We sincerely appreciate your reading and insightful comments on our paper. Your suggestions will certainly help improve both the general clarity and the more detailed aspects of models and their considerations. Below, we provide responses to your specific comments.

Specific comments:

- Abstract: in general, I’d suggest not using acronyms in the abstract. Although FEWE is frequently used, I do not think the use of ABM here is particularly needed.

We agree and will adjust that on the revised manuscript.

- L51: better “how to protect”?

We think it will be better in this situation: “know what to do to protect”. The term “know what to do” is often used in literature in the context of warning systems. 

- L90: “extensive variety of flood types” - please, be more precise and clearer on explaining what you mean here.

The LifeSim model can be used for many kinds of floods, such as flash floods, dike and dam breaches. In the revised manuscript, we will be more precise and provide some examples of floods.

- L107-109: very minor comment: I think it is more logic to invert the order of these two approaches. First the issuing of the flood warning and then the efficiency of the communication solutions and strategies.

Considering the timeline of the warning system procedures, it makes sense to reverse the order. However, we needed to define diffusion and mobilisation scenarios to analyse the warning issuance. Therefore, the order we used is effective for this purpose because it first introduces the discussion of diffusion and mobilisation, which we then apply again to analysing the effects of warning issuance.

- Figure 3: I suggest adding the hydraulic networks in panel c)

We agree and will provide that in the revised manuscript. 

- L228: what is the sensitivity of the LifeSim model to such error threshold? In other words, does a difference of nearly 0.5 m on water depth affect the way the model consider the level of people safety?

LifeSim employs several default hazard thresholds to define exposure analysis, categorising agent exposure as high hazard when these thresholds are exceeded. As a probabilistic simulation, many thresholds are defined using probability distributions, which allow, in some way, for the incorporation of flood hazard uncertainties into the model's sampling approach. The thresholds are listed below, and in the revised manuscript, this discussion will be included in the section on uncertainties and limitations.

For submersion inside structures, two main thresholds are considered based on agent mobility. When mobility is an issue, a triangular distribution is proposed, ranging from the floor to the upper floor of the building, with thresholds of 1.22 m (minimum), 1.52 m (best estimate), and 1.83 m (maximum). If there is no mobility issue, a triangular distribution at the upper floor ceiling level is used, with thresholds of 0.15 m (minimum), 0.30 m (best estimate), and 0.46 m (maximum).

For building stability, the default engineered criteria utilise a uniform distribution between 7 m².s⁻¹ and 10 m².s⁻¹.

For people caught during evacuation, the model applies thresholds ranging from 0.6 m².s⁻¹ to 1.2 m².s⁻¹. For vehicles caught during evacuation, a triangular distribution is used, with parameters of 0.3 m².s⁻¹ (minimum), 0.8 m².s⁻¹ (most likely), and 1.3 m².s⁻¹ (maximum) for low-clearance vehicles, and 0.6 m².s⁻¹ (minimum), 1.2 m².s⁻¹ (most likely), and 2.4 m².s⁻¹ (maximum) for high-clearance vehicle.

- L269-274: this part is not totally clear to me. Please, try to improve the explanation.

In this passage, we explain how various building attributes were defined, based primarily on data collected through a post-incident survey. Here, we provide a more detailed explanation of each attribute, and we will reorganise the information in the revised manuscript to enhance its clarity and comprehensibility.

To determine the number of floors for each building, we first analysed the distribution of responses, which indicated the following proportions: 1.1 % had one floor, 28.6 % had two floors, 49.3 % had three floors, 15.6 % had four floors, 3.6 % had five floors, 1.5 % had six floors, and 0.4 % had seven floors. Based on these distributions, we assumed that buildings with higher population densities are more likely to have more floors. For instance, the 42 buildings identified as having the highest population densities were assigned seven floors.

The foundation height, defined as the vertical distance between the street elevation and the ground floor, is an important parameter for estimating water levels inside buildings during floods. To calculate this, we used survey responses where water levels were reported to have reached the ground floor during flooding events. By comparing water levels inside and outside buildings in these instances, we determined an average foundation height of 0.57 metres, which was applied in the simulations.

Access to an attic is another important attribute for assessing the exposure of individuals within buildings. The survey indicated that 36.5 % of buildings are likely to have attic access. This probability was incorporated into the LifeSim model, with each simulation iteration applying it to determine the likelihood of agents without mobility issues accessing this building level.

Lastly, we applied the default "engineered" stability criterion in the LifeSim model for building stability. This criterion considers a range of drag coefficient values, from 7 m²·s⁻¹, as Clausen and Clark (1990) recommended for traditional and smaller constructions, to 10 m²·s⁻¹ recommended by LifeSim for higher-quality constructions. For each iteration, a value was sampled uniformly between these two limits to reflect variability in construction stability.

- L279: Here, you refer to the safe evacuation places defined within the model. I have the following questions: Were these locations known to people prior to the event (i.e., before the flood occurred)? Were their locations communicated in any way during the event? How were these evacuation places defined? Were they identified based on official reports or risk management plans, or were they determined based on your own expertise?

We lacked direct information regarding whether people were aware of safe locations to evacuate to or if they were informed about such places during the event. According to the survey, only 13.8 % of the population received an early warning with evacuation instructions, which may indicate a lack of awareness or preparedness among the population during the flood. Consequently, we identified these safe locations based on our own expertise. Points were strategically placed along the valley, and dynamic visualisations of test simulation results were used to refine these locations. This iterative process helped ensure that the identified points realistically represented the flow of people evacuating to non-flooded areas.

- Figure 5: some of the safe places seem to be very close to the river. It might be useful to add the maximum flood extent, in order to see if some of them were affected.

We ensured that these locations were carefully defined outside the flood extent. We also acknowledge that incorporating a maximum flood extent layer on the map could enhance visualisation.

 -L293: please check the brackets.

We will correct that in the revised manuscript.

- L300: RPL acronym is not introduced in the text.

A small mistake occurred here; it should have been RLP. While we introduced this abbreviation at the beginning of Section 2.2, it would be better to use the full name again in this context, as it appears in a different section.

- L303: you always refer to CEST when reporting the timing. Is it really necessary?

The general recommendation of the journal is to specify the time zone when discussing events across different regions. As we aimed to describe the timeline of a specific event that affected multiple regions in Europe, we followed this guideline. Although we dealt with only one time zone, this approach ensures clarity and makes the information universally accessible.

- L300-305: I believe adding a kind of temporal timeline in which you summarize the different steps and phases that took place during the event would clarify the sequence of what was done during the event.

This is indeed a great idea. We used the Post Event Review Capability (PERC) Flood event review 'Bernd' (Szönyi et al., 2022) as the primary reference for describing the event, which provides a thorough analysis in terms of risk assessment. Including a detailed description of the warning system used during the event in Germany, with specific times and actions, could further enhance the overall understanding of the event’s dynamics, particularly in terms of both the hazard itself and the functioning of the warning system.

- Figure 6 and explanation: this figure was not totally clear to me at first read. I guess both panels show a range (the area within the two curves) of the people warned or active in time. However, in both text and caption you always refer to single curve. I think I got the current interpretation of the figure after having look at figure 7, where the representation of curves, plus range of variability, is clearer. I suggest revising this part.

Despite the from the survey, some uncertainties remained regarding the representation of both diffusion and mobilisation aspects. To address this, we defined a range for each aspect, within which the LifeSim routine samples a middle curve for each iteration. For diffusion, the range was determined based on the survey question regarding the time the flood reached the population. This question was framed with a range of values indicating the hour they were impacted. For mobilisation, many respondents who reported leaving their homes did not provide information on the time between becoming aware of the situation and leaving. Therefore, we used extreme values, action taken immediately upon awareness or up to six hours later (the latest reported time), to define two potential curves. A better explanation of this approach will be provided in the revised manuscript.

- Scenarios: are scenarios A1.4 and A1.5 related to the curves (or range of variability) reported in figure 6. If this is the case, I suggest to better stress this link in the text.

Yes, these are related to the curves in Figure 6. Scenario A1.4 combines the empirical diffusion curve from the 2021 flood (Figure 6a) with the slow theoretical mobilisation curve. Conversely, Scenario A1.5 pairs the slow theoretical diffusion curve with the empirical mobilisation curve from the 2021 flood (Figure 6b). Further details are available in the supplementary material, where we include a table outlining the parameters for each scenario. However, we will include key aspects of this information in the revised manuscript to make it clearer.

- Figure 8: results in the figure show a significant equifinality of the model outcomes, although moving from 0 to 1 in the x-axis express a significant difference on the efficiency of the considered measure. Summing those two aspects make this problem even worse (panel c). Any comments on this and regarding how this could affect the considerations emerged from the study?

We discuss the general aspects of Figure 8 and its impacts on the model outputs in lines 449–464. The equifinality observed in the sampled curves due to probabilistic scenario sampling is influenced by other probabilistic parameters in the model, such as thresholds in submergence and stability criteria. These uncertainties collectively contribute to the observed outcomes, where different levels of warning and mobilisation result in similar outcomes.

- Figure 9: figure shows the outcomes of the model, but, how do they correlate with the spatial and quantitative distribution of the occurred fatalities? The comparison is always done considering the whole number, while their spatial allocation is not analyzed. This comparison could provide useful insight regarding the approach adopted ty LifeSim model and the representativeness of the proxy variables adopted to simulate exposure and vulnerability.

The highest level of detail regarding the location of fatalities available to us was the Local Administrative Unit  (LAU). Therefore, we used this level to compare indoor fatalities, as outdoor fatalities involve uncertainties in their exact location. Overall, the model represents fatalities well in Bad Neuenahr-Ahrweiler, located in the middle of the domain, but tends to overestimate fatalities in upstream regions and underestimate them in downstream regions. This observation may be attributed to the generalisation of the analysis for warning diffusion and mobilisation times across the entire domain.

- L598 (conclusions, first line): I suggest revising this sentence since I do not think the applicability itself of the ABM to the case study was the purpose of the study, right? Actually, you chose this model based on literature evidences, looking at previous experiences and applications of the model to similar events.

We use Lifesim due to its capabilities to dynamically simulate warning systems and life loss and its applicability in other studies. Indeed, it would be better to revise this sentence to align more closely with the aims of the paper.

- 603: where can I see evidences of this? It seems to me that this result is not supported by the study outcomes. See also my previous comment regarding Figure 9.

We defined the warning diffusion curves based on survey data. One key question for that in the survey asked respondents about the time the flood reached their house. This reported time was subtracted from the lead time to determine when respondents became aware of the flood after issuing the first warning. For cases where no lead time was reported, we used the time the flood reached the house as the point when respondents became aware. Consequently, our warning diffusion curve is closely tied to this specific survey question, the moment the flood reaches the houses.

Since the flood affected different parts of the domain at different times (Figure 4c), and we generalised the warning diffusion and mobilisation for the entire domain, this introduces a limitation. In regions where the flood arrived earlier, the population’s response might have been expected sooner, but our model may sample a later response. Conversely, in downstream regions, responses might appear earlier than expected. This limitation arises because we do not have precise building locations for the survey respondents. We attempted to explain this in lines 475–488 and 585–590, but we will further clarify this in the revised manuscript.

- L619: When you mention the potential fatalities reduction (24.3%), I suggest explicitly referencing the figures or the specific scenario associated with this number for clarity and context.

The reduction in this case is associated with the A2.3 scenario, based on the estimated median during the later warning issuance hour (8–9 hours before the flood peak). In this scenario, the median number of fatalities is 115, representing a 24.3 % reduction compared to the median estimation of 152 fatalities in the reconstructed scenario.

References

Clausen, L. and Clark, P. B.: The development of criteria for predicting dambreak flood damages using modelling of historical dam failures, in: International conference on river flood hydraulics, 369–380, 1990.

Szönyi, M., Roezer, V., Deubelli, T., Ulrich, J., MacClune, K., Laurien, F., and Norton, R.: Post Event Review Capability flood event review ‘Bernd,’ Zurich, Switzerland. Zurich Insurance Company, 74 pp., 2022.

 

Thank you for reading and commenting on the manuscript. Your suggestions and observations provide valuable insights for enhancing the paper. Overall, we agree with your recommendations regarding the organisation of the exposure section and will make the necessary improvements in the revised version. Additionally, we have addressed the factors related to resolution and the existing bias in the hydraulic modelling in more detail. These general comments and our responses to your additional notes are discussed below.

Some additional notes:

L18-22: in the abstract, it is unclear if the presented information is the result of the LifeSim model or the historical records as they happened. Please, re-write to become clearer.

It was based on historical records identified during our review of the survey data, and we will clarify this in the revised manuscript.

L23-24: is it possible to include along the paper (not in the abstract) an estimation of how many people would react accordingly even if they received the “timely warning”? The 80% of life loss reduction accounts for the possibility that people could mis-value/ignore the warning and don’t react by evacuating?

This estimation of the corresponding reaction cannot be directly identified, but it can be indirectly linked to the question “know what to do”, which we can incorporate into the paper. Of the 246 respondents identified as potential participants in the Ahr Valley, approximately 76% rated their knowledge of what to do as Class 1 or 2 on a scale of 1 to 6, where 1 indicates "totally clear" and 6 indicates "totally unclear."

The reduction of up to 80 % is associated with a scenario where diffusion and mobilisation are considered optimal in our work, based on the work of  Sorensen and Mileti (2015a, b) on warning systems. The curves used to represent this scenario closely correspond to the best-case examples of diffusion and mobilisation from the historical database. The authors fitted a model (see Equations S1 and S2 in the supplementary material) that reflects the behaviour of the population in these best-case scenarios. For mobilisation, the optimal response curve is derived from the fastest recorded case in the 1987 Confluence (USA) hazardous material flow database.

In general, this curve, along with other theoretical models, underscores the limitations of population mobilisation. For the most likely curve in each scenario, the percentage of the population evacuated six hours after receiving a warning is 89.49 %, 74.07 %, and 57.03 % for scenarios of good, moderate, and poor preparedness and perception, respectively. These percentages eventually increase to their limits of 98.63 %, 95 %, and 90.02 % after 72 hours.

L102-104: the numbering of items (1) and (2) can be deleted.

We will delete it in the revised manuscript. 

L215-… Section 2.2.1 Flood Hazard modelling

Although it is a long river section (30 km), a resolution of 5 m in the DTM, seems to me to be a good compromise between the computational requirement and the quality of the flood zone and flood height intended for the application of the methodologies for assessing the degree of preparedness and evacuation. Have higher DTM resolutions (2 m or even 1 m) been tested? It is one thing to evacuate people in a context of a depth of 0.1 m; it is a different thing to do with 0.5 m or more. A less good DTM can translate into more or less significant flood depths differences locally (and velocities), which can condition the assessment of residents' behavior in the face of the need for evacuation. The behavior might have been the correct one given the real flood hazard characteristics, and the wrong one according to the modeled depths and velocities. Could this be relevant?

Indeed, the effects of DEM resolution on the hydraulic modelling results have been tested and published in other papers (Khosh Bin Ghomash et al., 2024a, b). DEM resolutions of 1m, 2m, 5m, and 10m were used to simulate the 2021 flood event in the Ahr Valley. The studies show that the DEM has an effect on flood propagation and the simulated maximum inundation depths and flow velocities, but the differences between the results obtained with the different DEM resolutions are relatively small. It could be concluded that for getting the maximum flood extent and water depths even the 10m resolution would be sufficient to get valid results close to the observations. For flow velocities, a finer resolution model performs better, but mainly for the flow simulation within the river channel, which is not properly mapped in a 10m DEM. On the floodplain and in the urban environment, the flow velocities are very similar between the different DEM resolutions, with the exception that with the 10m resolution, in which some narrow streets or small houses are not well resolved and thus, the flow dynamics differ in these areas compared to the finer resolution. The differences between the finer resolutions, i.e. 1m, 2m, and 5m, are small and are thus negligible for the estimation of flood risk, including the risk of loss of life. Because of these findings, the 5m resolution DEM was selected for this study as the best compromise between the quality of the model results and simulation performance.   

Regarding the relevance of water depths and flow velocities differences in the Lifesim, the model employs several default hazard thresholds to define exposure analysis, categorising agent exposure as high hazard when these thresholds are exceeded. As a probabilistic simulation, many thresholds are defined using probability distributions, which is an indirect consideration of the flood hazard uncertainties in the model's sampling approach.  

Roughness coefficients could be more detailed. Buildings can also accommodate flood waters inside but the adopted simplification is acceptable .

We used the highest resolution land use classification available for Germany (Mundialis based on Landsat, resolution 10m) to derive spatially distributed roughness values. This data set has only 6 LU classes, for which standard literature values were assigned. Other LU mappings have more classes (e.g. CORINE), but have a coarser resolution and the LU classes are very similar, resulting in approximately the same assignment of roughness values, which cannot be distinguished as differentiated as detailed LU classifications. For example, it is difficult to assign differentiating roughness values to LU classes like lawns, soccer fields, meadows or parks. Moreover, roughness values are neither unique for particular classes, because the roughness of the classes can differ over seasons or different hydraulic models react differently to roughness values, nor roughness values absolute, but rather effective and subject to calibration. 

Keeping this in mind, the effect of roughness parameterisation on the flood modelling with RIM2D was also tested (Khosh Bin Ghomash et al., 2024b) in a multi-objective model calibration of RIM2D for the Ahr flood 2021. While the original simulation of Apel et al. (2022) using standard roughness values already provided valid and reliable simulation results, it could be shown that the roughness calibration could improve the results even more, but not significantly better. The main improvement could be achieved for the flow dynamics in the river channel. In this presented study, we used then the best-performing roughness data set, which best fulfilled the 3 objectives of the calibration (flood extent, flood depth in floodplains, and flood dynamics in the river channel).   

Figure 4-b: the two extreme colors are somewhat similar (blue tones). Can the Authors test a different selection of colors for the two classes of lower velocity (for example, using something between the yellow and the green)?

We will test other combinations of colours following the journal's guidelines for colour blindness.

Figure 4-c: the red class (on July 14) seems to be absent from the map. If this is the case, could it be removed?

Yes, it can be removed. We will do that in the revised version.

L242-... In section 2.2.2 Exposure at buildings scale, it would help the reader to have a clear picture of the number of buildings (or households or persons...) that are assessed. 100-m and 1-km grids are mentioned along with a survey of 277 households. It becomes unclear how these several sources were integrated into a values in the LifeSlim model. For example, the reader still doesn’t understand at this point of the paper why an allocation of “population inside buildings” was done (“proportional distribution of the population based on building size”) if the 277 surveys would clearly provide that information. A prior clear explanation of the study area, and of the number of buildings, households, or people considered in the model would greatly help.

When reaching the end of page 11, these totals are presented, but in my opinion the paper would improve its clarity if those figures were presented earlier.

L263: the concept of vulnerability (vulnerable) appears only twice in the paper. An index is mentioned but it seems to be based on age only (at L263-264 a “1km grid resolution” source is referred; so, what was the use of the survey?). I suggest to not refer to the concept of vulnerability or to improve its index with more variables. It is also mentioned on figure 1 but I think it makes no sense to do that reference. On the other hand, would the 277-residents survey allow for a refinement of vulnerability? Would it be valuable for the LifeSlim model?

The description of the collection of the buildings’ characteristics could also be improved. Which ones were obtained from the Census and which ones from the local survey?  

This comment covers all aspects of the exposure considerations in our study. We will reorganise the explanation in this section to provide more explicit and more detailed information in the revised manuscript. In general, building characteristics were defined using data from the survey, population distribution, and vulnerability attributes derived from the 100-metre grid provided by HANZE 2.0.3 and the German Census. A more detailed and structured description of the sources and their application in the model is provided below. 

The survey was used to gather supplementary building information required for the LifeSim model. It provided data on (1) the number of floors, (2) foundation height, and  (3) attic access. (1) The distribution of building floors was as follows: 1.1% had one floor, 28.6% had two floors, 49.3% had three floors, 15.6% had four floors, 3.6% had five floors, 1.5% had six floors, and 0.4% had seven floors. Based on this distribution, buildings with higher population densities were assumed to have more floors. For example, the 42 buildings with the highest population densities were assigned seven floors. (2) The foundation height, calculated as the vertical distance between the street elevation and the ground floor, was determined using survey data from flooding events and averaged 0.57 m, the value used for all buildings. (3) Additionally, 36.5% of buildings were found to have attic access, a parameter incorporated into the LifeSim model to assess the likelihood of individuals without mobility issues reaching attic levels.

The 100-m grid served as the base for population distribution. The population of each grid was distributed proportionally among buildings, using building footprint area multiplied by the number of floors as a weighting factor to approximate the 3D built area. Some grids contained populations but no overlapping building footprints. So before doing this population allocation, we performed a redistribution using a 1-km aggregation scale to correct the discrepancies.

The 1-km grid was used as the aggregation scale during the redistribution of population data from the 100-metre grid. This step ensured a consistent and accurate representation of the population in grids only where buildings from OpenStreetMap exists.

German census data provided the proportion of individuals aged over 65, used to determine mobility vulnerabilities. According to LifeSim assumptions, 5.1% of people under 65 years old have mobility issues, while 22.6% of those over 65 face similar issues.

Figure 5: the selected buildings footprint (yellow) is creating some noise with the higer two classes of the population grid (light green and yellow). It is also not obvious why the resolution differs from the blue cells (more coarse) to the others (in green and yellow, more fine).

We will enhance the representation of colours in accordance with the journal's guidelines for colour blindness.

L300-301: RPL appears only here along the paper. Can you explain what it means?

A small mistake occurred here; it should have been RLP. While we introduced this abbreviation at the beginning of Section 2.2, it would be better to use the full name again in this context, as it appears in a different section.

L430, Table 2: my concern still stands on the 5-meter resolution. It surely misses slope rupture breaklines (frequent in urban landscapes) that make a huge difference from standing in a safe place to standing in a dangerous one. Could this apport differences in the loss of life estimations, explaining partly the +28.8 % difference in your results (lines 444-445, in addition to the reasons you present?

Indeed, this could be associated with one of the reasons. The limitations raised some concerns for us as well. We have tried to clarify this and will improve the discussion throughout the text in the revised manuscript, particularly in the Discussion of Limitations section. Another aspect relates to hazards. Building footprints were excluded from the digital elevation model (DEM), which was transformed into a digital surface model (DSM) to enhance urban flow representation. Buildings were then relocated to the highest-hazard pixel adjacent to their original footprint for life-loss estimation, as the LifeSim model requires a building point to overlay the hazard layer in order to perform the risk analysis. While this approach improves flow representation in urban areas, it may result in buildings being placed in locations that do not accurately reflect their actual exposure conditions.

We conducted a test in LifeSim using the water depth time series, adjusting all pixels across all layers by subtracting the bias value of 0.46 from the hydraulic simulation. In this test, the estimated median, minimum, and maximum values of fatalities are reduced to 126, 115, and 136, respectively. This represents the influence of hydraulic modelling on the sensitivity of life-loss estimations.

L560-562: It is much valuable to assume these limitations. I didn’t notice before that bridges weren’t accounted for in the hydrodynamic simulations. They have a two-sided role in terms of hazard upstream and downstream the element.

This is a general limitation encountered in flood hydraulic modelling of urban systems, where singularities can affect flow representation. In many models, this representation remains limited. We have highlighted this situation in this section as we consider it important to mention it as a limitation.

L571-572: Yes, this may also introduce some error in the estimations. These models are very complex and it becomes hard to represent accurately the most probable real situation.

Certain complexities of real situations are challenging to represent accurately in this type of modelling, which may influence the results, as we have attempted to highlight in these lines.

References

Apel, H., Vorogushyn, S., and Merz, B.: 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, 3005–3014, https://doi.org/10.5194/nhess-22-3005-2022, 2022.

Khosh Bin Ghomash, S., Apel, H., and Caviedes-Voullième, D.: Are 2D shallow-water solvers fast enough for early flood warning? A comparative assessment on the 2021 Ahr valley flood event, Natural Hazards and Earth System Sciences, 24, 2857–2874, https://doi.org/10.5194/nhess-24-2857-2024, 2024a.

Khosh Bin Ghomash, S., Yeste, P., Apel, H., and Nguyen, V. D.: Monte-Carlo based sensitivity analysis of the RIM2D hydrodynamic model for the 2021 flood event in Western Germany, Natural Hazards and Earth Systems Sciences Discussions, https://doi.org/10.5194/nhess-2024-77, 2024b.

Sorensen, J. H. and Mileti, D. S.: First Alert and/or Warning Issuance Time Estimation for Dam Breaches, Controlled Dam Releases, and Levee Breaches or Overtopping, Lakewood, Colorado, 1–48 pp., 2015a.

Sorensen, J. H. and Mileti, D. S.: Protective Action Initiation Time Estimation for Dam Breaches, Controlled Dam Releases, and Levee Breaches or Overtopping, Lakewood, Colorado, 1–51 pp., 2015b.