This study evaluates the impact of potential future climate change on flood
regimes, floodplain protection, and electricity infrastructures across the
Conasauga River watershed in the southeastern United States through ensemble
hydrodynamic inundation modeling. The ensemble streamflow scenarios were
simulated by the Distributed Hydrology Soil Vegetation Model (DHSVM) driven
by (1) 1981–2012 Daymet meteorological observations and (2) 11 sets of
downscaled global climate models (GCMs) during the 1966–2005 historical and
2011–2050 future periods. Surface inundation was simulated using a
GPU-accelerated Two-dimensional Runoff Inundation Toolkit for Operational
Needs (TRITON) hydrodynamic model. A total of 9 out of the 11 GCMs exhibit an
increase in the mean ensemble flood inundation areas. Moreover, at the 1 %
annual exceedance probability level, the flood inundation frequency curves
indicate a
This paper has been authored by UT-Battelle, LLC, under
contract DE-AC05-00OR22725 with the US Department of Energy (DOE). The US
government retains and the publisher, by accepting the article for
publication, acknowledges that the US government retains a nonexclusive,
paid-up, irrevocable, worldwide license to publish or reproduce the
published form of this paper, or allow others to do so, for US
government purposes. DOE will provide public access to these results of
federally sponsored research in accordance with the DOE Public Access Plan
(
Floods are costly disasters that affect more people than any other natural hazard around the world (UNISDR, 2015). Major factors that can exacerbate flood damage include population growth, urbanization, and climate change (Birhanu et al., 2016; Winsemius et al., 2016; Alfieri et al., 2017, 2018; Kefi et al., 2018). Recent observations exhibit an increase in the frequency and the intensity of extreme precipitation events (Pachauri and Meyer, 2014), which have strengthened the magnitude and frequency of flooding (Milly et al., 2002; Langerwisch et al., 2013; Alfieri et al., 2015a, 2018; Mora et al., 2018). As a result, the damage and cost of flooding have substantially increased across the United States (US) (Pielke and Downton, 2000; Pielke et al., 2002; Ntelekos et al., 2010; Wing et al., 2018) and the rest of the world (Hirabayashi et al., 2013; Arnell and Gosling, 2014; Alfieri et al., 2015b, 2017; Kefi et al., 2018).
Since 1968, the National Flood Insurance Program (NFIP), administered by the
Federal Emergency Management Agency (FEMA), has implemented floodplain
regulation standards in the US to mitigate the escalating flood losses
(Bedient et al., 2013). For communities participating in the NFIP, flood insurance is
required for structures located within the 1 % annual exceedance
probability (AEP) flood zone (i.e., areas with probability of flooding
The increases in the magnitude and frequency of flooding, in addition to the inadequacy of floodplain measures and the high costs of hardening (Wilbanks et al., 2008; Farber-DeAnda et al., 2010; Gilstrap et al., 2015), have put electricity infrastructures at risk (Zamuda et al., 2015; Zamuda and Lippert, 2016; Cronin et al., 2018; Forzieri et al., 2018; Mikellidou et al., 2018; Allen-Dumas et al., 2019). In particular, electricity infrastructures which lie in areas vulnerable to flooding can experience floodwater damages that may lead to changes in their energy production and consumption (Chandramowli and Felder, 2014; Ciscar and Dowling, 2014; Bollinger and Dijkema, 2016; Gangrade et al., 2019). For instance, flooding can rust metals, destroy insulation, and damage interruption capacity (Farber-DeAnda et al., 2010; Vale, 2014; NERC, 2018; Bragatto et al., 2019). It is estimated that nearly 300 energy facilities are located on low-lying lands vulnerable to sea-level rise and flooding in the lower 48 US states (Strauss and Ziemlinski, 2012).
Several studies have assessed the vulnerability of electricity
infrastructures to flooding (Reed et al., 2009; Winkler et al., 2010;
Bollinger and Dijkema, 2016; Fu et al., 2017; Pant et al., 2017; Bragatto et
al., 2019; Gangrade et al., 2019). For highly sensitive water
infrastructures such as dams (McCuen, 2005), Gangrade et al. (2019) showed
that the surface inundation associated with probable maximum flood (PMF) is
generally projected to increase in future climate conditions. However, given
the extremely large magnitude of PMF (AEP
The objective of this study is to demonstrate the applicability of a
computationally intensive ensemble inundation modeling approach to better
understand how climate change may affect flood regimes, floodplain
regulation standards, and the vulnerability of existing infrastructures.
Extending from the framework developed by Gangrade et al. (2019) for
PMF-scale events (AEP The framework will evaluate the changes in flood regime using high-resolution ensemble flood
inundation maps. The ensemble-based approach is able to incorporate the
large hydrologic interannual variability and model uncertainty that cannot
be captured through the conventional deterministic flood map. The framework will enable direct frequency analysis of ensemble flood inundation maps that
correspond to historic and projected future climate conditions. This
approach provides an alternative floodplain delineation technique to the
conventional approach, in which a single deterministic design flood value is
used to develop a flood map with a given exceedance probability. The framework will evaluate the vulnerability of electricity infrastructures to climate-change-induced flooding and assess the adequacy of existing flood
protection measures using ensemble flood inundation. This information will
help floodplain managers to identify the most vulnerable infrastructures and
recommend suitable adaptation measures.
The following technique was adopted in this study. First, we generated streamflow projection by utilizing an ensemble of simulated streamflow hydrographs driven by both historical observations and downscaled climate projections (Gangrade et al., 2020) as inputs for hydrodynamic inundation modeling as presented in Sect. 2.2. Then, we set up and calibrated a 2D hydrodynamic inundation model, Two-dimensional Runoff Inundation Toolkit for Operational Needs (TRITON; Morales-Hernández et al., 2021), in our study area which is presented in Sect. 2.3. For inundation modeling, sensitivity analyses were conducted on three selected parameters to quantify and compare their respective influences on modeled flood depths and extents. The performance of TRITON was then evaluated by comparing a simulated 1 % AEP flood map with the reference 1 % AEP flood map from the Federal Emergency Management Agency (FEMA). Finally, as presented in Sect. 2.4 and 2.5, ensemble inundation modeling was performed to develop flood inundation frequency curves and maps and to assess the vulnerability of electricity infrastructures under a changing climate, respectively.
The article is organized as follows: the data and methods are discussed in Sect. 2; Sect. 3 presents the result and discussion; and the summary is presented in Sect. 4.
Our study area is the Conasauga River watershed (CRW) located in
southeastern Tennessee and northwestern Georgia (Fig. 1). The CRW is an
eight-digit Hydrologic Unit Code (HUC08) subbasin (03150101) with a total
drainage area of
Conasauga River watershed study area location, model extent, electric substations, and inflow locations. Background layer source: © OpenStreetMap contributors 2021. Distributed under a Creative Commons BY-SA License.
The ensemble streamflow projections were generated by a hierarchical modeling framework, which started with regional climate downscaling followed by hydrologic modeling (Gangrade et al., 2020). The climate projections were generated by dynamically downscaling 11 GCMs from the Coupled Model Intercomparison Project Phase-5 (CMIP5) data archive. Each GCM was used as lateral and lower boundary forcing in a regional climate model RegCM4 (Giorgi et al., 2012) at a horizontal grid spacing of 18 km over a domain that covered the continental US and parts of Canada and Mexico (Ashfaq et al., 2016) (Table 1). Each RegCM4 integration covered 40 years in the historic period (1966–2005; hereafter baseline) and another 40 years in the future period (2011–2050) under the Representative Concentration Pathway 8.5 (RCP8.5) emission scenario, with a combined 880 years of data across all RegCM4 simulations. To capture the multi-decadal climate variability, a minimum period of 30 years has been used in many studies (e.g., Alfieri et al., 2015a, b). Given the additional data available from Gangrade et al.(2020), we have adopted a longer 40-year period that may further enlarge the sample space to better support the statistical analyses in this study.
Summary of the 11 dynamically downscaled climate models (adopted from Ashfaq et al., 2016).
The RegCM4 simulated daily precipitation and temperature were further statistically bias-corrected to a spatial resolution of 4 km following a quantile mapping technique, described in Ashfaq et al. (2010, 2013). The 4 km Parameter-elevation Regressions on Independent Slopes Model (PRISM; Daly et al., 2008) data were used as the historic observations to support bias correction. In the baseline period, the simulated quantiles of precipitation and temperature were corrected by mapping them onto the observed quantiles. In the future period, the monthly quantile shifts were calculated based on the simulated baseline and future quantiles which were subsequently added to the bias-corrected baseline quantiles to generate bias-corrected monthly future data. Finally, the original daily values were rescaled to meet the corrected monthly precipitation and temperature. This approach substantially improves the biases in the modeled daily precipitation and temperature while preserving the simulated climate change signal. Further details of the bias correction are provided in Ashfaq et al. (2010, 2013) while the information regarding the RegCM4 configuration, evaluation, and future climate projections is detailed in Ashfaq et al. (2016).
The hydrologic simulations were then conducted using the Distributed Hydrology Soil Vegetation Model (DHSVM; Wigmosta et al., 1994), which is a process-based high-resolution hydrologic model that can capture heterogeneous watershed processes and meteorology at a fine resolution. DHSVM uses spatially distributed parameters, including topography, soil types, soil depths, and vegetation types. The input meteorological data include precipitation, incoming shortwave and longwave radiation, relative humidity, air temperature, and wind speed (Wigmosta et al., 1994, 2002; Storck et al., 1998). The DHSVM performance and applicability have been reported in various earlier climate- and flood-related studies (Elsner et al., 2010; Hou et al., 2019; Gangrade et al., 2018, 2019, 2020). A calibrated DHSVM implementation from Gangrade et al. (2018) at 90 m grid spacing was used to produce 3-hourly streamflow projections using the RegCM4 meteorological forcings described in the previous section (Table 1). In addition, a control simulation driven by 1981–2012 Daymet meteorologic forcings (Thornton et al., 1997) was conducted for model evaluation and validation. The hydrologic simulations used in this study are a part of a larger hydroclimate assessment effort for the ACT River basin, as detailed in Gangrade et al. (2020). Since there is no major reservoir in the CRW, the additional reservoir operation module (Zhao et al., 2016) was not needed in this study.
Note that while the ensemble streamflow projections based on dynamical downscaling and high-resolution hydrologic modeling from Gangrade et al. (2020) are suitable to explore extreme hydrologic events in this study, they do not represent the full range of possible future scenarios. Additional factors such as other GCMs, RCP scenarios, downscaling approaches, and hydrologic models and parameterization may also affect future streamflow projections. In other words, although these ensemble streamflow projections can tell us how likely the future streamflow magnitude may change from the baseline level, they are not the absolute prediction into the future. In practice, these modeling choices will likely be study-specific based on the agreement among key stakeholders. It is also noted that the new Coupled Model Intercomparison Project Phase-6 (CMIP6) data have also become available to update the ensemble streamflow projections but is not pursued in this study.
The ensemble inundation modeling was performed using TRITON, which is a computationally enhanced version of Flood2D-GPU (Kalyanapu et al., 2011). TRITON allows parallel computing using multiple graphics processing units (GPUs) through a hybrid Message Passing Interface (MPI) and Compute Unified Device Architecture (CUDA) (Morales-Hernández et al., 2021). TRITON solves the nonlinear hyperbolic shallow water equations using an explicit upwind finite-volume scheme, based on Roe's linearization. The shallow water equations are a simplified version of the Navier–Stokes equations in which the horizontal momentum and continuity equations are integrated in the vertical direction (see Morales-Hernández et al., 2021, for further model details). An evaluation of TRITON performance for the CRW is presented and discussed in Sect. 3.3.
TRITON's input data include digital elevation model (DEM), surface
roughness, initial depths, flow hydrographs, and inflow source locations
(Kalyanapu et al., 2011; Marshall et al., 2018; Morales-Hernández et
al., 2020, 2021). In this study, the hydraulic
and geometric parameters from the flood model evaluation section (Sect. 3.3) were used in the flood simulation. The topography was represented using
the one-third arc-second (
The simulated DHSVM streamflow was used to prepare inflow hydrographs for
ensemble inundation modeling. To provide a large sample size for frequency
analysis, we selected all annual maximum peak streamflow events (the maximum
corresponded to the outlet of CRW Fig. 1) from the 1981–2012 control
simulation (32 years), the 1966–2005 baseline simulation (440 years; 40 years
Given the nature of GCM experiments, each set of climate projections can be considered a physics-based realization of historic and future climate under specified emission scenarios. Therefore, an ensemble of multimodel simulations can effectively increase the data lengths and sample sizes that are keys to support frequency analysis, especially for low-AEP events. In this study, we conducted flood frequency analyses separately for the 1966–2005 baseline and 2011–2050 future periods so that the difference between the two periods represents the changes in flood risk due to climate change.
To prepare the flood frequency analysis, we first calculated the maximum flood depth at every grid in each simulation. A minimum threshold of 10 cm flood depth was used to judge whether a cell was wet or dry (Gangrade et al., 2019). Further, for a given grid cell, if the total number of non-zero flood depth values (i.e., of the 440 depth values) was fewer than 30, the grid cell was also considered dry. This threshold was selected based on the minimum sample size requirement for flood depth frequency analysis suggested by Li et al. (2018). Next, we calculated the maximum flooded area (hereafter used alternatively with “floodplain area”) for each simulation. A log-Pearson type III (LP3) distribution was then used for frequency analysis following the guidelines outlined in Bulletins 17B (USGS, 1982; Burkey, 2009) and 17C (England et al., 2019). Two types of LP3 fitting were performed. The first type of fitting is event-based that fitted LP3 on the maximum inundation area across all ensemble members. The second type of fitting is grid-based (more computationally intensive) that fitted LP3 on the maximum flood depth at each grid cell across all ensemble members. For both types of fittings, the frequency estimates at 4 %, 2 %, 1 %, and 0.5 % AEP (corresponding to 25-, 50-, 100-, and 200-year return levels) were derived for further analysis.
It is also noted that in addition to the annual maximum event approach used in this study, one may also use the peak-over-threshold (POT) approach which can select multiple streamflow events in a very wet year. While such an approach can lead to higher extreme streamflow and inundation estimates, the timing of POT samples is fully governed by the occurrences of wet years. In other words, if the trend of extreme streamflow is significant in the future period, the POT samples will likely occur more in the far future period. We hence select the annual maximum event approach that can sample maximum streamflow events more evenly in time, which can better capture the evolution of extreme events with time under the influence of climate change.
The vulnerability of electricity infrastructures to climate-change-induced flooding was evaluated using the ensemble flood inundation results. The 44 electric substations (Fig. 1) collected from the publicly available Homeland Infrastructure Foundation-Level Data (HIFLD, 2019) were considered to be the electrical components susceptible to flooding. To evaluate the vulnerability of these substations, we overlapped the maximum flood extent from each ensemble member with all substations to identify the substations that might be inundated under the baseline and future climate conditions. Further, as an additional flood hazard indicator, the duration of inundation was estimated at each of the affected substations using the ensemble flood simulation results.
The vulnerability analysis was performed for two different flood mitigation
scenarios. In the first scenario, we assumed that no flood protection
measures were provided at all substations. Hence, the substations that
intersected with the flood footprint were considered to be failed. In the
second scenario, it was assumed that flood protection measures were adopted
for all substations following the FEMA P-1019 recommendation (FEMA, 2014).
According to FEMA P-1019 (FEMA, 2014), for emergency power systems within
critical facilities, the highest elevation among (1) the base flood
elevation (BFE: 1 % FEMA AEP flood elevation) plus 3 ft (
During the vulnerability analysis, we also assumed that (1) the one-third arc-second spatial resolution DEM might reasonably represent the elevation of substations, (2) existing substations would remain functional and would not be relocated, and (3) no additional hardening measures (i.e., protections such as levees, berms, anchors, and housings) will be adopted in the future period. Also, the cascading failure of a substation due to grid interconnection was not considered in this study.
This section presents a comparison of the annual maximum peak streamflow (at
the outlet of CRW) used in the control, baseline, and future simulations.
The sample size included 32 events from the control (1981–2012) simulation,
440 events from the baseline (1966–2005) simulations, and another 440
events from the future (2011–2050) simulations. These samples are
illustrated in box-and-whisker plots in Fig. 2, where the central mark
indicates the median, while the bottom and top edges indicate the 25th and
75th percentiles, respectively. The whiskers extend to the furthest data
points not considered outliers, which correspond to approximately
A comparison of annual maximum peak streamflow at the outlet of the Conasauga River watershed. The sample size includes 32 events from the control (1981–2012), 440 from the baseline (1966–2005), and another 440 from the future (2011–2050) periods.
Under the future projection, an increase in the maximum peak streamflow is
shown, where the upper whisker in the future projection is
For a better understanding and selection of suitable TRITON parameters, a series of sensitivity analyses were conducted using different combinations of Manning's roughness, initial water depths, and river bathymetry correction factors (Table 2).
Summary of hydraulic and geometric parameters used in the sensitivity analysis.
Note:
In calibrating a hydraulic model, it is a common practice to adjust the
estimated Manning's
To establish an initial condition for TRITON, a sensitivity analysis was
performed on selected initial water depth values (ranging from 0 to 0.65 m, Table 2) to understand their relative effects. To select ranges for the
initial water depth, we summarized the observed water depth values that
correspond to low flow values at five USGS gauge stations inside the CRW.
The distribution of observed water depth values from the five gauges showed
average values ranging from 0.25 to 0.65 m. Existing DEM products, even those
with high spatial resolution (i.e., 10 m or finer), do not represent the
elevation of river bathymetry accurately (Bhuyian et al., 2014). For the
CRW, Bhuyian et al. (2019) found that the one-third arc-second spatial
resolution base DEM over-predicted the inundation extent because of the
bathymetric error, which reduced the channel conveyance. In this study, we
tested various bathymetry correction factors (ranging from
The sensitivity analysis was performed using the 13–22 February 1990 flood
event that has the maximum discharge among all 32 control simulation events.
To evaluate relative sensitivity of TRITON, we extracted simulated flood
depths at four arbitrary selected locations (Fig. 1) and estimated the
relative inundation area differences. The impacts of initial water depths
were significant only at the beginning where low flow values dominated the
hydrographs (Fig. 3a, d, g, and j). Larger initial water depth values
generated higher flood inundation depths for all sample locations. Although
the differences in flood inundation extents relative to the dry bed show an
increasing trend, the relative differences are less than 1.4 % (Fig. 4a). Similarly, the differences in average peak water depths and time to
peak relative to the 0.35 m initial water depth were less than 1.0 %
(Table 3). Increase in the channel and floodplain Manning's
Simulated flood inundation depths extracted at location 1
Change in simulated maximum flood inundation extents for
Change in peak water depth and time to peak.
Because of a lack of observed streamflow data in the CRW, the performance of TRITON was evaluated by comparing the simulated 1 % AEP flood map with the published 1 % AEP flood map from FEMA (FEMA, 2019). The purpose of this assessment is to understand whether TRITON can provide comparable results to the widely accepted FEMA flood estimates. While the FEMA AEP flood maps do not necessarily represent complete ground truth, such a comparison is the best option given the data challenge. A similar approach has been utilized by several previous studies in the evaluation of large-scale flood inundation (Alfieri et al., 2014; Wing et al., 2017; Zheng et al., 2018; Gangrade et al., 2019).
To derive the 1 % AEP flood map using TRITON, the ensemble-based approach used by Gangrade et al. (2019) was followed. The assessment started by preparing the streamflow hydrographs used to construct the 1 % AEP flood map. The 1981–2012 annual maximum peak events and their corresponding 10 d streamflow hydrographs were extracted from the control simulation. These streamflow hydrographs were then proportionally rescaled to match the 1 % AEP peak discharge estimated at the watershed outlet (Fig. 1), following the frequency analysis procedures outlined in Bulletin 17C (England et al., 2019). The streamflow hydrographs from control simulations were used for the peak discharge frequency analysis.
The results reported in the sensitivity analysis were also used to help
identify suitable TRITON parameters. In addition to streamflow hydrographs,
TRITON requires DEM, initial water depth, and Manning's
Three evaluation metrics, including fit, omission, and commission (Kalyanapu et al., 2011) were used to quantify the differences between the modeled and reference flood map. The measure of fit determines the degree of relationship, while the omission and commission statistically compare the simulated and reference FEMA flood maps (Kalyanapu et al., 2011). The comparison between the simulated maximum inundation and the corresponding 1 % AEP FEMA flood map showed 80.65 % fit, 5.52 % commission, and 15.36 % omission (Fig. 5), demonstrating that the TRITON could reasonably estimate flood inundation extent and depths in the CRW. The computational efficiency of TRITON can further support ensemble inundation modeling to provide additional variability information that cannot be provided by the conventional deterministic flood map.
Comparison of simulated maximum flood extent with the corresponding FEMA 1 % AEP flood map for the Conasauga River watershed. Background layer source: © OpenStreetMap contributors 2021. Distributed under a Creative Commons BY-SA License.
Although we have obtained satisfactory model performance for the purpose of our study, the flood model implementation has some limitations that may be enhanced in future studies. They include the following.
Spatially varying Manning's Apart from changes in future runoff and streamflow, projections of future
LULC and its corresponding surface roughness can be considered to understand
the broader impacts due to environment change. In this study, we corrected DEM bias along the river channel cells by
simplified bathymetry correction factors. More sophisticated bathymetric
configuration (i.e., channel shape and sinuosity) can be considered to
better represent channel conveyance. The current TRITON model does not provide capability to route local runoff
and external inflows through stormwater drainage systems. Coupling with
additional stormwater drainage models can be a potential future direction. Hydraulic and civil structures such as bridges, culverts, and weirs have not
been included since TRITON does not provide for the modeling of such
components. This can affect the accuracy of the flood depths, velocities,
and flood extents around these structures.
In this section, the projected changes in flood regime were calculated using the flooded area from the baseline and future simulations for each ensemble member. Figure 6 illustrates the box-and-whisker plots for each of the 11 dynamically downscaled GCMs. Given the small sample size in each distribution (40 compared to 440 in Fig. 2), the whiskers extend the largest and smallest data points with no outlier detection. For 9 out of the 11 downscaled climate models, the mean of 40 flood inundation showed an increase in the floodplain area in the future period. In terms of the 75th percentile and maximum, 10 out of 11 models showed an increase in the floodplain area. The distributions of maximum future inundation of four models are found to be statistically different than their baseline distributions at a 5 % significance level. Note that the spread in the future period is generally larger than the spread in the baseline period, suggesting an increase in the hydrologic variability in the future period. Also, while the results from different models were generally consistent, some inter-model differences were noted, which highlight the need of a multi-model framework to capture the uncertainty in the future climate projections. The multi-model approach provides a range of possible flood inundation extents, which is critical for floodplain management decision making. The potential increase in the floodplain area also demonstrates the importance of incorporating climate change projections in the floodplain management regulations.
A summary of simulated maximum flood inundation extents obtained
from the baseline and future scenarios. The mean flooded area values are
shown by
Figure 7 shows the relationship between the 440 flooded area values (across
11 downscaled GCMs) and their corresponding peak streamflow at the watershed
outlet, for both the baseline and future periods. Overall, both results
(Fig. 7a and b) exhibit strong nonlinear relationships with high
Relationship between floodplain areas and peak streamflow values
at the watershed outlet for
Figure 8 shows the event-based flood inundation frequency curves and their
corresponding 95 % confidence intervals in both the baseline and future
periods, for which each frequency curve was derived using an ensemble of 440 years of data. The use of long-term data helped reduce the uncertainty and
add more confidence in the evaluation of the lower AEP estimates. This type
of assessment cannot be achieved using only historic streamflow
observations, for which the limited records present a major challenge for
lower AEP estimates. For most of the exceedance probabilities, the flooded
areas projected an increase in the inundation areas in the future period
when compared to the baseline period. The 1 % AEP flood shows an
A summary of flood inundation frequency curves for the baseline and future periods.
Projected change (future minus baseline period) in flood depth
frequency maps for
The grid-based flood depth frequency results at 0.5 %, 1 %, 2 %, and
4 % AEP levels are illustrated in Fig. 9. In each panel, the projected
change (i.e., future minus baseline) at each grid is shown. The
corresponding histogram across the entire study area is presented in Fig. 10. As mentioned in Sect. 2.4, the LP3 distribution was used for frequency
analysis. In order to understand the suitability of LP3, we also conducted a
comparative analysis to test an alternative log-normal (LN) distribution. By
using the Anderson–Darling (Anderson and Darling, 1952) goodness-of-fit test
(
Histograms for the future changes (2011–2050) in the flood depth
relative to the baseline period (1966–2005) for
Based on the comparisons in Fig. 10, it is estimated that the flood depth
values at
Figure 11a shows the box-and-whisker plot for the distributions of maximum
flood depth values extracted at the substation location across all the
baseline and future simulations, assuming that no flood protection measures
were adopted (mitigation scenario 1). Of the 44 substations, 5 substations
could have been affected during the baseline period, while 7 substations are
projected to be affected during the future period (Fig. 11a). Increases
are indicated not only for the number of affected substations but also for
flood inundation depth values in the projected future climate. Overall, the
mean of the ensemble flood depth values shows an
A summary of maximum flood depths for substations that were
affected in the baseline and/or future periods
A summary of maximum inundation durations for substations that
were affected in the baseline and/or future periods
The maximum inundation durations at the affected substations are summarized in Fig. 12a (mitigation scenario 1) and b (mitigation scenario 2). For both mitigation scenarios and all affected substations, ensemble mean inundation durations exhibited an increase under future climate conditions. This increase in inundation duration would probably render substations out of service for longer periods of time by making it difficult to repair damaged substation equipment and restore grid services to customers. The potential hazards and consequences may also extend to critical facilities that are supplied by the affected substations. Similar to results presented in the previous sections, these results demonstrate the need for improving existing flood mitigation measures by incorporating the trends and uncertainties that originate from climate change. The vulnerability analysis approach presented in this study will better equip floodplain managers to identify the most vulnerable substations and to recommend suitable adaptation measures, while allocating resources efficiently.
This paper applies an integrated modeling framework to evaluate climate
change impacts on flood regime, floodplain protection standards, and
electricity infrastructures across the Conasauga River watershed in the
southeastern United States. Building on the ensemble concept used by
Gangrade et al. (2019) for PMF-scale inundation modeling (AEP
Although future climate conditions are uncertain, our results demonstrate the needs for (1) consideration of climate change in the floodplain management regulations, (2) improvements in the conventional deterministic flood delineation approach through the inclusion of probabilistic or ensemble-based methods, and (3) improvements in the existing flood protection measures for critical electricity infrastructures through enhanced hydro-meteorologic modeling capacities. In particular, rapidly advanced high-performance computing capabilities have enabled the incorporation of computationally intensive 2D hydraulics modeling in the ensemble-based hydroclimate impact assessment. While the computational cost demonstrated in this study may still seem steep, in the current speed of technology advancement, we will soon be able to implement such a computationally intensive assessment for wide applications. The approach presented in this study can be used by floodplain managers to develop flood depth frequency maps and to identify the most vulnerable electric substations.
The data that support the findings of this study are openly available in the figshare repository at the following URLs:
TTD, AJK, SCK, SG, and MMH developed the concept for the paper and designed the methodology. TTD performed all the simulations required for the study with feedback from all the co-authors. MBS, SG, and MMH focused on programming, software development, and testing of existing code components. MA and MMH provided access to supercomputing machine hours on ORNL's SUMMIT and RHEA computers. The manuscript was edited by TTD with inputs from the co-authors.
The authors declare that they have no conflict of interest.
This study was supported by the US Air Force Numerical Weather Modeling Program. Tigstu T. Dullo, M. Bulbul Sharif, Alfred J. Kalyanapu, and Sudershan Gangrade also acknowledge support by the Center of Management, Utilization, and Protection of Water Resources at Tennessee Technological University. Some portion of the project was funded by UT-Battelle subcontract no. 4000164401. The research used resources of the Oak Ridge Leadership Computing Facility at Oak Ridge National Laboratory. The input data sets are cited throughout the paper, as appropriate.
This research has been supported by the US Air Force Numerical Weather Modeling Program.
This paper was edited by David J. Peres and reviewed by two anonymous referees.