the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Technical Note: Resolution Enhancement of Flood Inundation Grids
Guy Schumann
Heiko Apel
Heidi Kreibich
Bruno Merz
Abstract. High-resolution flood maps are needed for more effective flood risk assessment and management. Producing these directly with hydrodynamic models is slow and computationally prohibitive at large scales. Here we demonstrate a new algorithm for post-processing low-resolution inundation layers by using high-resolution terrain models to disaggregate or downscale. The new algorithm is roughly eight times faster than current state-of-the-art algorithms, shows a slight improvement in accuracy when evaluated against observations of a recent flood, and is open source. The algorithm developed here can be applied in conjunction with a low-resolution hydrodynamic model and a high-resolution DEM to rapidly produce high-resolution inundation maps. For example, in our case study with a river reach of 20 km, the proposed algorithm generated a 4 m resolution inundation map from 32 m hydrodynamic model outputs in 33 seconds, compared to a 4 m hydrodynamic model runtime of 34 minutes. This 60-fold improvement in runtime is associated with a 25 % increase in RMSE when compared against the 4 m hydrodynamic model results and observations of a recent flood. Substituting downscaling into flood risk model chains for high-resolution modelling has the potential to drastically improve the efficiency of inundation map production and increase the lead time of impact-based forecasts, helping more at-risk communities prepare for and mitigate flood damages.
- Preprint
(2507 KB) - Metadata XML
-
Supplement
(8578 KB) - BibTeX
- EndNote
Seth Bryant et al.
Status: open (until 11 Oct 2023)
-
RC1: 'Comment on hess-2023-156', Anonymous Referee #1, 07 Sep 2023
reply
The technical note introduces a new algorithm for upscaling hydrodynamic flood inundation predictions using a high-resolution DEM. The new approach is compared to several simple upscaling approaches for one flood event in Germany. Overall, the manuscript is well-written and the new algorithm is a useful advance in flood inundation predictions. The methodology, analysis, and discussions are sound. As the authors discussed, there is nearly no improvement in the algorithm performance matrices compared to the Schumann algorithm, though there is a considerable improvement in run-time. In my view, the strongest result, and one that best showcases the superiority of the new algorithm, is in Figure 6. It shows that the new algorithm can much better represent small-scale inundation extent. This is very important and, clearly, cannot be well captured by standard evaluation metrics. I encourage the author to emphasize this result (including in the abstract and conclusions) and provide additional examples (using the same flood case study). Do not shy away from qualitative comparison. Adding building impact may also help showcase the improved capability at small scales. The new algorithm’s description needs to be extended.
Specific Comments:
Line 23 and later: Positive and negative bias – I find this confusing. Consider using over and under estimation or some other descriptor.
Line 29: ‘similar accuracy when upscaling’ – accuracy of DEM or flood map?
Line 67: ‘because of the resolution of the underlying global datasets.’ – need clarification
Title of section 2: ‘Resample Case’ add ‘Framework’
Section 3.1 – More detailed description of the algorithm is required. Provide a short explanation of the utility/reason for each step.
Figure S5 – show the river’s flow direction and the location of the HWM points.
Line 163 – change ‘Finally’ to ‘Additionally’ or ‘In addition,’ etc.
Line 177 – what was the performance of s2?
Figure 5 – ‘rvalue’ – is this R or R2? R2 should be reported. Explain the observations with values close to 0.
Citation: https://doi.org/10.5194/hess-2023-156-RC1 -
RC2: 'Comment on hess-2023-156', Anonymous Referee #2, 12 Sep 2023
reply
Resolution enhancement of flood inundation grids
Bryant, S., et al.
The manuscript proposes a method to obtain raster inundation maps with a certain resolution by running hydrodynamic models at lower resolution and then interpolating the obtained values, instead of running hydrodynamic models at the higher resolution. The motivation for this is a significant reduction of total computational time, yet avoiding a significant loss of accuracy. The method thus seems of potential interest for HESS readers.
I am writing this review before reading what another reviewer has written, to ensure independency of the reports. Sorry if I will repeat anything.
My first comment is that this manuscript depends too much on supplemental material. Figure S1 presents the synthetic case and a table used to define the possible cases one may encounter when downscaling, that is the graphical support to a conceptualization. Figure S2 and Table S1 present the case-study. Even though the supplemental material is available, I think that crucial information needs to be in the main manuscript. I do not know why the authors opted for this way, probably to fit some limitation length, but the result is unsatisfactory because no way the manuscript can be a stand-alone product.
Second, I think that a conceptual discussion to introduce the challenges related to the work needs to be added. Presently, it is left to figure S1 and eq. (1) that introduce the easy and the “partial” cases. I think that more emphasis should be given to the removal of cells that would be wet after the simple interpolation (the “isolated cell filter” is not well described) and to the extrapolation of inundation (that is another crucial step to correct the bare result of the interpolation).
Third, the methods are not well described. In the description of the CostGrow, what is a cost and how is it minimized? How is an isolated clump recognized? And so on. In addition, figures do not help. For example, values in the panels of fig 1 are not described, not even defined. I guess that panels (abc) contain elevations while panel (d) contains codes that will mean something (clump name?). Furthermore, overlapping wse and DEM in panel a does not help. It would be better to have panels up to (e) to see everything (I know that the DEM is also in fig 2a but we need it here). Also it would be helpful to make examples of what happens in some cells. Commenting some output would be important for understanding how the method works.
Fourth, I think that it would be important to stress that the method will refine and correct the inundation map up to what can be seen by the refined DEM. In many cases, particularly in urban contexts, the pattern of the inundation is determined by narrow objects (typically, narrow bank walls) that are invisible even in DEMs with a resolution of 1 m. A correct representation of the effects of these structures on the inundation dynamics requires a specific treatment in the hydrodynamic model. In cases where thin structures matter, I would not recommend using a downscaling algorithm. I trust that this could be irrelevant for the case-study presented here, but the authors may add their view on this or similar issues.
Fifth, I am somehow puzzled by figure S5, where the concentration of red at the right and blue at the left of the picture would mean that changing the resolution of the hydrodynamic model changes significantly the mean slope of the free surface along the reach, in turn indicating that at least one of the models will not represent correctly the reality.
After these comments that I consider most important, several detailed ones follow (with line number indicated).
27: In my experience there are two possible issues for channel misrepresentation: (1) a real thalweg being lower than how it appears in the DEM, and again (2) narrow bank walls being absent in the DEM. This could be mentioned.
52: it is unclear which hypothesis the authors are referring to; some explanation would be needed.
69: Unclear. To me, "is used in practice" would mean that it is a (relatively) standard tool used by engineers. But this contradicts the above, where it was written that only one published study exists. Clarification is needed.
74: This paper in under review. More information is needed here to understand the (synthetic?) case. This further supports a request to move material from the supplemental to the manuscript.
86: May be not just a matter of using an exceedance mask. in case the water is confined by an embankment and the coarse-mesh model returns a wide inundated area, the exceedance mask will dry the cells corresponding to the embankment but not the external ones. In this case, it is a matter of drying cells disconnected from the inundation area after the application of the exceedance mask. To me this is related to the second major comment and a need to give more emphasis to the removal of isolated flooded regions that will come with figure 1.
98-109: already mentioned above that the description of the method would need to be richer.
114: I am really shocked reading that the case-study flood had a return period ranging from 3 thousand to 60 thousand years. Without an explanation this statement just sounds crazy. I would suggest to provide some details to mention where these huge and very different values come from, or to remove the statement that is not crucial for the assessment of the algorithm’s performance.
116: again, are we really putting in a supplemental these key pieces of information?
118: "calibrated to the oberved inundation" is unclear. Which is the calibration parameter, and which the criterion? Later it will be mentioned that the calibration is aimed at reproducing the extent of the inundation (not the depths) by changing two roughness coefficients (river and urban), but the information is needed here.
Figure 1: why a subscript B? Before they were i and j.
125: this sounds strange. Avoiding a presumably better treatment because it is affected by resolution is kind of weird in a work aimed at resolution enhancement and providing (line 1) “high resolution flood maps". Sounds like this intends to be a preliminary study and validation in "easy" cases.
131: here and elsewhere, the measuring units for the Manning coefficient should be added.
133: weird writing, please rephrase.
137: 0.867 > 0.175, ok; 0.089 < 0.133, strange (considering how the n should change with grid size). Can a comment be added in this respect?
141: Here a discussion of the calibration was presented only for the hydrodynamic model. What about cost-grow, is it parameterless?
146: the order of presentation is different from the list in the above line. I would suggest to invert.
152: Figure S1 is a sketch with 2x2 coarse cells, while here it looks like we have 2x3 coarse cells. So, referring to figure S1 is unclear.
155: this is unclear. What is the horizontal plane for?
164: this statement may induce in a reader a doubt that the result presented here comes from the authors’ interpretation of a previous model, and may thus suffer from misinterpretation. In my understanding, this should not be the case since a Schumann is also in the authorship of the present paper, which may be made explicit.
207: here we find again a detail on model calibration that should have come much earlier.
The second part of the manuscript, showing the results, sounded more straightforward than the first. Maybe, comments on it may be stimulated by a revision of the first part clarifying the methods.
I hope these comments will help produce a clearer and more impactful manuscript.
Citation: https://doi.org/10.5194/hess-2023-156-RC2
Seth Bryant et al.
Seth Bryant et al.
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 350 | 97 | 14 | 461 | 23 | 4 | 6 |
- HTML: 350
- PDF: 97
- XML: 14
- Total: 461
- Supplement: 23
- BibTeX: 4
- EndNote: 6
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1