the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 22 Sep 2022
A climate-conditioned catastrophe risk model for UK flooding
Abstract. We present a climate-conditioned catastrophe flood model for the UK that simulates pluvial, fluvial and coastal flood risks at 1 arc second spatial resolution (~20–25 m). Hazard layers for ten different return periods are produced over the whole UK for historic, 2020, 2030, 2050 and 2070 conditions using the UKCP18 climate simulations. From these, monetary losses are computed for Great Britain only for five specific global warming levels (0.6, 1.1, 1.8, 2.5 and 3.3 °C). The analysis contains a greater level of detail and nuance compared to previous work and represents our current best understanding of the UK’s changing flood risk landscape. Validation against historical national return period flood maps yielded Critical Success Index values of 0.65 and 0.76 for England and Wales respectively, and maximum water levels for the Carlisle 2005 flood were replicated to an RMSE of 0.41 m without calibration. This level of skill is similar to local modelling with site specific data. Expected Annual Damage in 2020 was £730M, which compares favourably to the observed value of £714M reported by the Association of British Insurers. Previous UK flood loss estimates based on government data are ~3x higher and lie well outside our modelled loss distribution, which is plausibly centred on the observations. We estimate that UK 1 % annual probability flood losses were ~6 % greater in the average climate conditions of 2020 than for the period of historical river flow and rainfall observations (centred approximately on 1995) and can be kept to around ~8 % if all countries’ COP26 2030 carbon emission reduction pledges and ‘net zero’ commitments are implemented in full. Implementing only the COP26 pledges increases UK 1 % annual probability flood losses by ~23 % above recent historical values, and potentially ~37 % if climate sensitivity turns out to be higher than currently thought.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(3582 KB)
-
Supplement
(5005 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3582 KB) - Metadata XML
-
Supplement
(5005 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
RC1: 'Referee Comment on egusphere-2022-829', Anonymous Referee #1, 24 Oct 2022
This paper summarizes the impressive work done by Paul Bates and colleagues in flood risk modelling. While the details are left to other more technical papers, this work concentrates on the big picture and answers to the call for peer-reviewable flood risk assessments (in the UK but I would say also elsewhere). Because of this, I believe that the paper fits is appropriate for NHESS. I only have a couple of minor suggestions (and some more detailed comments below):
- Despite the title and abstract focus on climate scenarios, in the paper, a lot of emphasis is given to the transparency and consistency of the hazard and risk maps obtained through this modelling chain as opposed to the "opaque" official maps. The climate change analysis, instead, is not fully developed. The assessment of uncertainty with an envelope of climate models (and for different emission scenarios) would be a standard requirement (and the estimation of additional uncertainties of the hydrologic and hydraulic models would be even better). My suggestion would be to rephrase title and abstract to reflect the weight given in the paper to the product transparency (which has the appealing side effect of allowing coherent change analyses, for which this paper shows an example).
- The paper is a manifest for the consistent flood risk modelling at the national scale. It is a step toward the construction of the digital twin (of UK in this case). Is there any room for local knowledge in this game? I would expect, and the Authors acknowledge it, that locally tailored models may be more accurate than the national one. This is partly due to the calibration with local data but also to the better specification of boundary conditions, including hot-spots, that is possible because of local knowledge. Is there a way to incorporate local knowledge in a "consistent" large scale modelling effort maintaining its consistency? It would be nice to have a couple of lines discussing this point in the concluding section.
Detailed comments:
Line 88: Bloeschl et al. (2019, https://doi.org/10.1038/s41586-019-1495-6) show a significant increase of river flood magnitudes over the UK, specially the northern part,which is one of the clearest hotspots in Europe for that matter.
Line 212: under the RCP8.5 scenario only? Are therefore the different worming levels correspondent to different future times?
Line 216: uncertainty in hydrological modelling is accounted for. How? What are the regionalised "results"? How regionalised?
Line 220: for how many years are the stochastically generated events simulated over the UK? How many events per year are generated on average? (PS. at page 16 I see it is 10000 years)
Table 1: it would be very informative to also stratify the results by river flooding, pluvial flooding, and coastal flooding.
Line 267: labelled 2, in "red"?
Table 2: for "ABI" and "This paper" one could report also other statistics for the annual damages (not only the mean but, for example, the 25% and 75% quantiles, or more) which would show whether the distribution of "observed" annual damages is captured by the model, and that "This paper" is much more informative than NaFRA and CCRA3.
Line 479: for past changes, see e.g., Bertola et al. (2020, https://doi.org/10.5194/hess-24-1805-2020).
Citation: https://doi.org/10.5194/egusphere-2022-829-RC1 -
AC1: 'Reply on RC1', Paul Bates, 01 Nov 2022
Response to reviewer RC1
We are very grateful to the reviewer for their supportive comments and the suggestions for improvement.
The reviewer makes a good point regarding the title and abstract and suggests further highlighting how this work address the lack of transparency in ‘official’ flood maps. This is a problem not only in the UK, but also elsewhere to the best of our knowledge. In a revised version of the paper we will make changes as suggested to address this point and better emphasise this aspect of our contribution.
We also agree that future work should more fully develop the climate change analysis and look at projecting flood risk using ensembles of climate models and different emissions scenarios. This would be a substantial task and one that would need to build on the work presented here. As the current paper is already 23,000 words long (main text plus supplementary information) we think this would need to be as a separate contribution. We will however modify our paper to acknowledge the limitations of the basic climate change assessment undertaken in this proof-of-concept work and discuss what a robust assessment of climate uncertainty might look like.
The comment regarding the role of local knowledge in national scale models is also extremely pertinent. Whilst, national scale models need to be built from available and standardized data sets, there is a need to incorporate local knowledge in a consistent and traceable way that does not lead to local over-fitting for locations where validation data exists. Local over-fitting can give validation studies the appearance of rigour but may mean that this apparent level of skill cannot be generalised to other places.
Instead, we need to find ways to: (i) recover and assemble local data (e.g. on river bathymetry, flood defences and validation data) into consistent national databases and (ii) replicate the decision making of skilled local modellers in automated frameworks. The goal should be to create national models with local knowledge and skill, but, as we have seen in the US, doing this whole process manually is not a scalable solution. For example, the Federal Emergency Management Agency’s national flood mapping program is based on a patchwork of local models however the total cost from its inception in 1969 to 2020 was $10.6 billion, while covering only 33% of the rivers and streams in the country (Association of State Floodplain Managers, 2020).
In the revised paper we will add text to the conclusion to discuss this important issue and thank the referee for drawing our attention to this point.
Detailed comments will be addressed as follows (referee comments in italics):
Line 88: Bloeschl et al. (2019, https://doi.org/10.1038/s41586-019-1495-6) show a significant increase of river flood magnitudes over the UK, specially the northern part,which is one of the clearest hotspots in Europe for that matter.
Thanks. This is a useful reference which we will add to the paper.
Line 212: under the RCP8.5 scenario only? Are therefore the different worming levels correspondent to different future times?
The UKCP 12km regional model simulations we use for the climate projections represent 20-year time slices centred on 2030, 2050 and 2070 under RCP8.5 only. These are the ‘official’ UK climate projections produced by the UK Met Office and therefore an obvious choice and starting point for our work. We interrogate these simulations to find the points when particular specific global warming levels are crossed and then present the loss results based on the changed climate to this date. The different warming levels do therefore represent different future times, but an advantage is that the approach gives a degree of scenario-independence. Whilst the RCP8.5 trajectory is increasingly considered unlikely we only use this scenario to extract results at specific warming levels so are make no judgements about its probability.
It might also be useful to note that, at least until mid-century, the differences over the UK amongst the different emissions scenarios are relatively small. Because we consider near-future projections of flood risk, the impact of climate scenario choice is hopefully minimized. We will add further text to make these points clear.
Line 216: uncertainty in hydrological modelling is accounted for. How? What are the regionalised "results"? How regionalised?
We simply relate change factors to catchment physical characteristics in different UK regions to extrapolate the set of hydrological model outputs to basins that we have not explicitly modelled. We will add further text to make this point clear.
Line 220: for how many years are the stochastically generated events simulated over the UK? How many events per year are generated on average? (PS. at page 16 I see it is 10000 years)
The event rate is determined from an empirical distribution fitted to the annual event counts in the historic gauge data. For each year of the 10,000-year simulation, the number of events to be generated was sampled from this distribution. This resulted in ~343,000 events, ~170,000 of which have a >1 in 5-year magnitude event in at least one catchment (so ~17 per year). This is already detailed in the Supplementary Information on lines 432-439.
Table 1: it would be very informative to also stratify the results by river flooding, pluvial flooding, and coastal flooding.
This would indeed be nice to do, but the data sets we summarize rarely report the information in this way so this unfortunately cannot be done. We can however split our model results by flood hazard type and will add this to the revised manuscript.
Line 267: labelled 2, in "red"?
Good spot, this is a mistake and will be corrected.
Table 2: for "ABI" and "This paper" one could report also other statistics for the annual damages (not only the mean but, for example, the 25% and 75% quantiles, or more) which would show whether the distribution of "observed" annual damages is captured by the model, and that "This paper" is much more informative than NaFRA and CCRA3.
This is a good idea; we will add this.
Line 479: for past changes, see e.g., Bertola et al. (2020, https://doi.org/10.5194/hess-24-1805-2020).
Thanks for drawing our attention to this very useful reference. We will add this to the paper to support the point made here.
References
Association of State Floodplain Managers: Flood Mapping for the Nation: A Cost Analysis for Completing and Maintaining the Nation’s NFIP Flood Map Inventory, Madison, WI, 2020.
Citation: https://doi.org/10.5194/egusphere-2022-829-AC1
-
AC1: 'Reply on RC1', Paul Bates, 01 Nov 2022
-
RC2: 'Comment on egusphere-2022-829', Anonymous Referee #2, 02 Nov 2022
The paper presents a spatially consistent and transparent approach to model flood risk across the UK. The focus of the paper is on comparing the outcomes of a new coupled hydrodynamic – catastrophe model for fluvial, pluvial and coastal flooding with existing approaches, which are in the public domain but insufficiently documented. The authors show that existing approaches used for national flood and climate change risk assessments are likely overestimating the expected annual losses from flooding, due to a number of simplifications, such as when estimating the inundation from coastal flooding.
The paper is an important contribution to flood risk modelling in the UK and the resulting flood hazard and risk maps are the first high skill alternative to the official flood maps provided by UK government agencies and should therefore be published in NHESS. To walk the talk, I would like to encourage the authors to make the flood maps for different return periods and climate scenarios available for the academic community under an open-source license for non-commercial use.
General comments
The paper addresses the really important issue of a lack of alternatives to the official flood hazard and risk maps in the UK, which are spatially inconsistent and not well documented. The paper is well written, clearly structured and critically reflects on several caveats and limitations of the described approach. I have two main points of criticism, which have already been partly addressed by the authors but could be made clearer.
My first point is in regard to the validation of the estimated EAD from the model against insurance claims data from the ABI. The ABI data must be seen as the lower end of any damage estimation due to a number of reasons of which many are mentioned in the manuscript (e.g. data only covers insured residential properties, data on commercial flood damage not included etc.). I agree with the authors assessment that both NaFRA and CCRA3 likely overestimates the EAD, but I would argue that the author’s approach on the other hand is very likely an underestimation of the EAD, which should be discussed in more detail in the manuscript.
My second point is in regard to the climate scenarios. While the loss exceedance curves in Figure 5 and the EAD in Table 3 for different warming scenarios are scientifically interesting, I wonder what we can learn from a scenario that is above current warming levels but with current levels of exposure and vulnerability as we know that such a risk scenario is highly unlikely to occur (it would mean that we stop all human activity in the UK until the 2030s to make for example the 1.8°C scenario presented by the authors a credible one). In my opinion the spatial analysis shown in Figure 6 is more meaningful as it allows to see spatial changes in the hazard under climate change (although I would think it makes more sense to interpret those changes in qualitative terms).
I would not expect the authors to significantly change their results, but provide a bit more context how they would like readers to interpret their results.
Specific comments:
P7 L194ff: One main advantage of the local modelling approach used by the Environment Agency is that they have a good understanding of local flood defences and other protection infrastructure. Can you say something about how your approach compares to that? I have not checked Wing et al. 2019, but in case you have any information on the accuracy of your approach compared to data on local spatial flood defences that would be great.
P8 L199: How where the 10 different return periods selected? Olsen et al. (2015) (https://doi.org/10.3390/w7010255) show that the selection of return periods for the loss exceedance probability curve has a large effect on the EAD. Have you done any sensitivity analysis on how the selection of return periods is influencing your EAD estimates?
P9 L249: You mention the “Fathom model” for the first time in the manuscript. I am assuming this is the name of the model you are presenting in the paper, but would be good to formally introduce the name to avoid confusion.
P9 L257: If possible, it would be great to have the equations for each metric in the text as it makes it easier for the reader to understand how those metrics are calculated.
Figure 2: Would be nice to have an inset showing the location of each flood layer on a GB/UK map
Figure 2 caption: flood hazard maps on the right are shown in red not green
Table 2: Table 2 is an example, but comment is more general: it is sometimes not perfectly clear if values are for England, GB or the UK. As far as I am aware, the NaFRAs are conducted by each of the devolved nations individually. Is the number shown in Table 2, the sum of all NaFRAs or are these values for England only?
Citation: https://doi.org/10.5194/egusphere-2022-829-RC2 -
AC2: 'Reply on RC2', Paul Bates, 07 Nov 2022
We are very grateful to the reviewer for their supportive comments and the suggestions for improvement.
First of all, we’d like to reassure the reviewer that we are indeed “walking the talk” by making the data fully available for non-commercial research use under a standard academic licence. This is already mentioned in the “Data availability” section on lines 523-525 of the main text but we will see if this could be made clearer.
General comments
In terms of the ABI data and whether this is an under-estimate or not, it is worth noting that the ABI data have already been substantially adjusted to deal with many of their limitations. For this we follow the approach given in Penning-Rowsell (2021). This method corrects the data (as far as can reasonably be accomplished at present) for inflation, territorial basis, betterment, taxation, missing pluvial flood losses, underinsurance, ABI market share, missing non-residential losses and changing GDP over time. There is a long section on this in the SI on lines 170-225 and the approach is summarised in the main text on lines 153-163 and 332-337. Of course, this is not to claim that the corrected ABI data are therefore ‘truth’ or that the correction factors determined by Penning-Rowsell are exact, but it does mean that it is not at all clear that the corrected ABI Expected Annual Damage value is an under-estimate. Post-correction, the ABI EAD is just as likely to be too high as too low. The ABI data will of course have error and we do already note in the conclusions on line 488 that “the ABI data need careful handling and adjustment because of the way they have been collected”. However, the referee is correct that we could have said more about their likely uncertainty and will add this discussion to a revised version of the paper.
We agree that including socio-economic as well as climate scenarios would be interesting, however understanding changes in risk due to climate alone is extremely useful in its own right. Moreover, only by controlling for socio-economic change can the impact of particular climate emission policy responses be clearly identified. Demarcating the impact on flood risk of the COP26 commitments and ‘net zero’ ambitions is major outcome of the paper and should have wide impact. The reviewer is however correct that the next step is to look at the interplay between climate and development in modulating future risk, although this is not trivial because of the granularity of socio-economic projections that are required. The IPCC Shared Socio-economic Pathways (SSPs) are at country level and downscaling of these to 1km (i.e., still much coarser that the ~20-25m resolution inundation model) has only just been completed for the UK (see https://uk-scape.ceh.ac.uk/our-science/projects/SPEED/shared-socioeconomic-pathways). These downscaled SSP data will need careful evaluation prior to their use in a flood risk study and some careful methodological development will be needed to bridge the remaining resolution gap. This will be a substantial task and one that realistically will need to be described in a stand-alone paper. Including socio-economic projections in the present (rather overlong) manuscript is probably too much. Instead, we do already include statements about the likely impact of including socio-economic change on lines 390-393 and indicate that this should be looked at in future work.
Specific comments:
P7 L194ff: One main advantage of the local modelling approach used by the Environment Agency is that they have a good understanding of local flood defences and other protection infrastructure. Can you say something about how your approach compares to that? I have not checked Wing et al. 2019, but in case you have any information on the accuracy of your approach compared to data on local spatial flood defences that would be great.
Primarily, we use the exactly same government flood defence database as the UK environmental agencies, namely AIMS (https://www.data.gov.uk/dataset/cc76738e-fc17-49f9-a216-977c61858dda/aims-spatial-flood-defences-inc-standardised-attributes). A reference to this is already included in the bibliography. Most flood defences in our model are based on this ‘official’ view and therefore the majority should be exactly the same between local and national studies. It is also worth noting that local models are only employed in the UK to produce estimates of flood hazard, and these hazard maps are not currently used in the production of national risk estimates. Instead, flood risk is determined separately using large scale simplified inundation models built using national data sets including AIMS for the flood defences (e.g., the NaFRA methodology in England). Official national scale risk estimates thus do not benefit from local knowledge either (at least as far as we can tell from the limited information about these methods that is in the public domain).
The referee is correct however that that EA, SEPA, NRW and DfI local flood hazard modelling studies may possibly supplement the AIMS data with knowledge that is not systematically recorded in an open-source form. Large scale studies, as conducted here, need to work with available published data and their results may diverge from local modelling where this such information has a significant impact. To address some of these limitations we use the method of Wing et al. (2019) to automatically identify flood defences in high resolution terrain data and apply this everywhere such data exists. The Wing et al (2019) paper showed that this method could added important information to official flood defence records and for a test reach of the River Po led to improved model predictions. Importantly, the method can identify structures which impact flood propagation on floodplains, such as causewayed roads and railway embankments, which are not officially classified as flood defences. It is difficult to generalize the River Po findings, but in general we would expect this automatic detection approach to miss some flood defences that local knowledge would pick up, but at the same time it may identify relevant terrain features that might otherwise be overlooked. NaFRA does not include a similar methodology to supplement AIMS flood defence information (as far as we can tell).
We will add some further comments to the paper to discuss this.
P8 L199: How where the 10 different return periods selected? Olsen et al. (2015) (https://doi.org/10.3390/w7010255) show that the selection of return periods for the loss exceedance probability curve has a large effect on the EAD. Have you done any sensitivity analysis on how the selection of return periods is influencing your EAD estimates?
Actually, each loss-exceedance probability curve comes from the catastrophe model part of the workflow so is based on 10,000 years of synthetic flood events with realistic spatial footprints. The return period maps are used to turn each of these spatially variable event intensity footprints into a composite flood depth map for which we can calculate a loss. This gives a distribution of losses with which to form a loss exceedance curve. The Expected Annual Damage is therefore just the integral of the loss exceedance curves. This approach differs significantly from the simpler method of calculating loss for a series of ‘constant in space’ return period maps and using these to compute an EAD as Olsen et al have done which. This is already discussed in the SI on lines 377-382. The choice of return periods in our method will somewhat influence the granularity with which footprints can be generated but the results are not expected to be significantly sensitive to this choice. Accordingly, the return periods were simply chosen to form a spread across the range of typical loss creating flood events and we will add some text to better explain this.
P9 L249: You mention the “Fathom model” for the first time in the manuscript. I am assuming this is the name of the model you are presenting in the paper, but would be good to formally introduce the name to avoid confusion.
“Fathom” was included in error, and we will remove this. The model was produced by Fathom (www.fathom.global) but as this is an academic work we did not want to be accused of advertising.
P9 L257: If possible, it would be great to have the equations for each metric in the text as it makes it easier for the reader to understand how those metrics are calculated.
Of course. We will add these.
Figure 2: Would be nice to have an inset showing the location of each flood layer on a GB/UK map
We will try to do this. The tension here is that the plot is already a whole page figure so where to add an inset without reducing the size of each sub-panel (and hence losing detail) could be a problem. We will experiment with some potential solutions.
Figure 2 caption: flood hazard maps on the right are shown in red not green
Thanks! This is a mistake and will be corrected.
Table 2: Table 2 is an example, but comment is more general: it is sometimes not perfectly clear if values are for England, GB or the UK. As far as I am aware, the NaFRAs are conducted by each of the devolved nations individually. Is the number shown in Table 2, the sum of all NaFRAs or are these values for England only?
NaFRA is the name of the flood risk mapping programme in England only (although it did also cover Wales pre-2013). Wales, Scotland and Northern Ireland have their own programmes with different methodologies and only report number of properties exposed and not financial losses. To create a GB loss we therefore scale the NaFRA result for England using the ratios reported in Penning-Rowsell (2021). These were taken from the emulation methodology used in the 2017 UK Climate Change Risk Assessment (Sayers, 2017). This suggested that England accounts for 79% of flood losses, Scotland 12%, Wales 6% and Northern Ireland 2%.
You are right however that this is not very clear in the main text, and we will correct this.
References
Penning-Rowsell, E. C.: Comparing the scale of modelled and recorded current flood risk: Results from England, Journal of Flood Risk Management, 14, e12685, https://doi.org/10.1111/jfr3.12685, 2021.
Sayers, P.: Projections of future flood risk in the UK, Climate Change Committee, London, UK, 2017.
Wing, O. E. J., Bates, P. D., Neal, J. C., Sampson, C. C., Smith, A. M., Quinn, N., Shustikova, I., Domeneghetti, A., Gilles, D. W., Goska, R., and Krajewski, W. F.: A New Automated Method for Improved Flood Defense Representation in Large-Scale Hydraulic Models, Water Resources Research, 55, 11007–11034, https://doi.org/10.1029/2019WR025957, 2019.
Citation: https://doi.org/10.5194/egusphere-2022-829-AC2
-
AC2: 'Reply on RC2', Paul Bates, 07 Nov 2022
Interactive discussion
Status: closed
-
RC1: 'Referee Comment on egusphere-2022-829', Anonymous Referee #1, 24 Oct 2022
This paper summarizes the impressive work done by Paul Bates and colleagues in flood risk modelling. While the details are left to other more technical papers, this work concentrates on the big picture and answers to the call for peer-reviewable flood risk assessments (in the UK but I would say also elsewhere). Because of this, I believe that the paper fits is appropriate for NHESS. I only have a couple of minor suggestions (and some more detailed comments below):
- Despite the title and abstract focus on climate scenarios, in the paper, a lot of emphasis is given to the transparency and consistency of the hazard and risk maps obtained through this modelling chain as opposed to the "opaque" official maps. The climate change analysis, instead, is not fully developed. The assessment of uncertainty with an envelope of climate models (and for different emission scenarios) would be a standard requirement (and the estimation of additional uncertainties of the hydrologic and hydraulic models would be even better). My suggestion would be to rephrase title and abstract to reflect the weight given in the paper to the product transparency (which has the appealing side effect of allowing coherent change analyses, for which this paper shows an example).
- The paper is a manifest for the consistent flood risk modelling at the national scale. It is a step toward the construction of the digital twin (of UK in this case). Is there any room for local knowledge in this game? I would expect, and the Authors acknowledge it, that locally tailored models may be more accurate than the national one. This is partly due to the calibration with local data but also to the better specification of boundary conditions, including hot-spots, that is possible because of local knowledge. Is there a way to incorporate local knowledge in a "consistent" large scale modelling effort maintaining its consistency? It would be nice to have a couple of lines discussing this point in the concluding section.
Detailed comments:
Line 88: Bloeschl et al. (2019, https://doi.org/10.1038/s41586-019-1495-6) show a significant increase of river flood magnitudes over the UK, specially the northern part,which is one of the clearest hotspots in Europe for that matter.
Line 212: under the RCP8.5 scenario only? Are therefore the different worming levels correspondent to different future times?
Line 216: uncertainty in hydrological modelling is accounted for. How? What are the regionalised "results"? How regionalised?
Line 220: for how many years are the stochastically generated events simulated over the UK? How many events per year are generated on average? (PS. at page 16 I see it is 10000 years)
Table 1: it would be very informative to also stratify the results by river flooding, pluvial flooding, and coastal flooding.
Line 267: labelled 2, in "red"?
Table 2: for "ABI" and "This paper" one could report also other statistics for the annual damages (not only the mean but, for example, the 25% and 75% quantiles, or more) which would show whether the distribution of "observed" annual damages is captured by the model, and that "This paper" is much more informative than NaFRA and CCRA3.
Line 479: for past changes, see e.g., Bertola et al. (2020, https://doi.org/10.5194/hess-24-1805-2020).
Citation: https://doi.org/10.5194/egusphere-2022-829-RC1 -
AC1: 'Reply on RC1', Paul Bates, 01 Nov 2022
Response to reviewer RC1
We are very grateful to the reviewer for their supportive comments and the suggestions for improvement.
The reviewer makes a good point regarding the title and abstract and suggests further highlighting how this work address the lack of transparency in ‘official’ flood maps. This is a problem not only in the UK, but also elsewhere to the best of our knowledge. In a revised version of the paper we will make changes as suggested to address this point and better emphasise this aspect of our contribution.
We also agree that future work should more fully develop the climate change analysis and look at projecting flood risk using ensembles of climate models and different emissions scenarios. This would be a substantial task and one that would need to build on the work presented here. As the current paper is already 23,000 words long (main text plus supplementary information) we think this would need to be as a separate contribution. We will however modify our paper to acknowledge the limitations of the basic climate change assessment undertaken in this proof-of-concept work and discuss what a robust assessment of climate uncertainty might look like.
The comment regarding the role of local knowledge in national scale models is also extremely pertinent. Whilst, national scale models need to be built from available and standardized data sets, there is a need to incorporate local knowledge in a consistent and traceable way that does not lead to local over-fitting for locations where validation data exists. Local over-fitting can give validation studies the appearance of rigour but may mean that this apparent level of skill cannot be generalised to other places.
Instead, we need to find ways to: (i) recover and assemble local data (e.g. on river bathymetry, flood defences and validation data) into consistent national databases and (ii) replicate the decision making of skilled local modellers in automated frameworks. The goal should be to create national models with local knowledge and skill, but, as we have seen in the US, doing this whole process manually is not a scalable solution. For example, the Federal Emergency Management Agency’s national flood mapping program is based on a patchwork of local models however the total cost from its inception in 1969 to 2020 was $10.6 billion, while covering only 33% of the rivers and streams in the country (Association of State Floodplain Managers, 2020).
In the revised paper we will add text to the conclusion to discuss this important issue and thank the referee for drawing our attention to this point.
Detailed comments will be addressed as follows (referee comments in italics):
Line 88: Bloeschl et al. (2019, https://doi.org/10.1038/s41586-019-1495-6) show a significant increase of river flood magnitudes over the UK, specially the northern part,which is one of the clearest hotspots in Europe for that matter.
Thanks. This is a useful reference which we will add to the paper.
Line 212: under the RCP8.5 scenario only? Are therefore the different worming levels correspondent to different future times?
The UKCP 12km regional model simulations we use for the climate projections represent 20-year time slices centred on 2030, 2050 and 2070 under RCP8.5 only. These are the ‘official’ UK climate projections produced by the UK Met Office and therefore an obvious choice and starting point for our work. We interrogate these simulations to find the points when particular specific global warming levels are crossed and then present the loss results based on the changed climate to this date. The different warming levels do therefore represent different future times, but an advantage is that the approach gives a degree of scenario-independence. Whilst the RCP8.5 trajectory is increasingly considered unlikely we only use this scenario to extract results at specific warming levels so are make no judgements about its probability.
It might also be useful to note that, at least until mid-century, the differences over the UK amongst the different emissions scenarios are relatively small. Because we consider near-future projections of flood risk, the impact of climate scenario choice is hopefully minimized. We will add further text to make these points clear.
Line 216: uncertainty in hydrological modelling is accounted for. How? What are the regionalised "results"? How regionalised?
We simply relate change factors to catchment physical characteristics in different UK regions to extrapolate the set of hydrological model outputs to basins that we have not explicitly modelled. We will add further text to make this point clear.
Line 220: for how many years are the stochastically generated events simulated over the UK? How many events per year are generated on average? (PS. at page 16 I see it is 10000 years)
The event rate is determined from an empirical distribution fitted to the annual event counts in the historic gauge data. For each year of the 10,000-year simulation, the number of events to be generated was sampled from this distribution. This resulted in ~343,000 events, ~170,000 of which have a >1 in 5-year magnitude event in at least one catchment (so ~17 per year). This is already detailed in the Supplementary Information on lines 432-439.
Table 1: it would be very informative to also stratify the results by river flooding, pluvial flooding, and coastal flooding.
This would indeed be nice to do, but the data sets we summarize rarely report the information in this way so this unfortunately cannot be done. We can however split our model results by flood hazard type and will add this to the revised manuscript.
Line 267: labelled 2, in "red"?
Good spot, this is a mistake and will be corrected.
Table 2: for "ABI" and "This paper" one could report also other statistics for the annual damages (not only the mean but, for example, the 25% and 75% quantiles, or more) which would show whether the distribution of "observed" annual damages is captured by the model, and that "This paper" is much more informative than NaFRA and CCRA3.
This is a good idea; we will add this.
Line 479: for past changes, see e.g., Bertola et al. (2020, https://doi.org/10.5194/hess-24-1805-2020).
Thanks for drawing our attention to this very useful reference. We will add this to the paper to support the point made here.
References
Association of State Floodplain Managers: Flood Mapping for the Nation: A Cost Analysis for Completing and Maintaining the Nation’s NFIP Flood Map Inventory, Madison, WI, 2020.
Citation: https://doi.org/10.5194/egusphere-2022-829-AC1
-
AC1: 'Reply on RC1', Paul Bates, 01 Nov 2022
-
RC2: 'Comment on egusphere-2022-829', Anonymous Referee #2, 02 Nov 2022
The paper presents a spatially consistent and transparent approach to model flood risk across the UK. The focus of the paper is on comparing the outcomes of a new coupled hydrodynamic – catastrophe model for fluvial, pluvial and coastal flooding with existing approaches, which are in the public domain but insufficiently documented. The authors show that existing approaches used for national flood and climate change risk assessments are likely overestimating the expected annual losses from flooding, due to a number of simplifications, such as when estimating the inundation from coastal flooding.
The paper is an important contribution to flood risk modelling in the UK and the resulting flood hazard and risk maps are the first high skill alternative to the official flood maps provided by UK government agencies and should therefore be published in NHESS. To walk the talk, I would like to encourage the authors to make the flood maps for different return periods and climate scenarios available for the academic community under an open-source license for non-commercial use.
General comments
The paper addresses the really important issue of a lack of alternatives to the official flood hazard and risk maps in the UK, which are spatially inconsistent and not well documented. The paper is well written, clearly structured and critically reflects on several caveats and limitations of the described approach. I have two main points of criticism, which have already been partly addressed by the authors but could be made clearer.
My first point is in regard to the validation of the estimated EAD from the model against insurance claims data from the ABI. The ABI data must be seen as the lower end of any damage estimation due to a number of reasons of which many are mentioned in the manuscript (e.g. data only covers insured residential properties, data on commercial flood damage not included etc.). I agree with the authors assessment that both NaFRA and CCRA3 likely overestimates the EAD, but I would argue that the author’s approach on the other hand is very likely an underestimation of the EAD, which should be discussed in more detail in the manuscript.
My second point is in regard to the climate scenarios. While the loss exceedance curves in Figure 5 and the EAD in Table 3 for different warming scenarios are scientifically interesting, I wonder what we can learn from a scenario that is above current warming levels but with current levels of exposure and vulnerability as we know that such a risk scenario is highly unlikely to occur (it would mean that we stop all human activity in the UK until the 2030s to make for example the 1.8°C scenario presented by the authors a credible one). In my opinion the spatial analysis shown in Figure 6 is more meaningful as it allows to see spatial changes in the hazard under climate change (although I would think it makes more sense to interpret those changes in qualitative terms).
I would not expect the authors to significantly change their results, but provide a bit more context how they would like readers to interpret their results.
Specific comments:
P7 L194ff: One main advantage of the local modelling approach used by the Environment Agency is that they have a good understanding of local flood defences and other protection infrastructure. Can you say something about how your approach compares to that? I have not checked Wing et al. 2019, but in case you have any information on the accuracy of your approach compared to data on local spatial flood defences that would be great.
P8 L199: How where the 10 different return periods selected? Olsen et al. (2015) (https://doi.org/10.3390/w7010255) show that the selection of return periods for the loss exceedance probability curve has a large effect on the EAD. Have you done any sensitivity analysis on how the selection of return periods is influencing your EAD estimates?
P9 L249: You mention the “Fathom model” for the first time in the manuscript. I am assuming this is the name of the model you are presenting in the paper, but would be good to formally introduce the name to avoid confusion.
P9 L257: If possible, it would be great to have the equations for each metric in the text as it makes it easier for the reader to understand how those metrics are calculated.
Figure 2: Would be nice to have an inset showing the location of each flood layer on a GB/UK map
Figure 2 caption: flood hazard maps on the right are shown in red not green
Table 2: Table 2 is an example, but comment is more general: it is sometimes not perfectly clear if values are for England, GB or the UK. As far as I am aware, the NaFRAs are conducted by each of the devolved nations individually. Is the number shown in Table 2, the sum of all NaFRAs or are these values for England only?
Citation: https://doi.org/10.5194/egusphere-2022-829-RC2 -
AC2: 'Reply on RC2', Paul Bates, 07 Nov 2022
We are very grateful to the reviewer for their supportive comments and the suggestions for improvement.
First of all, we’d like to reassure the reviewer that we are indeed “walking the talk” by making the data fully available for non-commercial research use under a standard academic licence. This is already mentioned in the “Data availability” section on lines 523-525 of the main text but we will see if this could be made clearer.
General comments
In terms of the ABI data and whether this is an under-estimate or not, it is worth noting that the ABI data have already been substantially adjusted to deal with many of their limitations. For this we follow the approach given in Penning-Rowsell (2021). This method corrects the data (as far as can reasonably be accomplished at present) for inflation, territorial basis, betterment, taxation, missing pluvial flood losses, underinsurance, ABI market share, missing non-residential losses and changing GDP over time. There is a long section on this in the SI on lines 170-225 and the approach is summarised in the main text on lines 153-163 and 332-337. Of course, this is not to claim that the corrected ABI data are therefore ‘truth’ or that the correction factors determined by Penning-Rowsell are exact, but it does mean that it is not at all clear that the corrected ABI Expected Annual Damage value is an under-estimate. Post-correction, the ABI EAD is just as likely to be too high as too low. The ABI data will of course have error and we do already note in the conclusions on line 488 that “the ABI data need careful handling and adjustment because of the way they have been collected”. However, the referee is correct that we could have said more about their likely uncertainty and will add this discussion to a revised version of the paper.
We agree that including socio-economic as well as climate scenarios would be interesting, however understanding changes in risk due to climate alone is extremely useful in its own right. Moreover, only by controlling for socio-economic change can the impact of particular climate emission policy responses be clearly identified. Demarcating the impact on flood risk of the COP26 commitments and ‘net zero’ ambitions is major outcome of the paper and should have wide impact. The reviewer is however correct that the next step is to look at the interplay between climate and development in modulating future risk, although this is not trivial because of the granularity of socio-economic projections that are required. The IPCC Shared Socio-economic Pathways (SSPs) are at country level and downscaling of these to 1km (i.e., still much coarser that the ~20-25m resolution inundation model) has only just been completed for the UK (see https://uk-scape.ceh.ac.uk/our-science/projects/SPEED/shared-socioeconomic-pathways). These downscaled SSP data will need careful evaluation prior to their use in a flood risk study and some careful methodological development will be needed to bridge the remaining resolution gap. This will be a substantial task and one that realistically will need to be described in a stand-alone paper. Including socio-economic projections in the present (rather overlong) manuscript is probably too much. Instead, we do already include statements about the likely impact of including socio-economic change on lines 390-393 and indicate that this should be looked at in future work.
Specific comments:
P7 L194ff: One main advantage of the local modelling approach used by the Environment Agency is that they have a good understanding of local flood defences and other protection infrastructure. Can you say something about how your approach compares to that? I have not checked Wing et al. 2019, but in case you have any information on the accuracy of your approach compared to data on local spatial flood defences that would be great.
Primarily, we use the exactly same government flood defence database as the UK environmental agencies, namely AIMS (https://www.data.gov.uk/dataset/cc76738e-fc17-49f9-a216-977c61858dda/aims-spatial-flood-defences-inc-standardised-attributes). A reference to this is already included in the bibliography. Most flood defences in our model are based on this ‘official’ view and therefore the majority should be exactly the same between local and national studies. It is also worth noting that local models are only employed in the UK to produce estimates of flood hazard, and these hazard maps are not currently used in the production of national risk estimates. Instead, flood risk is determined separately using large scale simplified inundation models built using national data sets including AIMS for the flood defences (e.g., the NaFRA methodology in England). Official national scale risk estimates thus do not benefit from local knowledge either (at least as far as we can tell from the limited information about these methods that is in the public domain).
The referee is correct however that that EA, SEPA, NRW and DfI local flood hazard modelling studies may possibly supplement the AIMS data with knowledge that is not systematically recorded in an open-source form. Large scale studies, as conducted here, need to work with available published data and their results may diverge from local modelling where this such information has a significant impact. To address some of these limitations we use the method of Wing et al. (2019) to automatically identify flood defences in high resolution terrain data and apply this everywhere such data exists. The Wing et al (2019) paper showed that this method could added important information to official flood defence records and for a test reach of the River Po led to improved model predictions. Importantly, the method can identify structures which impact flood propagation on floodplains, such as causewayed roads and railway embankments, which are not officially classified as flood defences. It is difficult to generalize the River Po findings, but in general we would expect this automatic detection approach to miss some flood defences that local knowledge would pick up, but at the same time it may identify relevant terrain features that might otherwise be overlooked. NaFRA does not include a similar methodology to supplement AIMS flood defence information (as far as we can tell).
We will add some further comments to the paper to discuss this.
P8 L199: How where the 10 different return periods selected? Olsen et al. (2015) (https://doi.org/10.3390/w7010255) show that the selection of return periods for the loss exceedance probability curve has a large effect on the EAD. Have you done any sensitivity analysis on how the selection of return periods is influencing your EAD estimates?
Actually, each loss-exceedance probability curve comes from the catastrophe model part of the workflow so is based on 10,000 years of synthetic flood events with realistic spatial footprints. The return period maps are used to turn each of these spatially variable event intensity footprints into a composite flood depth map for which we can calculate a loss. This gives a distribution of losses with which to form a loss exceedance curve. The Expected Annual Damage is therefore just the integral of the loss exceedance curves. This approach differs significantly from the simpler method of calculating loss for a series of ‘constant in space’ return period maps and using these to compute an EAD as Olsen et al have done which. This is already discussed in the SI on lines 377-382. The choice of return periods in our method will somewhat influence the granularity with which footprints can be generated but the results are not expected to be significantly sensitive to this choice. Accordingly, the return periods were simply chosen to form a spread across the range of typical loss creating flood events and we will add some text to better explain this.
P9 L249: You mention the “Fathom model” for the first time in the manuscript. I am assuming this is the name of the model you are presenting in the paper, but would be good to formally introduce the name to avoid confusion.
“Fathom” was included in error, and we will remove this. The model was produced by Fathom (www.fathom.global) but as this is an academic work we did not want to be accused of advertising.
P9 L257: If possible, it would be great to have the equations for each metric in the text as it makes it easier for the reader to understand how those metrics are calculated.
Of course. We will add these.
Figure 2: Would be nice to have an inset showing the location of each flood layer on a GB/UK map
We will try to do this. The tension here is that the plot is already a whole page figure so where to add an inset without reducing the size of each sub-panel (and hence losing detail) could be a problem. We will experiment with some potential solutions.
Figure 2 caption: flood hazard maps on the right are shown in red not green
Thanks! This is a mistake and will be corrected.
Table 2: Table 2 is an example, but comment is more general: it is sometimes not perfectly clear if values are for England, GB or the UK. As far as I am aware, the NaFRAs are conducted by each of the devolved nations individually. Is the number shown in Table 2, the sum of all NaFRAs or are these values for England only?
NaFRA is the name of the flood risk mapping programme in England only (although it did also cover Wales pre-2013). Wales, Scotland and Northern Ireland have their own programmes with different methodologies and only report number of properties exposed and not financial losses. To create a GB loss we therefore scale the NaFRA result for England using the ratios reported in Penning-Rowsell (2021). These were taken from the emulation methodology used in the 2017 UK Climate Change Risk Assessment (Sayers, 2017). This suggested that England accounts for 79% of flood losses, Scotland 12%, Wales 6% and Northern Ireland 2%.
You are right however that this is not very clear in the main text, and we will correct this.
References
Penning-Rowsell, E. C.: Comparing the scale of modelled and recorded current flood risk: Results from England, Journal of Flood Risk Management, 14, e12685, https://doi.org/10.1111/jfr3.12685, 2021.
Sayers, P.: Projections of future flood risk in the UK, Climate Change Committee, London, UK, 2017.
Wing, O. E. J., Bates, P. D., Neal, J. C., Sampson, C. C., Smith, A. M., Quinn, N., Shustikova, I., Domeneghetti, A., Gilles, D. W., Goska, R., and Krajewski, W. F.: A New Automated Method for Improved Flood Defense Representation in Large-Scale Hydraulic Models, Water Resources Research, 55, 11007–11034, https://doi.org/10.1029/2019WR025957, 2019.
Citation: https://doi.org/10.5194/egusphere-2022-829-AC2
-
AC2: 'Reply on RC2', Paul Bates, 07 Nov 2022
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 661 | 341 | 17 | 1,019 | 82 | 3 | 2 |
- HTML: 661
- PDF: 341
- XML: 17
- Total: 1,019
- Supplement: 82
- BibTeX: 3
- EndNote: 2
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
James Savage
Oliver Wing
Niall Quinn
Christopher Sampson
Jeffrey Neal
Andrew Smith
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(3582 KB) - Metadata XML
-
Supplement
(5005 KB) - BibTeX
- EndNote
- Final revised paper