the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
FOREWARNS: Development and multifaceted verification of enhanced regional-scale surface water flood forecasts
Cathryn E. Birch
Steven J. Böing
Thomas Willis
Linda Speight
Aurore N. Porson
Charlie Pilling
Kay L. Shelton
Mark A. Trigg
Abstract. Surface water flooding (SWF) is a severe hazard associated with extreme convective rainfall, whose spatial and temporal sparsity belies the significant impacts it has on populations and infrastructure. Forecasting the intense convective rainfall that causes most SWF on the temporal and spatial scales required for effective flood forecasting remains extremely challenging. National scale flood forecasts are currently issued for the UK and are well regarded amongst flood responders, but there is a need for complimentary enhanced regional information. Here we present a novel SWF forecasting method, FOREWARNS (Flood fOREcasts for surface WAter at a RegioNal Scale), that aims to fill this gap in forecast provision. FOREWARNS compares reasonable worst-case rainfall from a neighbourhood-processed, convection-permitting ensemble forecast system against pre-simulated flood scenarios, issuing a categorical forecast of SWF severity. We report findings from a workshop structured around three historical flood events in Northern England, in which forecast users indicated they found the forecasts helpful and would use FOREWARNS to complement national guidance for action planning in advance of anticipated events. We also present results from objective verification of forecasts for 82 recorded flood events in Northern England from 2013–2022, and for 725 daily forecasts spanning 2019–2022, using a combination of flood records and precipitation proxies. We demonstrate that FOREWARNS offers good skill in forecasting SWF risk, with high spatial hit rates and low temporal false alarm rates, confirming that user confidence is justified, and that FOREWARNS would be suitable for meeting the user requirements of an enhanced operational forecast.
- Preprint
(2570 KB) - Metadata XML
-
Supplement
(1756 KB) - BibTeX
- EndNote
Ben Maybee et al.
Status: open (extended)
-
RC1: 'Comment on nhess-2023-83', Pierre Javelle, 31 Aug 2023
reply
General comment
This paper presents a new method, FOREWARNS, which allows regional scale forecasting of surface water floods (SWF). The subject of SWF is of great importance and the present study gives some interesting insights that could interest the whole community of operators and researchers dealing with this issue. The paper is very well written and documented. A comprehensive database was used. Results are well detailed and analysed. This is a very good paper.
My only concern is the lack of information on the calculation of flood return periods, in particular how initial soil moisture is taken into account by FOREWARNS. It seems to me that it is not (but I maybe misunderstood). For me, this is a critical point that should be better explained and where the main limitations of the method should lie.
Detailed comments
L39-40, « severity and frequency » What is the distinction here ? In the text after, you seem to make no distinction. Please homogeneise the vocabulary (see also rq . L141)
L110 , « Tennant, 2015 » : this reference is missing
L141 , « severity » : Say clearly that for you, severity = frequency, expressed in term of return period. Note that this definition does not take into account the "risk"but only the "hazard" (see also rq L39-40, where you seems to do the distinction.)
L147 , « multiple ensemble member fields » : Please explain how do you deal with ensemble to obtain one unique FOREWARNS forecast ? In all the study, it seems you deal with deterministic forecasts (see fig 1 or 2 for example).
L165, figure 1 : Please indicate the legend legend for black contours. Local authority boundaries ?
L170-178 : this part is very unclear. You don’t explain how exactly your flood return periods are obtained. You explain how you pass from a multi-frequency hyetogram to a mono frequency (using your national scenari), but you don’t explain how you pass from rainfall to flood (since rainfall return period is not equal to flood return period, it also depends on initial condition on the catchement). If I understand, you don't use G2G runoff outputs (runoff) as in SWFHIM… is it correct ?
L186 : why do you choose the centroid ? Wouldn’t be possible to study all the return period estimated using your method, and then take the higest return period in each catchment ?
L187-188 : What is the physical motivation in the definition of those catchments. In particular, why do you choose smaller catchments only in urban areas (and not for all) ? Why do you say it is « hydrologically consistent » ? What exactely is behind this agregation / sampling procedure in term of final results (just for mapping, or more deep consequencecies on the results?)
L200, « MOGREPS-UK ensemble » : could you explain how do you deal with different members to obtain one unique forecast (see also rq L147)
L213, « more than twice that could be subjectively verified as SWF » : What do you mean by that ? What is your "subjective" definition of SWF ? A return period > 5 year ?
L220, Figure 2 : the quality of the image should be improved (difficult to read the FGS column)
L235 « RoSWF » : I don t understand how RoSWF have been used in this study. For me it has not been used but I maybe misunderstood (same for G2G, see my remark L170-178).
L238 « Fig S3 » : this figure is cited before S2
L240 « inevitably overestimates » : For me, there is also 2 explanations for this strong over estimation (median at 90% of catchments in "false alarm" if I understand correctly Fig S3) :
1) You use "only" rainfall threshold to define a flood, thus, antecedent conditions are not taken into account. Better to use net rainfall or output of an hydrological model
2) You consider "only" the hazard level to define a SWF and not the intensity of damage. So maybe your level of return period has been reached, but without damage, and nobody has noticed/reported.
L251 « contingency tables » : please say that n = nb of day for temporal, and nb of catchment for spatial verification (if I correctly undersand)
L270 « from a visual inspection of forecast-proxy pairs » : This is the only explanation you give. It is not possible for the reader to understand what has been made (see remark L524 ).
L276-282 « equitable » : I understand that the number of d (true negatives) are not accounted, but please explain it is more « equitable ».
L305 figure 3 : Please indicate your threshold defining an event (5 or 30-year)?
L307-412, all : this part is very interesting. However, I would put it after the results of 4., and I would try to make it shorter. Indeed, there is already an interesting discussion about the usefulness of the method (from an user point of view) that should be said at the end, rather than here.
L399-412 : I think for a « far » comparison with actual SWF FGS, it should be said somewhere that FOREWARNS does not tale into account initial moisture condition (from my view this is its main limitation, I maby misunderstood this point), and concerning the « improved level of local detail » (see L400), that SWF FGS products have been aggregated to the county level, but results of SWFHIM (see Aldridge et al., 2020) could be aggregated over any areas, for example the FOREWARNS catchments. So I think in the future, what should be done is to merge SWFHIM and FOREWARNS (ie apply SWFHIM with RWCRSs and aggregated at a catchment levels (and using smaller catchments). This is just a personnal remark. No need to take it into account in your paper.
L383 « figure S2 » is cited after figure S3
L424 «for all but 4 events. R » : Which ones ? Which criteria ? Is it the same visual analysis than later (see remark L524)
L437 Figure 5 : Please indicate your threshold defining an event (5 or 30-year)?
L437 Figure 5 , bis : Do we to have 28 points of the figure (one by day) ? Because I see only 24.
L448 «half of recorded flood locations were still successfully » : Do you see this result on figure 4, event 27 for the forecast (left column) ? I see only one recorded flood location correctly identified. I maybe misunderstood.
L449« Fig. S2 » is cited after figure S3
L489 « 4.3 » : should be 4.2 ?
L491-493 « There are … in total » : it is easier to understand if you change this by : « There are 41 days during this period with recorded SWF events, of which **29 were correctly detected** by the radar proxy. An additional 79 days showed SWF in the radar proxy, yielding 108 proxy flood days in total. »
L491-493 bis : this means that for 79 over the 108 « proxy flood days » no flood has been « observed/reported ». If I correctly understood, this seems huge. For me, this is because you don’t take into account initial soil moisture conditions AND damage impacts in your method (see my rq L240)
L498 « for spatial skill scores » : replace by « for spatial ** and temporal ** skill scores
L522-523 « do not account for the overall spatial distribution » : if I understood you gave spatial scores into Fig 6S, first row.
L524 « see Sect. 2.3.2 » : there is very few explanation in section 2.3.2. (see remark L270). I don t understand the results. Could you please give more details after line 270 ?
L565-end « Summary and Recommendations » should be more nuanced (taking into account my remarks L240, L399-412, L491-493 bis...)
in particular sentences as :
L576 « FOREWARNS would complement existing efforts to improve global flash flood guidance ». I agree that it is a very good idea to use RWCRSs in any other warning methods. But I don’t understand the way FOREWARNS translates that in ** flood ** return period. For instance FFG methods (as SWFHIM in UK) use a rainfall runoff-model taking into account initial soil moisture conditions. If this is not the case for FOREWARNS, this is a limitation, and this has to be said clearly (that does not mean that FOREWARNS is useless).
L582 « Our solution, a combination of recorded events and radar proxy observations of SWF ». Again, I would be more nuanced. With these solution, many non flood event are identified as flood event. This is a « first » level verification that has to be done… By doing that, you are verifying the « forecasted » RWCRSs against the « radar observed » RWCRSs, which is good. But say also clearly that you are staying on the ** rainfall hazard ** part of the verification, and that, in the present paper, you have not yet verified the ** flood damage impact ** part.
Citation: https://doi.org/10.5194/nhess-2023-83-RC1
Ben Maybee et al.
Ben Maybee et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 391 | 105 | 10 | 506 | 24 | 5 | 6 |
- HTML: 391
- PDF: 105
- XML: 10
- Total: 506
- Supplement: 24
- BibTeX: 5
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1