Summary:

This article proposes a method that combines methodological improvements to the HAND method with a simple 1D hydraulic model in order to generate fluvial flood inundation maps. The proposed model is shown to be competitive when compared to a baseline model [2D HEC RAS] for a case study in Ontario, Canada. The computational efficiency and model performance of the proposed approach is highlighted to demonstrate the novelty of the work. Overall, the proposed methods represent simple but effective improvements over commonly used methods. The model and the approach described in this “Development and technical paper” would be of interest to readers of “Geoscientific Model Development” – both in Canada and internationally due to its broad applicability. I recommended accepting this manuscript for publication since it satisfies the criteria:

+ Significance: a new approach that improves on a widely used method is highly important

+ Scientific quality & reproducibility: the methods are described fully, with enough information to replicate the work, and can be applied to a wide range of locations results in significant scientific results

+ Presentation quality: the paper is well-written (minor comments are included below, as well as suggestion to improve the figures).

The main limitation of the proposed manuscript is the analysis of model performance which has the potential to be expanded. Currently, only two (similar) performance metrics are used. Doing so has the potential to generate more discussion points, and will better highlight the methodological improvements introduced in this model/manuscript. Details of this are provided below. 

I hope these specific and minor comments are helpful to the authors

Specific Comments:

Abstract

L17: “10,000 times less” Is it possible to provide a non-relative measure of this computational speed, I think this would make it clearer. For example, is the difference between 1 s or 10,000 s, or 1 day or 10,000 days? – whilst both have a 10,000 factor difference, the implications are different. So perhaps in brackets, the average computational time improvement, or perhaps another metric to measure computational performance (computational complexity, bit complexity, etc.)

Introduction

L82: “hydraulic properties for 1D calculations” – for the reader’s benefit, please include some specific examples of hydraulic properties, perhaps in parentheses, to aid understanding of the proposed method

L87: first mention of the Blackbird software package in the main text – it would be nice to identify that the proposed method is indeed the Blackbird program/software here.

Methods

L120 – 150: I am not sure if this is a common method (bullet points) of listing all the steps of the workflow. It can easily be converted to paragraph form to improve readability and flow. 

L130: Are things like “boundary conditions” settings within Blackbird? A diagram may be useful here

L136: Some more info on calibration would be useful here (e.g., computational expense/scaling, number of iterations, loss function ,etc.). How sensitive are the various parameters? How is accuracy with and without calibration? Some of this is presented later, so perhaps a note for the reader would be helpful

L163: The program here is Blackbird, I assume?

L188: Is this necessarily a bad thing: “This is a known issue with the processing behind the HAND method (Aristizabal et al., 2023). In any case, the generation of the HAND raster itself requires manipulating the DEM to determine flow pathways.”

L291: I do not agree with his justification of the selected performance metrics. While depth error does contain more information than a binary projection of error, neither of the selected metrics indicate whether the depths are over- or under-predicted. In other words, the model is unable to distinguish whether the predicted flood extent is over- or underestimated, which is critical for characterizing model performance, as the latter is much worse in the context of flood damage. Additionally, I don’t think that the two selected metrics are very complementary, as they are just two variations of the mean error. So some more thought on the selection of metrics is necessary. 

L296: The first tests are probably better described as a “proof of concept” rather than a “case study”

Results

L424: Figure 5 and the related analysis can be strengthened. The figure shows 36 values, half of which are from the uncalibrated models, and both error metrics are quite similar [see earlier comment]. I wonder if there would be a way to enhance it to better distinguish the improvements offered by Blackbird, such as using bootstrapping or some uncertainty propagation to generate error bars? 

Discussion

L430: following on from the comment in the Abstract, “which is approximately ×10,000 faster” it would be easier for a reader if all model types and cases were summarised in a table with the associated run times listed next to them to ease comparison. 

L439: “regulatory maps” – for an international audience this term needs to be defined. Regulatory maps are not often used for real-time decisions, but rather planning. I don’t disagree with the main point here, but the set of regulatory maps is known to be not useful for this type of scenario. The proposed method provides several advantages for flood mitigation, and these can be discussed without reference to regulatory maps too. Agencies will have different types of models (2D) for more detailed flood maps in sensitive areas rather than relying on regulatory maps exclusively.

L456: “requiring manual intervention”: while this is true, much of the model development process can also be automated. The automation property of the proposed method is not unique. And again, there are many positive aspects of the proposed approach, but perhaps this argument can be rethought or reframed. 

L550: I had some confusion during my review to separate the GASS method from the Blackbird software. It would be helpful if this could be defined early in the manuscript. I assume the GASS method may be implemented in any modelling environment, and that the version presented here is hosted within Blackbird? 

Minor comments:

L73: perhaps add a comma after “to” to improve readability: “since each domain or catchment area the method is applied to is computed independently.”

L105: Confusing wording, should just state ‘builds on existing strategies’ or similar.

L110: a word missing after “bathymetric”? information?

L173: “0” missing after “Section 2.4”

Figure 1: 1b) – would it be better to illustrate this as a raster, since I’m assuming this portion is grid-based?

Figure 2 caption: “southwest” is used but a compass rose is not present in the figure. Perhaps say “bottom left corner”, instead? Secondly, it would easier if the berm is outlined in the figure as well. The flood outline is hard to see as well

Page 8: Equation 1 is presented, however, the other equations in the paper thus far have not been labelled similarly. 

L301: add “Canada”

L351: Could use “10-year return period” etc. instead

Figure 3: this map is oddly low quality (markers to Home Hardware etc.) are not needed. It can be easily replaced with a more neutral base map with a clear identification of the reach and the HEC RAS cross section extents.

L383: good practice to include the test statistic results in parentheses

L386: “not accumulating with station” homoscedastic may be an easy way to describe this phenomenon

L411: “Different model formulations leading to different calibrated Manning’s roughness values” this sentence seems incomplete, or should “leading” be “leads”?

Figure 5: I would prefer if subplot C had the same y axis limits as subplot A

Review summary

The manuscript presents a number of novel developments to application of the HAND algorithm in combination with a steady-state hydrodynamic model for fluvial flood mapping. These novel developments include this very combination with a 1D model that consider backwater effects and the so-called dynamic HAND algorithm to determine flood outlines, which preserves the floodplain connectivity and results in more realistic flood areas compared to the standard HAND approach. The manuscript is well-written and clearly structured. The proposed improvements are substantial and are demonstrated to improve the previous procedures and provide more accurate flood mapping results. In overall, this is very interesting model development and a well-presented manuscript which may benefit from a few suggestions listed below.

It would be great to improve the clarity at some points in the method description.

L173: What are the reach-integrated hydraulic properties you are talking about? Is this conveyance, i.e., flow cross-sectional area and hydraulic radius? Would be good to mention this here.

The core of the proposed DHAND algorithm is described in L197-200 and is not quite clear to me. I would be great to provide a more detailed description, maybe with a visual. This a key novel development, but appears very concise, at least to me. I understand that the terrain is discretized to k layers. Why conditioning the DEM in the specific layer k with a breaching algorithm and all other cells with a fill algorithm results in a desired connectivity? What do you mean by “all other cells”?

L219-221: please, integrate the visual for slices (b) into Figure 1c. this would helpful. You may consider to integrate Figure 1d showing the raster-based discretized view, where each raster cell represents a slice.

L254: This is not clear to me, how a cross-sectional area becomes a flooded volume. Here again a raster-based visual (Figure 1d) would be helpful to show these peculiarities.

L291-294: The authors argue that they omit contingency metrices and indices derived from them in favor of depth. I cannot really follow this argument. Particularly, in flat areas errors in depth might be relatively small, but discrepancies in predicting inundation areas are substantial. I would therefore suggest to include these metrices in evaluation and also showcase the performance in terrains with steep and flat floodplains. Here, I also support the first reviewer who suggest more in-depth evaluation of the algorithm performance. Maybe it is not needed to compute these metrices for simplified synthetic cases, but for a couple real-world examples as suggested above.

L316ff: Why do you list all the test SB1-SB5 if you do not show their results. Please, remove and focus on the real-world examples. As suggested above, I recommend to include at least two different cases with different terrain slopes and resulting inundation area extent. The presented Waldemar example seem to rather confined valley – an easy task for a flood mapping algorithm.

Minor comments:

L42: check citation „e.g.“ also thereafter

L55: Recent near-real time applications (in a hindcast mode) of the GPU-accelerated 2D models seem to overcome this barrier (Apel et al., 2022, Najafi et al., 2023).

L60ff: here also a set of hydrogeomorphic mapping methods can be mentioned (e.g., Tavares da Costa et al., 2019).

L137-150: it seems to me that this is just a repletion of the steps summarized in L120-136, but not structured according to preprocessing, simulation and postprocessing. Please, clarify.

L160: alpha – please, note here that it is explained in L250ff in more details.

L162: correct the head loss notation (h_e). In Equation it is given as delta_h_j.

L419: HEC-RAS

Figure 3 – I would rather appreciate a figure showing elevation as a background

L439-440: remove the word “event”. Typically, flood hazard maps do not correspond to a specific flood event, but represent a mosaic of maximum flood extents corresponding to at-point 100-year flood estimates at various locations. There is a subtle but important difference!

I particularly appreciate a frank and transparent discussion on challenges and limitations, especially L503-516!

References:

Apel, H., Vorogushyn, S., & Merz, B. (2022). Brief communication: Impact forecasting could substantially improve the emergency management of deadly floods: case study July 2021 floods in Germany. Natural Hazards and Earth System Sciences, 22(9), 3005–3014. https://doi.org/10.5194/nhess-22-3005-2022

Najafi, H., Kumar Shrestha, P., Rakovec, O., Apel, H., Vorogushyn, S., Kumar, R., Thober, S., Merz, B., & Samaniego, L. (2024). High-resolution impact-based early warning system for riverine flooding. Nature Communications, 15, 3726. https://doi.org/10.1038/s41467-024-48065-y

Tavares da Costa, R., Manfreda, S., Luzzi, V., Samela, C., Mazzoli, P., Castellarin, A., & Bagli, S. (2019). A web application for hydrogeomorphic flood hazard mapping. Environmental Modelling and Software, 118, 172–186. https://doi.org/10.1016/j.envsoft.2019.04.010