2010), Extreme value modeling of areal rainfall from weather radar, Water Resour. Res.,46, W09514, doi:10.1029/2009WR008517
Citation: https://doi.org/10.5194/egusphere-2023-1530-RC1 - AC2: 'Reply on RC1', Talia Rosin, 04 Dec 2023
-
RC2: 'Comment on egusphere-2023-1530', Anonymous Referee #2, 17 Oct 2023
Review of Rosin et al. Exploring patterns in precipitation intensity-duration-area-frequency relationships using weather radar data
The paper explores precipitation patterns and rainfall statistics in space and time based on weather radar data and the application of the method SMEV. This method is described in detail in Marra et al (2022) for radar pixel values however here the method is extrapolated to also include spatial rainfall and thus dependence of area.
The paper is generally well-written and understandable. Several things are indeed interesting – especially the pixel-area relations depending on rainfall duration as well as the significant climatological differences in the study area. Below I have added a few clarification and additions that can help to clarify the manuscript:
- The novelty of the paper should be emphasized more in the introduction. I guess compared to former efforts, the novelty here is the application of SMEV also on the areal component rather than single pixels?
- In my view, it is not clear what the differences between MEV and SMEV are. The authors refer to past publications, but could be relevant to describe the SMEV in a bit more detail here in order to understand how it differs from other methods of extreme value statistics. For example line 157-158 could be detailed further.
- The equation in line 128: Shouldn’t it be Z=316R^1.5?
- Do you think the discrepancy between gauge and radar is due to differences in climatology, bias adjustment, or radar artifacts? Consulting figure 4 it seems that there are some radar issues close to the radar – maybe something related to scanning and CAPPI generation?
- Figure 5. I think it would be very interesting also to include the rain gauges statistics in this figure – and in principle also in figures 6-7. Typically, you would see an underestimation of the radar estimates at short durations. See e.g. Schleiss et al (2020) or Andersen et al 2021. The point scale is indeed interesting to study in addition to the study of the difference between pixel scale and areas of 10, 100 and 500 km2. In regards to point 1 in the conclusion the actual comparison in the manuscript is not made on point scale but on a pixel scale. Even going from point to pixel scale will result in scaling – which will be more dominant for shorter durations than larger ones.
- It would be relevant to present fitted parameter values of shape, scale, kappa, lambda in order to study how the vary over different durations and area. Maybe present selected values in a table. Potentially an empirical relation between model parameter estimates and duration and area could be sought.
Andersen, C.B.; Wright, D.B.; Thorndahl, S. Sub-Hourly to Daily Rainfall Intensity-Duration-Frequency Estimation Using Stochastic Storm Transposition and Discontinuous Radar Data. Water 2022, 14, 4013.
Schleiss, M., Olsson, J., Berg, P., Niemi, T., Kokkonen, T., Thorndahl, S., Nielsen, R., Nielsen, J.E., Bozhinova, D., Pulkkinen, S. (2020). The accuracy of weather radar in heavy rain: a comparative study for Denmark, the Netherlands, Finland and Sweden. Hydrology and Earth System Sciences, 24, 3157–3188,
Citation: https://doi.org/10.5194/egusphere-2023-1530-RC2 - AC1: 'Reply on RC2', Talia Rosin, 04 Dec 2023
Data sets
Radar precipitation data Efrat Morin, Talia Rosin & Francesco Marra https://ims.gov.il/en
Rain gauge data Efrat Morin, Talia Rosin & Francesco Marra https://ims.gov.il/en
Model code and software
A Unified Framework for Extreme Sub-daily Precipitation Frequency Analyses based on Ordinary Events Francesco Marra https://zenodo.org/record/3971558
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 267 | 180 | 30 | 477 | 49 | 16 | 18 |
- HTML: 267
- PDF: 180
- XML: 30
- Total: 477
- Supplement: 49
- BibTeX: 16
- EndNote: 18
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1