Towards Resilient Critical Infrastructures: Understanding the Impact of Coastal Flooding on the Fuel Transportation Network in the San Francisco Bay

Round 1

Reviewer 1 Report

Review of the paper "Towards Resilient Critical Infrastructures: 2
Understanding the Impact of Coastal Flooding on the Fuel Transportation Net-3 work in the San Francisco Bay"  Yiyi He , Sarah Lindbergh, Yang Ju, Marta Gonzalez, John Radke

In my opinion the topic of the paper is very interesting and actual. New methodologies  and approaches to assess the vulnerability of infractructures in coastal region to Sea level Rise, taking into account  incentainties in climate projections,   constitute an important contribution to identify and to design adaptation policies and coastal protecion structures. 

Furthermore the paper is well written, the methodology is clearly explained and the results are consistent. Therefore I think that  manuscript  is worth to be published on IJGI. 

Just I suggest to authors to  describe   better the floodding  simulations. They carried out 120 hydraulic simulations with a 3D code by Delft.  This is fine, but it  is a very time consuming work and computer power. Why did they not use a shallow water code ? Furthermore what is the spatial resolution of their hydraulic model ? How did take into account buildings and road and others geometrically very complex coastal infrastructures.  Have they used subgrig models to represent such complexity? 
I am sure that such informations would improve the quality and reliability of the study. 

Author Response

Comment: I suggest to authors to describe better the flooding simulations. They carried out 120 hydraulic simulations with a 3D code by Delft. This is fine, but it is very time-consuming work and computer power. Why did they not use a shallow water code? Furthermore, what is the spatial resolution of their hydraulic model? How did take into account buildings and roads and others geometrically very complex coastal infrastructures? Have they used subgrid models to represent such complexity?

Response: Thank you for recognizing our contribution and for providing us with constructive comments. To our knowledge, four different flow domains are defined in 3Di. One of the domains-2D surface flow is based on the 2D depth-averaged shallow water equations. These equations are based on the conservation of momentum. 3Di considers the various processes; inertia, advection, pressure, and friction for computing the horizontal flow. This 3Di model is based on Stelling’s original work (Stelling, G.S. Quadtree Flood Simulations with Sub-Grid Digital Elevation Models. Proceedings of the Institution of Civil Engineers - Water Management 2012, 165, 567–580, doi:10.1680/wama.12.00018) and although it is computationally heavy, 3Di uses a quadtree structure to refine and allow accurate tracking of levees, roads and other surface objects that are finer than the spatial resolution. We did not include buildings in this 50 meter run of the model but we did generate a 1-meter resolution DEM from a Lidar dataset, extract flood blocking features like levees, roads, and railroads, and included them in the final 50-meter dataset. Addressing your question on whether we used sub-grid models to represent the complexity in bathymetry and surface structures, the answer is yes. For clarification purposes, we made changes to section 2.2 as follows:

Section 2.2, Line 145-161:

‘To achieve this goal, we leverage coastal flooding model outputs from a research project led by Radke et al. [22] in which a hydrodynamic model-3di [35,36] was used to model coastal flooding for 120 climate scenarios. The 3Di hydrodynamic model developed by the Delft University of Technology is a commercial model that dynamically simulates the movement of tides and flood events over digital representations of land surfaces. The inputs of 3Di include time-series water levels as boundary forcing to generate water flows, and digital surface data containing topography, bathymetry, and aboveground objects such as levees to direct the waterflows. It should be noted that building structures are excluded in our model since they are too granular to be represented. However, critical flood prevention structures namely levees are preserved [34]. A unique advantage of the 3Di model for this study is its ability to feasibly compute over large regions at fine spatial resolutions, which is enabled by the model's quadtree-based compression algorithm that simplifies the input digital surface while preserving significant topographic variations such as those from levees [34]. Our model simulates an entire tidal cycle and calculates, in a series of time steps, the flow direction, velocity, and water depth as a flood event progresses. A detailed description of the modeling process is well documented in reference [22] under Appendix C’

Reviewer 2 Report

The paper is well written. My only comments is related to the hydrodynamic model 3di the authors used. It would be helpful to the readers to include some brief introduction of the model (line 153) on the model setup, resolution etc

Also, please double check if the model is 3D (line 366), most urban flooding models are 1D or 2D or 1D-2D coupled for performance reason, I double the hydrodynamic model is 3D.

Author Response

Point 1: It would be helpful to the readers to include some brief introduction of the model (line 153) on the model setup, resolution, etc.

Response 1: Thank you for your suggestion. We modified section 2.2 in the revised manuscript to clarify questions and concerns about the flood model:

Section 2.2, Line 145-161:

‘To achieve this goal, we leverage coastal flooding model outputs from a research project led by Radke et al. [22] in which a hydrodynamic model-3di [35,36] was used to model coastal flooding for 120 climate scenarios. The 3Di hydrodynamic model developed by the Delft University of Technology is a commercial model that dynamically simulates the movement of tides and flood events over digital representations of land surfaces. The inputs of 3Di include time-series water levels as boundary forcing to generate water flows, and digital surface data containing topography, bathymetry, and aboveground objects such as levees to direct the waterflows. It should be noted that building structures are excluded in our model since they are too granular to be represented. However, critical flood prevention structures namely levees are preserved [34]. A unique advantage of the 3Di model for this study is its ability to feasibly compute over large regions at fine spatial resolutions, which is enabled by the model's quadtree-based compression algorithm that simplifies the input digital surface while preserving significant topographic variations such as those from levees [34]. Our model simulates an entire tidal cycle and calculates, in a series of time steps, the flow direction, velocity, and water depth as a flood event progresses. A detailed description of the modeling process is well documented in reference [22] under Appendix C’

Point 2: Also, please double-check if the model is 3D (line 366), most urban flooding models are 1D or 2D or 1D-2D coupled for performance reasons, I doubt the hydrodynamic model is 3D.

Response 2: Thank you for your comment. The 3Di model is indeed 2D. In the revised manuscript, we changed back to the model’s original name ‘3Di’ and added additional references and notes in section 2.2 (where we first talk about the model) to help readers understand the model.

Section 4, Line 382-384:

‘Our preliminary experiments in the Bay Area using the 3Di hydrodynamic model that created the 120 coastal flooding scenarios show impact potential from inland flooding caused by extreme precipitation events.’

Reviewer 3 Report

This paper investigates the impact of coastal flooding on fuel transportation networks in the San Francisco Bay at regional and local scales. A total of 120 coastal flooding scenarios, including four General Circulation Models, two Representative Concentration Pathways, three percentiles of future SLR estimates, and five planning horizons (20-year intervals from 2000 to 2100) are considered to perform a comprehensive analysis.

This paper is well written and easy to follow. Investigating coastal flooding impacts on fuel transportation networks is very important and much needed to achieve a better and sustainable future under climate change. I believe the topic of this paper will be of great interest to the readers. However, some key points are unclear and need to be elaborated to improve the readability of this paper. Therefore, this paper can be accepted with minor revisions. My specific comments are provided below.

L52-63: An example of Colonial Pipeline incident due to a ransomware attack is provided here as a good demonstration of the impact of critical infrastructure (CI) failures. However, this example is not related to the topic of this paper with a focus on the impact of coastal flooding on the fuel transportation network. In other words, I think this example is unnecessary and should thus be deleted.

L92: I am confused about the short-term elevation of sea levels from dynamic coastal flooding. Did you mean that coastal flooding can further increase sea levels? Please clarify.

L157-164: The spatial resolution of General Circulation Models (GCMs) is coarse (approximately 100-300 km). How did you create flood inundation time series maps at 50-meter spatial resolution? What about the reliability of flood inundation simulations using GCMs? How did you validate the 3Di hydrodynamic model used for flood inundation simulations? In addition, how did you obtain the hourly time series of water levels under different RCPs?

L302: Please change “Figure8Figure9” to “Figures 8 and 9”.

L306-307: Are you sure that the variation of exposure increases over time due to the uncertainties in future climate projections? Is it not because of increasing warming levels?

L315: What kinds of uncertainties are you referring to?

L330-331: Please explicitly explain why the drop in network efficiency happens earlier in the fourth time horizon (2060–2080) and then slightly increases in the last time horizon.

Author Response

Point 1: (L52-63) An example of Colonial Pipeline incident due to a ransomware attack is provided here as a good demonstration of the impact of critical infrastructure (CI) failures. However, this example is not related to the topic of this paper with a focus on the impact of coastal flooding on the fuel transportation network. In other words, I think this example is unnecessary and should thus be deleted.

Response 1: Thank you for this suggestion. We agree that the Colonial Pipeline ransomware attack example is relatively tangent from the topic of this paper and therefore removed it in the revised manuscript.

Point 2: (L92) I am confused about the short-term elevation of sea levels from dynamic coastal flooding. Did you mean that coastal flooding can further increase sea levels? Please clarify.

Response 2: In this study, we model coastal flooding as a result of both sea level rise and storm surge. We agree the previous sentence is therefore confusing so we changed the term “coastal flooding” here for “storm surge” . Based on the most recent report from IPCC (AR5), the frequency and magnitude of climate-change-induced extreme weather events such as storm surges are expected to increase. Therefore, the impact of long-term sea level rise at regional and global scales is amplified by short-term storm surges. For clarification purposes, we changed the original sentence to “Furthermore, the short-term elevation of sea levels from storm surges are also expected to intensify the effects of global and regional sea level rise.” (Line 83-85 in the revised manuscript)

Point 3: (L157-164) The spatial resolution of General Circulation Models (GCMs) is coarse (approximately 100-300 km). How did you create flood inundation time series maps at 50-meter spatial resolution? What about the reliability of flood inundation simulations using GCMs? How did you validate the 3Di hydrodynamic model used for flood inundation simulations? In addition, how did you obtain the hourly time series of water levels under different RCPs?

Response 3: We used projected sea levels at the San Francisco NOAA tidal gauge as the water level inputs (boundary forcing) to our flood model. The projected sea levels include probabilistic sea level rise values under different emission scenarios, storm surge projected with the GCMs, and tide (Cayan, Kalansky, Iacobellis, & Pierce, 2016, ref. 38). We used a 50 m resolution digital surface model as the second input. The sea levels and the digital surface model were used in 3Di to produce flood inundation time series.

We validate the model setup by simulating a historical extreme flood event and then comparing observed and simulated water levels at five tidal gauges in the SF Bay. The simulated water levels, in general, match the observation (Pearson’s r>=0.81, RMSE<=0.69 m, depending on the station). We used extreme sea level events to simulate flooding. Each event is a 72-h window, starting with the highest sea level projected under a given climate scenario during a 20-year period between 2000 and 2100. A detailed description of the modeling and validation processes are well documented in two key references in the revised manuscript (provided in section 2.2 where the coastal flood model is introduced):

• Radke, J.D, G.S. Biging, K. Roberts, M. Schmidt-Poolman, H. Foster, E. Roe, Y. Ju, S. Lindbergh, T. Beach, L. Maier, Y. He, M. Ashenfarb, P. Norton, M. Wray, A. Alruheili, S. Yi, R. Rau, J. Collins, D. Radke, M. Coufal, S. Marx, A. Gohar, D. Moanga, V. Ulyashin, A. Dalal. (University of California, Berkeley) 2018. Assessing Extreme Weather-Related Vulnerability and Identifying Resilience Options for California’s Interdependent Transportation Fuel Sector. California’s Fourth Climate Change Assessment, California Energy Commission. Publication Number: CCCA4-CEC-2018-012.
• Ju, Y.; Lindbergh, S.; He, Y.; Radke, J.D. Climate-Related Uncertainties in Urban Exposure to Sea Level Rise and Storm Surge Flooding: A Multi-Temporal and Multi-Scenario Analysis. Cities 2019, 92, 230–246, doi:10.1016/j.cities.2019.04.002.

Point 4: (L302) Please change “Figure8Figure9” to “Figures 8 and 9”.

Response 4: Thank you for pointing out this error. We have corrected this formatting mistake in the revised manuscript.

Point 5: (L306-307) Are you sure that the variation of exposure increases over time due to the uncertainties in future climate projections? Is it not because of increasing warming levels?

Response 5: We draw a distinction between the uprising trend in terms of asset exposure across different time horizons and the variation within the same time horizon. The former is related to increasing warming levels and the latter is related to variation in climate change projections. We agree that the original sentence is confusing and therefore made the following changes:

Section 3.1, Line 314-316:

‘The variation of exposure (within the same time horizon) increases over time due to the uncertainties in future climate projections.’

Point 6: (L315) What kinds of uncertainties are you referring to?

Response 6: Here we aim to highlight the variation in the number of connected components over five time horizons (shown in Figure 9). As we move from the first time horizon (2000-2020) to the last (2080-2100), we are able to observe a dramatic increase. We do realize that there is a difference between uncertainty (quantified by a probability distribution) and variability (quantified by a distribution of frequencies of multiple instances) and therefore changed the wording from uncertainty to variability in the revised manuscript.

Section 3.1, Line 322-323

‘This result suggests that, due to coastal flooding, the multimodal network becomes more and more fragmented over time with increasing variation.’

Point 7: (L330-331) Please explicitly explain why the drop in network efficiency happens earlier in the fourth time horizon (2060–2080) and then slightly increases in the last time horizon.

Response 7: Thank you for your suggestion. We added a detailed description of the calculation of global efficiency and potential explanations for the observed changes under RCP 8.5 scenarios in the revised manuscript.

Section 3.1, Line 336-350:

‘The global efficiency metric, measures and compares flow exchanges between nodes across different time horizons (Figure 10). Our results show that, in general, the efficiency within the multimodal network decreases over time for both RCP 4.5 and RCP 8.5 sce-narios. For the first three time horizons (2000 – 2020, 2020 – 2040, 2040 – 2060), network efficiency remains stable. For RCP 4.5 scenarios, a steep drop in network efficiency occurs during the last time horizon (2080 – 2100) yet for RCP 8.5 scenarios, the drop in network efficiency happens earlier in the fourth time horizon (2060 – 2080). In the last time horizon, the minimum value for global efficiency is higher compared with the fourth time horizon. This is potentially caused by the significant divergence between RCP 4.5 and RCP 8.5 after the third time horizon (Figure 7) which results in heterogeneous exposure of fuel transportation infrastructure to coastal flooding under different climate scenarios. As is previously mentioned in section 2.4, the shortest path distance between node pairs is a key component of the global efficiency metric. It is possible that the shortest path distances for some node pairs reduce in some flood scenarios resulting in a higher global efficiency value.’