Projections of coastal flood hazards and flood potential for the U.S. Atlantic coast

All processes and methods are outlined in Nederhoff and others (2024); please refer to that for more information beyond the summary in this document. To generate time-series of forcings for coastal flooding models in order to map future coastal flooding hazards along the south Atlantic United States coast due to sea level rise and plausible future storm conditions that consider the changing climate, hurricanes, and natural variability, we gathered available atmospheric forcing data (specifically precipitation, sea-level pressure, and near-surface wind for this study) from CMIP6 Global Climate Models (GCM). At the time of this study, only products for Representative Concentration Pathway 8.5 for the projected time-period 2020-2050 were available and used. Output was gathered for specific High-Resolution Model Intercomparison Project (HighResMIP) experiments: HadGEM3-GC31, EC-Earth3P-HR, CNRM-CM6-1-HR, GFDL-CMC4C192 and CMCC-CM2-VHR4

We analyzed multi-model trends in future (2020-2050) tropical cyclone climatology depicted in GCMs throughout the study area (Nederhoff and others, 2024). This included detailed comparisons to historical runs in probability functions of tropical cyclone sea-level pressure, propagation speed and maximum wind speed throughout the study area, to highlight future changes in tropical cyclone characteristics by geographical position

We obtained Global Surge and Tide Model (GTSM) output (run for all aforementioned CMIP6 experiments’ sea-level pressure and wind) for nearshore water levels for projected period 2016-2050, and historical period (1976-2015). As described in Nederhoff and others (2024), we conducted initial comparisons of datasets and analysis of extreme water level changes, before preparing data for use in following process steps.

As described by Nederhoff and others (2024), we tested the Super-Fast Inundation of CoastS model (SFINCS; Leijnse and others, 2021) resolutions and computational efficiency. Determined running the SFINCS at 200 m spatial resolution, with sub-gridding was optimum for this study, providing balance between fast simulations and accuracy of coastal water levels (tested for Hurricane Florence,14 September 2018, with historical NOAA water levels). The study area was covered by three rectilinear SFINCS domains, aligned shore-normal for each respective area, with the offshore boundary as the nearshore GTSM output locations. Model boundaries extend outside the study area to encompass and include necessary hydrodynamics. Elevation for the SFINCS domains was extracted from the corresponding DEMs in the region and resampled from 1 meter resolution to the SFINCS model's computational grid. SFINCS simulations were run with soil infiltration rates derived using the Curve Number Method (U.S. Dept. of Agriculture, Soil Conservation Service, 1985) to capture absorption/run-off of precipitation in the model. Curve Numbers were derived using the National Land Classification Dataset (NLCD 2016) and the Digital General Soil Map of the United States (NRCS).

We conducted initial comparisons of WW3 data for projections (run with wind conditions for all aforementioned CMIP6 experiments) at the 15-20 m isobath and analysis of extreme nearshore wave changes, before preparing data for use in following process steps.

Hindcasted water levels were compared to NOAA tide station observations and were used to guide any necessary bias corrections (see the Nearshore water level, tide and non-tidal residual projections (2016-2050) and hindcasts (1979-2016) for the U.S. Atlantic coast, also available in this data release). Bias corrections were applied to the projected water levels. See Nederhoff and others (2024) for more details.

In collaboration with U.S. Army Corps of Engineers (USACE), we used a synthetic database available from Nadal-Caraballo and others (2020) of approximately 1,200 tropical cyclone events to establish a baseline of boundary conditions for tropical storms. As described in Nederhoff and others (2024), changes in tropical storm parameters, computed from the previous tropical cyclone analysis comparing GCM data for historical to future periods, were used to shift the hazard curves to represent future cyclone conditions and changes in frequency of occurrence and magnitude.

We derived future time-series data of river/fluvial discharge through the study area for 48 rivers, using the relationship between historical NLDAS precipitation and NWM reanalysis data and applying it to future GCM precipitation output (Nederhoff and others, 2024). The upstream watershed of each of the 48 rivers was identified from the network of river reach IDs used by the NWM (NFIE). Historical precipitation (1993-2018) over each individual watershed was used for each respective river. Future discharge was then estimated by applying future GCM precipitation data (2020-2050) over watersheds and using the established relationships between historical precipitation/pluvial rates and discharge. When no precipitation was projected in data, baseline river discharge rates (from NWM historical periods) were used. An additional river time series consisted solely of its historical baseline discharge, due to its watershed being too small for this process.

Using the GTSM output and computed wave setup, we identified extreme water levels along the open coast and associated fluvial inputs and precipitation for extreme coastal water elevation events. As described by Nederhoff and others (2024), the largest coastal storm events (from GTSM storm tide and wave setup) of each GCM were identified, equivalent to an average of the largest 5 storms per year. The overland flow model (SFINCS) was run for all anomalously high-water level events (top 150 from each contributing GCM, plus all tropical cyclone events from USACE) with each event’s commensurate GTSM coastal water levels, wave setup, SLR, point-source river discharge (at each river), and precipitation data fields included as forcing for the simulation. Additionally, all simulations were repeated for seven SLR scenarios: 0, 0.25, 0.50, 1.00, 1.50, 2.00 and 3.00 meters of SLR added compared to baseline water levels in the year 2005.

Detailed quality control was conducted for test outputs from the model system. After identifying initial sources of error, all simulations were rerun.

Return period (RP) statistics (1/20/100-year storm, or no storm/daily average conditions) were calculated per grid cell for each SLR scenario to yield a composited raster of water levels for each SLR and storm combination (Nederhoff and others, 2024). With each composited raster, by RP and SLR, a depth threshold of 5 cm (at native 200-m scale of SFINCS computational grid) was used to preserve legitimate flood projections in high-relief areas. Raster outputs were run through an iterative function (in Matlab) to identify cells connected to coastally driven flooding (such as, physically connected to contiguous coastal flood surface and ocean). For cells not connected to coastal flooding, output was labeled "ponding", to signify vulnerability to flood hazards driven by river discharge or precipitation. Water levels/elevations in each cell were then depth-differenced to underlying DEM data (sub-sampled to horizontal resolution of 10 m) to resolve fine-scale features in coastal flood hazards and ponding areas, as well as return corresponding water depth information. Water depths were only calculated for areas identified as coastal flooding (not ponding), as that was the focus of the study. Uncertainty was calculated as a sum of contributions, including DEM uncertainty (35 cm), projected vertical land motion (VLM) based on SLR (spatially variable per SLR scenario), and uncertainty with the model and model processes (spatially variable, derived from water level return-period curves at each grid point, dependent on scenario). This total uncertainty is applied to the final water elevation and extrapolated outward to depict the maximum and minimum potential flood area considering total uncertainty (labeled as ‘flood potential’). Water depths are accurate within these bounds.

Data from all domains were merged to make geoTIFFs of the originating rasters for each data layer (coastal flood hazard, ponding, and maximum/minimum flood potential). The geoTIFFs were exported as shapefiles from ArcMap for all combinations of seven SLRs (0, 0.25, 0.5, 1.0, 1.5, 2.0 and 3.0 m), 3 storms (1-year, 20-year, and 100-year return period coastal events), and the non-storm condition for a total of 28 scenarios. Final shapefiles were separated by state (Projections_FloodHazards_*STATE*.zip) for file-size considerations. Shapefiles are further organized by storm scenario (’RP’), with flood hazards and ponding under one directory, and flood potential in another directory. Shapefiles depicting the boundary of output for the study area are also included (FloodHazard_boundaries_*STATE*.zip). Despite obtaining the most recent DEM information available, some locations in Georgia and Florida had DEM data that was too coarse or unreliable for flood projection outputs, and these areas are outlined (FloodHaz_Areas_Not_Mapped_*STATE*). Data for North Carolina and South Carolina can be found in Barnard and others (2023). Other U.S. Atlantic coast data are included in this dataset.

Edits were made to correct spelling in author name. No data were changed. The metadata available from a harvester may supersede metadata bundled within a download file. Users are advised to compare the metadata date of this file to any similar file to ensure they are using the most recent version. (scochran@usgs.gov)

Metadata was modified to include updated doi# and full citation for Nederhoff and others (2024) Cross Reference. No data information was changed. The metadata available from a harvester may supersede metadata bundled within a download file. Users are advised to compare the metadata dates to determine which metadata file is most recent. (pcmsc_data@usgs.gov)