Abstract
Floods have been occurring with increasing frequency, leading to damage to communities worldwide. These impacts are expected to continue to rise due to increases in the intensity of extreme rainfall. Global climate model (GCM) output, while imperfect in reproducing daily rainfall, is the only practical source of future projections of extreme rainfall intensification. This article presents a practical method for translating GCM precipitation output into usable outputs for stormwater and flood management planning at a regional or local level. The method estimates the impact of extreme storm intensification on riverine flooding using available runoff estimates from GCM precipitation and variable infiltration capacity models, focusing on changes in elevation and frequency due to climate change. It allows communities and utilities to obtain a screening-level estimate of climate change impacts to peak discharge rate statistics without conducting hydrologic modeling. This article outlines the method, its implementation for the 48 contiguous states of the United States, and an example calculation for a river in the eastern United States. Changes in extreme storm runoff intensity vary significantly by region, but much of the United States is projected to see increases of 25 and 50% by 2060 and 2090, respectively, for the RCP8.5 scenario.
HIGHLIGHTS
Provides estimates of percent increases in runoff and precipitation by decade from 2020 to 2090 for the United States.
Outlines a method to use existing hydraulic models to estimate changes future flood elevations based on the projected increases to runoff.
Future increases in runoff will be significantly larger than the more-often-studied percent increases in precipitation.
By the end of century, a significant fraction of the southern and eastern portions of the United States can expect runoff increases of 30% or more for RCP4.5 and 50% or more for RCP8.5.
Graphical Abstract
INTRODUCTION
In recent years, floods have been occurring with an increasing frequency. People around the world are being displaced from their homes by flooding each year, while damages from flood events run into the hundreds of billions of US dollars in direct asset losses annually (Gabriels et al. 2022; Maiwald et al. 2022; Taguchi et al. 2022; Whitehurst et al. 2022). These impacts are expected to continue to rise in the future due to expected increases in precipitation intensity for short-term extreme rainfall events. There is increasing interest in improving the understanding of future flood hazards in a changing climate (Judi et al. 2018; Gangrade et al. 2020). Flood risk management can be improved by better estimates of future flood hazards that are based on the best current understanding of the future climate system (Alfonso et al. 2016; Srikrishnan et al. 2019; Zarekarizi et al. 2020).
For many communities, estimating changes to either urban or riverine flood risk is a challenge as many of the available approaches require technical sophistication and modeling. In the United States, the Federal Emergency Management Agency (FEMA) is responsible for producing Flood Insurance Rate Maps and reports specifying the boundaries of zones deemed to be most vulnerable to fluvial and coastal flooding (FEMA 2019). The FEMA flood maps and reports can provide very useful information on existing flood risk but have limitations (Pappenberger et al. 2005, 2006; Bales & Wagner 2009). These products assume a stationary flood-event time series, an assumption that is often invalidated due to a variety of changes, including changes in land cover, improvements in urban stormwater infrastructure (Villarini et al. 2009), and above all, climate change.
One of the most important impacts of a future warmer climate is the projected increase in the frequency and the intensity of extreme rainfall events. IPCC (2021) concluded that the frequency and intensity of heavy precipitation events have likely increased at the global scale over most land regions with good observational coverage. Clear observational data support increases in North America, Europe, and Asia, confirming what the climate model projections have indicated. However, the physical mechanisms linking climate change and extreme storms can result in variable responses, making detection and attribution of trends in extreme storm characteristics difficult. Projecting changes in severe storms is also challenging because of global climate model (GCM) limits on the grid size that make it hard to capture and represent the small-scale, highly local physics of extreme storms.
The analysis of future extreme storms can be based directly on extending trends in historical observations (e.g., Cheng et al. 2014; Ganguli & Coulibaly 2017) or can be based on historical observations to calibrate or downscale GCM projections of precipitation with the use of nonstationary statistics (Lima et al. 2016). The usual intent is to update design storms or intensity–duration–frequency (IDF) curves to account for nonstationarity and plausible future changes in rainfall. Statistical methods often focus on key parameters and vary those variables known to have a relationship with rainfall (Cheng & AghaKouchak 2014; Agilan & Umamahesh 2016). This approach does not generally directly incorporate future projections of precipitation but instead focuses on better characterizing the time-dependent variability in rainfall observed in the historic record. Another approach estimates future precipitation from GCMs by modifying the time series used to generate IDF curve values (Maimone et al. 2019).
For most regions of the world, downscaled GCM output, imperfect as it is in reproducing daily rainfall, is the only practical source of future projections of extreme rainfall intensification. However, estimating the intensification of rainfall extremes from GCM output is complex. Some challenges are as follows:
Estimates vary by rainfall extreme frequency and duration and are influenced by topography, land use, and proximity to large water bodies.
Estimates are related to temperature increases, but some studies indicate that longer return period events could lead to larger increases per degree Celsius than the Clausius–Clapeyron assumption (sometimes referred to as super Clausius–Clapeyron scaling).
The intensification of extreme rainfall is now established and documented at the daily scale, but it is less certain what occurs at the sub-daily scale.
There are many indications that rainfall scaling may also increase as a function of duration, such that shorter duration events will likely see the largest rainfall increases in a warmer climate.
Despite these challenges, the growing realization that climate change is causing a great deal of damage requires that practical methods of projecting extreme storm intensification are needed as a basic first screening step in accounting for climate change in riverine flooding.
A PRACTICAL, INEXPENSIVE APPROACH FOR RIVERINE FLOODING
The development of the method presented in this article arose from the need to address the challenge related to estimating changing flood impacts due to climate change. There is an inherent difficulty in translating unrealistic daily GCM precipitation output into usable outputs for stormwater and flood management planning on a regional or local level. GCM daily precipitation output generally suffers from what is often referred to as the ‘drizzle effect’. The simulated daily rainfall events are usually less intense and more frequent than what is actually occurring. This combination of overestimation of the number of daily rainfall events and underestimation of intensity results in relatively accurate total rainfall on a monthly, seasonal, and annual basis. However, in dealing with extreme events and flooding caused by storms of 24 h or less, the daily GCM output is unrealistic. During the course of developing our approach, a second common misconception came to light. We realized that projected change in runoff resulting from extreme precipitation is a better measure of flood potential than projected change in precipitation. The required riverine or urban flood modeling needed to use extreme precipitation estimates for flooding projections is often beyond the reach of many communities, yet relying on precipitation changes may result in a significant underestimation of flood potential. Applying GCM output to estimate increases in precipitation intensity has been successfully done on an individual location basis (Trenberth 2011; Maimone et al. 2019; Maimone & Malter 2020). Doing so across the entire United States using runoff estimates, with results that are specific to each region, has not been previously accomplished.
Assessing changes to potential flooding due to extreme rainfall events in an urban setting can often suffice with estimates of changing precipitation due to the high percentage of impervious cover. For riverine flooding, changing runoff is more appropriate. For both applications, one of the more practical and widely applied approaches is to develop delta change factors (DCFs) that relate the future precipitation intensity to the current precipitation intensity (Padulano et al. 2019). It relies on changes to readily available daily downscaled precipitation output from GCMs to provide estimates of future trends under different greenhouse gas concentration scenarios.
The method described herein illustrates a stepwise process for estimating the climate change-related shift in precipitation and runoff. The use of changes in runoff as opposed to changes in precipitation is critical for riverine flooding assessments as will be shown later. For urban flood applications, changes in precipitation can be used to provide input to hydrologic/hydraulic models. For riverine flood applications, changes in runoff can be used in our proposed method to estimate future riverine flood elevations. Certain ground rules were established in developing the method for riverine flooding to ensure it could be conducted with a reasonable level of effort. First, the proposed riverine method needs to provide a screening-level estimate of climate change impacts to peak discharge rate statistics without conducting hydrologic modeling of each stream. It was understood that remapping the various floodplain frequency-based maps would be a prohibitively large effort and was not within the scope of this analysis. In addition, the analysis must only use existing, publicly available data sources and cover the entire contiguous United States.
readily available daily GCM precipitation output for urban applications, or
estimates of runoff based on the same daily precipitation output.
The runoff estimates used in this study are available for the contiguous United States from the website ‘Downscaled CMIP3 and CMIP5 Climate and Hydrology Projections’: (https://gdo-dcp.ucllnl.org/downscaled_cmip_projections/).
Though not the focus of this article, the boxes on the right show that the same basic technique for calculating runoff-based DCFs can be applied to calculating precipitation-based DCFs. Precipitation-based DCFs can be used to alter design storms or IDF curves for use in urban flooding applications.
PRECIPITATION AND RUNOFF DELTA CHANGE FACTORS
It is important to understand that there are two types of DCFs. One relates current precipitation to future precipitation, and the other relates current stormwater runoff in a watershed to future runoff in the watershed. Runoff-based DCFs are likely to be considerably higher than precipitation-based DCFs. Thus, using DCFs based on precipitation to assess riverine flooding might significantly underestimate future flooding. To illustrate the difference between runoff-based and precipitation-based DCFs, a simple example is shown in Table 1 for a number of United States Geological Survey (USGS) hydrologic unit code (HUC4) watersheds. For each watershed, an estimate of both precipitation-based and runoff-based DCFs was made using the 20 largest storms for two future periods, 2050 and 2090, compared to a baseline period of 1986–2005. The DCFs for runoff are calculated from daily runoff output from a variable infiltration capacity (VIC) model whose input is the same precipitation values used to calculate the precipitation-based DCFs. The relationship between DCFs for precipitation and DCFs for runoff depend to a degree on the land use of the watershed and the intensity of the storm, along with antecedent conditions. For extreme events, the absolute change in runoff and precipitation is often quite similar (ground is fully saturated or impervious). Because the runoff total is always less than the precipitation total over the watershed areas, the percent change for runoff can be significantly higher. This can be seen in Table 1, where in many watersheds, runoff-based DCFs are twice as high as the related precipitation-based DCFs.
HUC4 . | HUC4 name . | Delta change factors . | |||
---|---|---|---|---|---|
Calculated for RCP8.5 . | |||||
Precipitation . | Runoff . | ||||
2050 . | 2090 . | 2050 . | 2090 . | ||
0107 | Merrimack | 13.5% | 25.2% | 24.7% | 47.6% |
0204 | Delaware – Mid Atlantic Coastal | 9.5% | 20.0% | 16.2% | 35.1% |
0314 | Choctawhatchee – Escambia | 7.0% | 15.2% | 11.7% | 25.9% |
0415 | NE Lake Ontario – St Lawrence | 12.8% | 22.5% | 24.3% | 43.8% |
0510 | Kentucky – Licking | 10.3% | 23.3% | 19.1% | 44.9% |
0602 | Middle Tennesee – Hiwassee | 16.5% | 28.0% | 30.5% | 53.5% |
0705 | Chippewa | 10.9% | 22.9% | 19.1% | 41.5% |
0803 | Lower Mississippi – Yazoo | 12.5% | 21.4% | 19.7% | 34.1% |
0902 | Red | 7.2% | 15.0% | 12.4% | 26.2% |
1003 | Missouri – Marias | 9.4% | 18.4% | 17.9% | 36.2% |
1104 | Upper Cimarron | 4.4% | 11.8% | 8.3% | 22.7% |
1206 | Middle Brazos | 7.9% | 14.8% | 13.7% | 26.1% |
1305 | Rio Grand Closed Basins | 8.2% | 12.7% | 16.3% | 25.7% |
1408 | San Juan | 7.8% | 15.0% | 16.0% | 31.8% |
1505 | Middle Gila | 12.6% | 18.6% | 25.1% | 37.8% |
1606 | Central Nevada Desert Basin | 15.0% | 28.3% | 31.5% | 62.6% |
1708 | Lower Columbia | 19.1% | 30.9% | 35.5% | 59.0% |
1802 | Sacramento | 11.6% | 24.7% | 21.2% | 46.6% |
HUC4 . | HUC4 name . | Delta change factors . | |||
---|---|---|---|---|---|
Calculated for RCP8.5 . | |||||
Precipitation . | Runoff . | ||||
2050 . | 2090 . | 2050 . | 2090 . | ||
0107 | Merrimack | 13.5% | 25.2% | 24.7% | 47.6% |
0204 | Delaware – Mid Atlantic Coastal | 9.5% | 20.0% | 16.2% | 35.1% |
0314 | Choctawhatchee – Escambia | 7.0% | 15.2% | 11.7% | 25.9% |
0415 | NE Lake Ontario – St Lawrence | 12.8% | 22.5% | 24.3% | 43.8% |
0510 | Kentucky – Licking | 10.3% | 23.3% | 19.1% | 44.9% |
0602 | Middle Tennesee – Hiwassee | 16.5% | 28.0% | 30.5% | 53.5% |
0705 | Chippewa | 10.9% | 22.9% | 19.1% | 41.5% |
0803 | Lower Mississippi – Yazoo | 12.5% | 21.4% | 19.7% | 34.1% |
0902 | Red | 7.2% | 15.0% | 12.4% | 26.2% |
1003 | Missouri – Marias | 9.4% | 18.4% | 17.9% | 36.2% |
1104 | Upper Cimarron | 4.4% | 11.8% | 8.3% | 22.7% |
1206 | Middle Brazos | 7.9% | 14.8% | 13.7% | 26.1% |
1305 | Rio Grand Closed Basins | 8.2% | 12.7% | 16.3% | 25.7% |
1408 | San Juan | 7.8% | 15.0% | 16.0% | 31.8% |
1505 | Middle Gila | 12.6% | 18.6% | 25.1% | 37.8% |
1606 | Central Nevada Desert Basin | 15.0% | 28.3% | 31.5% | 62.6% |
1708 | Lower Columbia | 19.1% | 30.9% | 35.5% | 59.0% |
1802 | Sacramento | 11.6% | 24.7% | 21.2% | 46.6% |
A few things stand out from the results shown in Table 1.
Mid-century increases in precipitation intensity for extreme storms range from a low of 4–5%, and as high as 19%. These indicate the variability of GCM projections around the country and highlight the importance of the need for localized projections.
Mid-century projections for increases in runoff from extreme storms range from a low of around 8 to over 35%.
The difference between precipitation DCFs and runoff DCFs can be quite significant, ranging from a factor of 1.6–2.2. For the watershed sizes included in this study, this range was fairly consistent as the watersheds all contained a mixture of urban and more rural land use. The more urbanized the watershed, the smaller this difference will be.
DEVELOPING URBAN AND RIVERINE DELTA CHANGE FACTORS
In brief, our approach to estimating future changes in riverine flood elevations can be summarized in two steps:
apply available daily GCM runoff estimates for the United States to estimate DCFs for each decade from 2020 to 2090 for precipitation or runoff;
use the calculated runoff DCFs and FEMA's existing floodplain studies to estimate the shift in water surface elevations related to the 10-, 50-, 100-, and 500-year discharge rates that will occur due to climate change impacts.
While we focus on the development of runoff DCFs in this method section, precipitation DCFs can also be calculated in a similar manner using downscaled daily precipitation values from GCMs for use in urban flood modeling applications.
To estimate runoff DCFs, our method uses daily GCM runoff values for the United States based on the VIC model (Hamman et al. 2018). Because these outputs are freely available for download across the contiguous United States and are directly linked to downscaled GCM precipitation data, no additional hydrologic modeling is required to obtain runoff values. We followed three basic steps to calculate these runoff DCFs:
select a watershed database and an appropriate watershed scale,
select GCM models and download downscaled GCM daily VIC model runoff data output, and
create DCF lookup tables based on the downscaled runoff data for both representative concentration pathway (RCP)4.5 and RCP8.5 and for each decade from 2020 through 2090.
Selection of watershed database
Selection of GCMs and downscaled data
Under the World Climate Research Program (WCRP), the Working Group on Coupled Modelling (WGCM) established the Coupled Model Intercomparison Project as a standard experimental protocol for studying the output of coupled atmosphere-ocean general circulation models. Under CMIP5, there are more than 30 GCMs from modeling centers around the globe that are used as a starting point and downscaled for regional or local climate change modeling. Downscaling of each model provides daily precipitation output, but results can differ widely between models. This is referred to as model uncertainty, and there is no consensus as to whether one model is superior to another. To account for the variation in future predictions, the climate community suggests the use of multiple models, as it provides additional and more reliable information than one single model (Knutti 2008). Localized constructed analog (LOCA) daily precipitation and VIC model runoff estimates for RCP8.5 and RCP4.5 are available from the downscaled CMIP5 Climate and Hydrology Projections website (https://gdo-dcp.ucllnl.org/downscaled_cmip_projections/).
Output is available for 32 GCMs covering the contiguous United States, and all were used to develop both precipitation-based and runoff-based DCFs. Daily data for all GCM grid cells for each of the 202 HUC4 subregions were downloaded for the periods 1986–2005 and 2010–2099 for use as either a reference period or a future projection period. In total, 311,464 grid cells were assigned to a relevant HUC4 subregion based on the location.
Development of delta change factors
The method to calculate DCFs for the chosen subregions shown in Figure 2 was developed via extensive sensitivity testing. A full description of the sensitivity testing goes beyond the scope of this article. Testing included studying changes to results due to averaging approaches and sequences, number of extreme events per time period, and the length of the averaging period. This testing led to decisions related to the selection of duration periods, reference periods, and the use of the mean or median for averaging. The sequence of averaging was also tested, as averaging occurred across daily runoff, the 32 GCMs, and the thousands of grid cells to reach a single DCF per watershed.
The development of runoff-based DCFs was carried out for both the RCP4.5 and RCP8.5 scenarios. Note that the following steps focus on the runoff DCFs, but the only difference in developing the precipitation DCFs is the replacement of daily runoff values with daily precipitation values.
A mean runoff amount was determined for each grid cell. The calculated value entailed the 20 largest runoff days from the VIC model output in the 20-year reference period as well as each subsequent 20-year period. This step established that each grid cell had 288 different mean runoff values corresponding to the reference period and the eight decades spanning 2020–2090 (including 2020 and 2090) and the 32 GCM-based VIC model daily runoff values used (32 × 9 = 288).
The mean runoff from the reference period of each grid cell was compared with each future period to calculate a DCF for each of the 32 VIC model results. This represents the percentage increase in runoff for each period's extreme runoff events compared with the reference period.
The DCFs were spatially summarized using the median of all the grid cells within the watershed to obtain a DCF value for each decade and the VIC model. This resulted in each watershed having 256 DCFs corresponding to the eight decades spanning 2020–2090 and the 32 VIC models.
The mean DCF of the GCM/VIC model outputs was calculated to obtain a single DCF for the HUC4 watershed for each decade. The result was that each HUC4 subregion had eight DCFs corresponding to the eight decades spanning 2020–2090.
Estimates of riverine elevation and probability shift
The last steps in the process relate the runoff-based DCFs to riverine flooding. The runoff-based DCFs provide an estimate of projected changes in runoff within the watershed for each of the coming decades though 2090. To meet the study objective, these changes in runoff must be used to project changes in flood elevation without developing a watershed specific hydrologic model. FEMA Flood Insurance Study (FIS) reports provide both discharge rates and associated water surface elevations at cross sections along rivers and creeks for much of the United States. These values were developed through a nationwide modeling effort and represent the best available estimates of current conditions for many locations in the United States.
To calculate the changes in flood elevation for a location, four FEMA-FIS flood elevations and the four associated discharge rates for the 10-, 50-, 100-, and 500-year recurrence intervals can be used. These provide the link between flow and flood elevation that represents the underlying hydraulic modeling in the FIS. Along with the decade-by-decade DCFs calculated for each watershed, future flood flows and frequencies can be calculated using the following five steps.
- 1.
Locate the nearest cross section to the desired location along the river of interest.
- 2.
Access the 10-, 50-, 100-, and 500-year flood elevations shown in the FIS report as well as the associated river flows used in the model to calculate the river flood stage.
- 3.
Fit a curve to the four data points relating the flow to the elevation using an appropriate curve fit equation line (e.g., linear, exponential).
- 4.
Use the DCFs by decade to increase the FEMA flows for the selected RCP.
- 5.
Calculate the decade-by-decade projected flood elevations associated with the increased flood flows using the estimated curve fit equation but staying within the model elevation domain (do not extrapolate beyond the 500-year elevation).
RESULTS AND EXAMPLES
Figures 3 and 4 indicate that the GCMs show a considerable range of expected increases in the largest storm events throughout the United States. The largest increases occur in the East, with the more arid regions showing smaller projected changes. Increases of over 30–40% for RCP4.5 and 50% for RCP8.5 start to appear by mid-century. By the end of the century, a significant portion of the southern and eastern portions of the United States can expect increases of 30% or more for RCP4.5 and 50% or more for RCP8.5.
Examples of delta change factors
For two example subregions, Table 2 lists the runoff-based DCF for each decade as the percentage difference in runoff during the 1986–2005 reference period and each 20-year period centered on the decades between 2020 and 2090. Note that the 2020 DCF for the two subregions already shows a significant percent increase from the reference period. This indicates that current conditions have already been impacted by climate change. This is common for many areas of the country.
HUC4 . | HUC4 name . | Decade . | RCP4.5 change factor . | RCP8.5 change factor . |
---|---|---|---|---|
0315 | Alabama | 2020 | 19.5% | 26.7% |
2030 | 21.1% | 28.0% | ||
2040 | 25.6% | 35.0% | ||
2050 | 31.2% | 40.4% | ||
2060 | 35.1% | 44.4% | ||
2070 | 32.8% | 48.1% | ||
2080 | 35.8% | 53.0% | ||
2090 | 38.9% | 59.3% | ||
1705 | Middle Snake | 2020 | 7.2% | 7.5% |
2030 | 7.8% | 8.2% | ||
2040 | 9.6% | 7.3% | ||
2050 | 7.0% | 8.7% | ||
2060 | 10.0% | 10.9% | ||
2070 | 9.1% | 13.6% | ||
2080 | 8.8% | 18.3% | ||
2090 | 11.3% | 24.3% |
HUC4 . | HUC4 name . | Decade . | RCP4.5 change factor . | RCP8.5 change factor . |
---|---|---|---|---|
0315 | Alabama | 2020 | 19.5% | 26.7% |
2030 | 21.1% | 28.0% | ||
2040 | 25.6% | 35.0% | ||
2050 | 31.2% | 40.4% | ||
2060 | 35.1% | 44.4% | ||
2070 | 32.8% | 48.1% | ||
2080 | 35.8% | 53.0% | ||
2090 | 38.9% | 59.3% | ||
1705 | Middle Snake | 2020 | 7.2% | 7.5% |
2030 | 7.8% | 8.2% | ||
2040 | 9.6% | 7.3% | ||
2050 | 7.0% | 8.7% | ||
2060 | 10.0% | 10.9% | ||
2070 | 9.1% | 13.6% | ||
2080 | 8.8% | 18.3% | ||
2090 | 11.3% | 24.3% |
The two examples were selected to show an important point. Not only can the magnitude of intensification be very different in different areas of the country but also the timing of intensification. For the Middle Snake watershed in Idaho and Oregon, the runoff DCFs by 2090 range from 11 to 24% for RCP 4.5 and 8.5, respectively. Interesting in this watershed is the fact that intensification is not projected to occur until mid-century. In contrast, the Alabama watershed has much higher DCFs of approximately 40–60% by 2090, with intensification fairly steady throughout the entire period between 2020 and 2090.
Example of flood elevation projection
An example of using the DCFs to project future changes to flood elevations can best illustrate the steps to develop changes to riverine flood elevations discussed in the section ‘Estimates of Riverine Elevation and Probability Shift’. A FEMA-FIS report provided both flows and associated flood elevations at a cross section located along a river in the eastern United States (Steps 1 and 2). These are shown in Table 3.
Flow elevation from FEMA-FIS . | ||
---|---|---|
Return interval (years) . | Flow (m3/s) . | Elevation (m) . |
10 | 2,095 | 7.0 |
50 | 3,115 | 8.3 |
100 | 3,625 | 9.1 |
500 | 4,842 | 10.9 |
Flow elevation from FEMA-FIS . | ||
---|---|---|
Return interval (years) . | Flow (m3/s) . | Elevation (m) . |
10 | 2,095 | 7.0 |
50 | 3,115 | 8.3 |
100 | 3,625 | 9.1 |
500 | 4,842 | 10.9 |
Step 3 of the calculation method was carried out by taking the FEMA flows for each of the four return intervals and matching them to the associated flood elevations as shown in Table 3. These are then fitted to a linear equation with a resulting R2 value in excess of 99%. This is done to calculate shifts in elevation related to shifts in flow due to climate change based on the underlying FEMA model results.
In step 4, the DCFs that were developed for this river location are used to increase the FEMA flows for the 10-, 50-, 100-, and 500-year return intervals. In Table 4, the DCFs for RCP8.5 are shown in the second column for each decade between 2020 and 2090. Using the decade-by-decade DCFs, the FEMA flows shown in Table 4 for each of the four return intervals are increased by the decade-by-decade percentages represented by the DCFs, resulting in estimates of the 10-, 50-, 100-, and 500-year flows influenced by climate change.
In step 5, the adjusted future flows for each return interval are related to the associated flood elevations using the fitted linear equation (Figure 7). This makes use of the FEMA watershed model results estimated from the flow and elevation data provided in the FIS report to estimate the higher flood elevations for each of the return intervals. Table 4 provides the results for RCP8.5.
The FEMA models provide the flow–elevation relationship up to the FEMA estimated 500-year return interval, which in this example is 10.9 m as shown in Table 3. Because no FEMA model results are available beyond 10.9 m, it is not known whether the same relationship will hold as the channel configuration above a 11 m elevation is likely to differ. For this reason, in Table 4, the extrapolated flood elevations above 11 m are shaded to indicate that these values are likely to have an unknown accuracy. All the unshaded elevations are within the FEMA model domain and are accurate to the degree that the FEMA model was calibrated. As shown, all the future 500-year elevations are extrapolations.
DISCUSSION
There are now several regional and country-wide estimates of extreme storm intensification. One such resource is provided as a part of the USEPA Climate Resilience Evaluation and Awareness Tool (CREAT) (https://epa.maps.arcgis.com/apps/MapSeries/index.html?appid=3805293158d54846a29f750d63c6890e).
This tool used a combination of 22 GCMs and downscaled daily precipitation data to develop precipitation DCFs for extreme storms for the contiguous United States. The approach used a 20-year averaging period much like our approach. The focus was to provide DCFs for a 24-h duration storm for a variety of return intervals. Results are available between 5- and 100-year return intervals for two periods: 20-year averages around the years 2035 and 2060.
There are a number of significant differences between our results and the USEPA results.
We provide decadal results through 2090, whereas CREAT focuses on only two near and mid-century periods.
Our focus is to provide runoff-based DCFs. CREAT provides only precipitation-based DCFs. As noted, runoff-based DCFs will be significantly higher than precipitation-based DCFs.
CREAT provides two values based on selecting the five GCMs that show the least storm intensity increases and those that show the greatest intensity increases. Our approach uses all GCMs but also focuses on the highest estimates all 32 GCMs.
CREAT gives a range, and our approach gives a single value.
Our results provide more flexibility in application than the CREAT site. This is illustrated in the example shown in Table 4. The example illustrates an application to a particular location in the eastern United States. The table provides DCFs that indicate that current extreme storms are already almost 25% higher than the reference period of the latter part of the last century. The DCFs shown in the table suggest that this increase has flattened and will start to increase once again in the second half of this century. This process of intensification can vary by location. In some subregions, precipitation- and runoff-based GCM output can show local anomalies that are not evident in the temperature projections at the same location. Because GCM precipitation and runoff outputs are much more variable and less stable than GCM temperature output, we are currently working on temperature-based approaches to develop DCFs to supplement the runoff-based approach discussed in this article. Despite these local anomalies in the runoff-based DCFs, the results are clearly important for planners and designers to begin to account for these significant changes in extreme storm intensities that are, even today, being seen around the world.
Method limitations
The approach developed in this article is specifically designed for riverine flood planning purposes in those areas where more locally specific climate projections and hydrologic–hydraulic models are not available. Runoff estimates are calculated by the VIC model (Liang et al. 1994). VIC is a grid-based model used for coupled land surface model – global circulation model applications. It is a two-layer, soil vegetation model where each grid cell is independently simulated from precipitation on a daily time step. There is no channel flow, subsurface flow, or recharge to soil from rivers, and there is no routing of streamflow from one grid to another. It cannot simulate sewers or collection systems that might exist within a model cell. Because the grid cells are large (0.25° or roughly 27.5 km wide), the assumption is that the groundwater flow is small relative to surface flow and that lakes/wetlands that do not significantly impact the estimates are considered reasonable.
In addition to any limits imposed by the large-scale constraints of the VIC model runoff estimates, there are a number of other points to keep in mind when applying the approach.
The DCFs are based on GCM precipitation and VIC model results. It is well established that GCMs, and consequently VIC runoff estimates, underestimate extreme, convective storm events. This makes it difficult to assess whether the GCM changes to extreme storms are similarly underestimated.
These DCFs are calculated using an approach that selects only the largest events as simulated by the GCMs. These larger events are subject to considerable instability and randomness. Much of this is smoothed out by averaging across 32 GCMs, using 20-year averaging periods, and using the spatial mean. Nevertheless, some anomalous results still appear.
The DCFs, though varying by the watershed, are still representing relatively large areas. Runoff-based and even precipitation-based DCFs can vary quite significantly within each area depending on topography and land use. As such, they should be viewed as reasonable initial estimates for planning purposes but are not suitable for design.
The accuracy of the flood elevation changes related to changes in runoff is only as accurate as the models used in developing the FEMA flood plains.
In addition to the aforementioned limitations, there are certain data requirements that are necessary and might not be available in all locations.
The first part of the study requires that a grid of simulated runoff coupled to rain events is available. Such data are available for the continental United States (VIC model results) but are not generally available outside the continental United States for downscaled GCM daily output. Where no gridded runoff estimates are available related to the GCM precipitation output, only DCFs derived from precipitation are possible. These will probably underestimate the expected change in flood elevations as discussed in this article.
Applications of the method to assess changes in river flood elevations require that estimates of flood flow and flood elevation for a number of return intervals are available. FEMA provides these estimates at river cross sections for many watersheds within the United States, but not all. In many countries, similar studies provide flood flows and elevations for a variety of return intervals, and the same method can be applied. Where no such flow elevation results are available, then the application of DCFs to assess changes in flood elevation cannot be carried out without developing a hydraulic model of the river system.
All these results are based on CMIP5 simulations. CMIP6 is currently producing new model runs with updated climate scenarios. At present, VIC model results for CMIP6 are not available.
Future work could include an update of the method using CMIP6 results when available, as well as a comparison and discussion of CMIP5 and CMIP6 results.
Despite these caveats, these results should help to understand and explain the extreme storm events that are already occurring, and to help to reset our hydrologic assumptions in the face of current and projected climate change.
CONCLUSIONS
Flooding from extreme rain events is a growing problem throughout the world. For most communities, it has become clear that the current flood plain delineations, precipitation event recurrence probabilities, and associated flood elevations are not only out of date but are also subject to the continued change due to the climate change. As the results show, extreme rainfall events are likely to increase in intensity in almost all areas of the United States. This climate change-related phenomenon is likely to be similar in many areas of the globe. Changes in extreme rainfall intensity vary significantly by region but can ultimately result in extreme rainfall events that are 50% or more intense in many areas of the United States. There are many emerging statistical approaches to estimating expected changes to intense precipitation events due to the climate change. These approaches can provide insight into expected extreme event intensification. However, for most communities, initiating these studies represents a financial burden. A less-expensive and accessible approach to incorporating climate change in developing flood management plans is a critical need.
The approach described in this article is designed to provide reasonable initial estimates of projected changes to intense rainfall events and associated changes in runoff. It then provides insight into what those changes might mean for local flood elevations without the need for developing hydraulic models of the watershed. It makes use of existing, readily available data from GCMs, and applies it to the existing flood study information available in FEMA-FIS reports. The results, available by HUC4 watershed, are suitable for planning studies and initial risk estimates and can help to prepare for more detailed studies when adaptation measures are being designed. The approach can be easily updated as new GCM results become available or other methods of developing DCFs are developed.
As noted earlier, DCFs developed based on the GCM output for precipitation or runoff can be subject to local anomalies. One approach we are presently exploring is to use GCM temperature projections and various assumptions of atmospheric moisture holding capacity to assess changes in extreme precipitation. It is expected that this approach will provide more consistent and stable results than precipitation-based DCFs when used to develop runoff-based DCFs.
The implications of climate change for flooding are obvious and are being observed around the world today. The results of this study for the United States should be a wake-up call for what we might expect from these extreme rainfall events both now and in the future. Although designed as a planning level estimate, this method can provide any community with previously mapped riverine floodplains with estimates to help understand current and future flood conditions. The same approach applied to develop future changes to precipitation intensity of extreme events can also help in urban flood planning.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.