Predicting Coastal Change from Hurricanes

Transcription

Predicting Coastal Change from Hurricanes
Predicting Coastal Change from Hurricanes and Nor’easters: A Prototype GISbased Coastal Impact Assessment Model
By Brian H. Bossak
Dept. of Physics, Astronomy, and Geosciences
Valdosta State University 1
Abstract
Forecasting coastal changes as a result of hurricanes and nor’easters is still in
relative infancy. While models exist to predict estimated storm surge, and the resultant
engineering damage to structures, forecasting beach erosion and barrier island breach
occurrences remains a challenge. The U.S. Geological Survey accelerated efforts to
develop such models during the active North Atlantic hurricane season of 2004. This
model predicts coastal changes from storm events by utilizing Sallenger Storm Impact
Scale forecasts, and is incorporated within a customized ArcGIS project. The GIS-based
prototype model, termed the Coastal Impact Assessment Tool or CIAT, is revealed here,
as well as the results of a verification exercise.
Background
The Atlantic hurricane season of 2004 resulted in multiple tropical cyclone
landfalls along the southeastern U.S. coastline. Hurricanes made three direct landfalls in
Florida, one in Alabama, and one in South Carolina (Figure 1). Although much of the
popular media focused on the direct impact of the storms on the people and property
affected, as a result of the intense landfall activity, dramatic changes to shorelines took
place. In an extremely rare climatological event, hurricanes Frances and Jeanne made
landfall approximately 2 miles apart; as a result, much of the central Florida east coast
1
Former Affiliation: U.S. Geological Survey, Florida Integrated Science Center, St. Petersburg, Florida
was pounded by storm waves and surge, with effects compounded by the temporal and
proximal occurrence of the storms (Figure 2). In addition, Hurricane Ivan, a powerful
major hurricane, made landfall in the north Gulf coast, and battered the coastline of
Alabama and the western Florida panhandle. The U.S. Geological Survey, utilizing
previously published research methodologies, accelerated attempts to predict and forecast
coastal impacts during 2004’s busy hurricane season.
Sallenger (2000) established a Storm Impact Scale (SSIS) based on research
conducted along the Atlantic coastline. The SSIS is divided into four different coastal
impacts (Figure 3); these include Swash, Collision, Overwash, and Inundation, all of
which focus on the interaction between the wave/storm surge interactions with the
primary beach dune, also known as the “first line of defense”. The Swash regime (Impact
1) indicates wave runup with no antecedent dune erosion. The Collision regime (Impact
2) indicates heightened wave runup with coincident dune erosion. Overwash (Impact 3)
results in the landward displacement of beach dunes over a short distance, while the
Inundation regime (Impact 4) results in a landward displacement of beach dunes on the
order of 1 km. An initial forecast of potential change to shorelines was rapidly created by
USGS researchers and posted to the web prior to the landfall of Ivan (Figure 4) and
Jeanne.
Development of the CIAT
To improve on these estimated shoreline impacts, and progress toward real-time
coastal impact prediction from hurricanes and nor’easters, a pilot project was initiated to
examine the feasibility of a digital GIS-based tool for SSIS prediction along a shoreline.
The end result is an experimental product termed the Coastal Impact Assessment Tool
(CIAT). The CIAT is a tool embedded within the ArcGIS 8.x series of products, and in
functional form, is designed to appear as a toolbar within the GIS environment. The
CIAT is primarily menu driven, with minor user-input required into dialog boxes. The
CIAT utilizes the ArcGIS Spatial Analyst extension to conduct operations on raster files,
such as those produced from remotely-sensed data. The current experimental version of
CIAT is version 1 (v.1); future versions are expected to be available for use by coastal
managers, disaster management professionals, and hazard researchers.
The CIAT pilot project is focused on Assateague Island, Maryland/Virginia
(Figure 5). Assateague is an ideal location for such a project because: 1) it possesses both
large and small dune structures, some artificial, as well as areas of the beach in which
only a sand berm is present, 2) Assateague experiences large impacts from coastal storms
such as nor’easters, 3) several previous storm impacts on the island, such as the enormous
1962 Ash Wednesday nor’easter, have been documented (Morton et al. 2003; Morton and
Sallenger 2003), and 4) the island has been the scene of multiple studies and
topographical surveys utilizing LIDAR (Light Detection and Ranging) technology. Past
LIDAR surveys have generally taken place with the Airborne Topographic Mapper
(ATM) sensor. The ATM sensor, considered much more accurate than topographic
elevation estimation via photogrammetric techniques, was designed to detect the first
return from the laser pulse, and thereby is subject to derived elevation distortion due to
vegetation coverage or man-made structures. Previous studies incorporating ATM
LIDAR have been conducted along the U.S. shoreline (Brock et al. 1999, 2003; Stockdon
et al. 2002). Recently, a new NASA-created LIDAR known as the Experimental Airborne
Advanced Research LIDAR has been utilized for coastal elevation surveys, coral reef and
bathymetry mapping, and land use/land cover classification (Figure 6). The EAARL
LIDAR improves on the capabilities of the ATM sensor by detecting an entire waveform,
and not just a single returned pulse, thereby allowing for multiple processing possibilities.
For example, waveform detection in the EAARL sensor allows for post-processing of the
vegetation canopy top, if desired, or bare earth. EAARL also has the distinct advantage
over ATM in that it is water-penetrating to approximately 1-secchi-depth. EAARL has a
spatial resolution of 1 m and a swath width of approximately 240 m at a nominal flying
height of 300m. Assateague Island has near-full-island coverage of EAARL LIDAR
surveys adding to its attractiveness as a study site.
The CIAT utilizes LIDAR-derived beach slopes and dune crest raster data to
determine SSIS regimes. Here, beach slope and dune crest data were derived from
EAARL LIDAR data through processing in remote-sensing-analysis software (ERDAS
Imagine). The beach slope files were created by utilizing land-only-filtered topographic
raster files. The land/non-land cutoff values were set at 0.81 m, based on personal
communication with H. Stockdon of the USGS, and include a value of 0.31 m for Mean
High Water (MHW) plus variability at that location of the U.S. coastline of 0.5 m. The 1
m spatial resolution elevation data was resampled to 5 m spatial resolution elevation data
using a focal analysis technique in which elevations in a moving 5 m by 5 m grid of raster
pixels are averaged to an elevation value in the center pixel. The two land-only pixels
closest to the MHW line were then utilized to generate beach slopes using a commonlyaccepted formula (change in elevation between pixels/distance between pixels). The end
units of slope are unitless, although since, for small slopes, the tanθ (rise/run) is
essentially equal to θ (angle measured in radians), the radian unit can be substituted.
Figure 7 illustrates the procedure utilized to determine beach slopes. The end result is a
line of 5 m pixels containing a beach slope for use in the CIAT.
The CIAT also requires a line of pixels containing values for the primary dune
crest (D-high) along a barrier island’s shoreline. Transects superimposed over the LIDAR
data were generated every 50 m along the Assateague coastline utilizing a GIS-based
program known the Digital Shoreline Analysis System (DSAS). A spatial profile was
generated along each transect in order to digitize the D-high point (Figure 8) (Elko et al.
2002). Additional dune crest elevation points were digitized between the transected Dhigh “control points” by manually examining the LIDAR data. The resulting file is a line
of pixels containing a value for D-high in meters.
After completing the processing procedures for the D-high file and beach slope
files (Figure 9), a mask can be generated for the actual model area. The mask covers the
area on the beach from the pixel line of beach slope values to the pixel line of D-high
values. In essence, this means that the CIAT model covers that portion of the beach from
MHW to the first major dune crest. A file containing this model mask must be created
and added to the table of contents in the GIS in order for the CIAT model to operate. For
further detail on the design, characteristics, and operating requirements of the CIAT, see
the CIAT v.1 User’s Guide (Bossak et al. 2005).
The formulae utilized in the formulation of the SSIS have been tested previously
(Sallenger et al. 1999; Sallenger 2000), but not in a GIS-based model, as has been
accomplished here. The formulae for the R-high and R-low calculations are:
Rhigh = R2% + tide, where
[ H 0 L0 (0.563β f + 0.004]1 / 2
2
R2% = 0.35β f ( H 0 L0 )
1/ 2
+
2
and
Rlow = Rhigh – S2%, where
S 2% = H 0 (0.85 *
βf
H 0 L0
+ 0.06)
The parameters of the Iribarren number, ξ0, used to determine reflective versus
dissipative beaches, are incorporated into the model formulae, as reflected by the
variables β (beach slope), H0 (wave height), and L0 (wave period). For more information
on the wave runup equations utilized in the SSIS and CIAT models, see Stockdon et al.
(2002; in review) or Holman (1986).
Testing the CIAT on Actual Nor’easters
Two strong nor’easters struck the eastern seaboard of the U.S. in the winter of
1998 within a time span of two weeks. Figure 10 illustrates peak wave heights from the
two storms. ATM LIDAR surveys were performed before the storms impact, on
September 15, 1997, and just after the storms, on February 9-10, 1998. This data was
utilized, along with ancillary data such as aerial photographs and DOQQs, to model the
impact of the nor’easters on Assateague’s coastline. Figure 11 illustrates changes to two
dune structures located along the shore of Assateague. In the top graphic, the dune has
been eroded in a classic Overwash Regime profile. In the bottom graphic, the dune has
been displaced in a classic Collision Regime profile.
The CIAT was tested sequentially on both the January 28th and February 5th peaks
in storm intensity. The NOAA deepwater wave buoy 44009, located offshore of Ocean
City, Maryland, recorded wave heights during the storm, and this data forms the basis of
nearshore interpolation of wave data. Nearshore interpolation was accomplished by
USGS researchers (using the SWAN model) soon after the storms took place, and this
data was utilized as the CIAT parameters for coastal impact prediction. The nearshore
parameter data utilized for January 28th was a wave height of 3.96 m, a wave period of
12.8 s, and a tide/surge value of 1.04 m. The nearshore parameter data for February 5th
was a wave height of 4.11 m, wave period of 10.7 m, and a tide/surge value of 1.44 m.
Model results were not compounded between each storm.
Qualitative Assessment
Results of the January 28th and the February 5th model runs were analyzed via
qualitative methods. Several assessments of model output were generated from wholeisland model runs. On each of the assessments, blue triangles and green squares on the
image represent the location of samples taken for quantitative model validation purposes
(discussed in next section). The underlying aerial photograph used in the assessments was
taken on October 24, 2003, and is therefore five years post-storm occurrence. The values
in each legend represent the predicted SSIS.
Figure 12 depicts model output for a portion of the northern end of Assateague
Island. The left panel depicts model output for Jan. 28th, while the right panel depicts
model output for Feb. 5th. This portion of the coastline consists of sandy overwash
deposits, and likely breaches in the dune that fill in with time. Model predictions for
January demonstrate a universal model prediction of Overwash (Impact level 3). For the
February nor’easter, a near universal prediction of Overwash is also indicated. Based on
the aerial photography of the modeled region, overwash deposits are conceptually likely
in this region during a storm consisting of the parameters decribed above.
Figure 13 depicts model output for a central portion of Assateague Island. Legend
components are consistent with those in Figure 12. This central portion of Assateague is
composed of an artificial dune ridge. Model results for January (left) and Februrary
(right) indicate either no dune erosion (Swash) or some dune erosion (Collision), with
very little, if any, overtopping of the dunes during an Overwash or Inundation event.
Figure 14 illustrates model output for a portion of the southern end of Assateague
Island. Here a narrow sandy portion of the beach subject to frequent Overwash events is
succeeded lower on the coast by an accreting coastal zone that also experiences
Overwash, and in some cases Inundation. The model predictions indicate that slightly
more Inundation would occur due to the February 5th storm than would result from the
January 28th storm. In all of these instances, and elsewhere along the coast of Assateague,
the model results appear reasonable based on qualitative assessment.
Quantitative Assessment
In order to conduct a brief quantitative assessment of model results, 122 samples
were gathered in total from the January and February nor’easter model results (depicted
with blue triangles – January – and green squares – February – in the previous three
figures). For each of these sample sites, collected via stratified random sampling along
the shoreline, the LIDAR elevations from the pre-storm and post-storm data, pre- and
post- storm D-high values, and horizontal change in the location of D-high were
recorded. The goal of the quantitative analysis was to determine if, when the model
predicts a specified SSIS value, there is a measurable difference between recorded values.
The samples from January and February were combined to form the dataset used
the next four figures. Three of these figures are boxplots. Boxplots depict a large volume
of information about a dataset in one graphic. The outer “whiskers” on each boxplot
represent the 95% boundary of the dataset, meaning that 95% of all the values in the
distribution for an individual variable lie within the area contained by the whiskers. The
shaded box between the whiskers represents the range of 50% of the data (from the 25th
percentile to the 75th percentile), and the horizontal line with a solid diamond shape
superimposed represents the median value for any particular variable. Individual lines
with a solid triangle superimposed represent values for “outliers”, which are values of the
variables that fall outside of 95% of the entire range (Norman and Streiner, 2003).
Figure 15 is a boxplot of the sample data from the model verification test on the
two nor’easters. The horizontal axis presents the predicted SSIS regime, with the vertical
axis presenting the displacement of D-high in meters. The data in this figure represents
the distance that the location of the crest of the first major dune landward from MHW has
moved from the pre-storm LIDAR survey to the post-storm LIDAR survey. Predicted
values 1 (Swash) and 2 (Collision), in general, exhibit little landward displacement in Dhigh, and this observation was expected from an accurate model result. Median
displacement in meters for the Swash (Collision) regime was approximately 5 m (10 m).
Median displacement values for the Overwash (Inundation) regimes were much more
amplified at approximately 30 m (55 m). A clear distinction is noted in the figure for
locations predicted to have little to no movement of the location of the dune crest (Swash
and Collision) and the locations where movement of the coastal dunes was predicted
(Overwash and Inundation).
Figure 16 is also a boxplot, but this figure depicts the change in the elevation of
D-high pre- and post- storms rather than its displacement, as in Figure 15. The figure
illustrates that, in general, there was a slight net loss in the elevation of dune crests after
the 1998 nor’easters for both the Swash and Collision regimes, and a slight net gain in the
elevation of D-high for the Overwash and Inundation regimes. The greatest median
increase in the elevation of D-high occurred in those samples classified as Inundation
(SSIS value 4) in the model results, while the greatest variability in D-high elevation
occurred in the predicted Overwash regime locations (SSIS value 2).
Figure 17 is a boxplot of the change in elevation (m) of the 122 sample points. In
regimes 1-3, the median elevation change was approximately 1 m of erosion. The
Inundation category (4) experienced the least variability, with the median elevation
change of approximately 0.25 m. Figure 18, a scatterplot of the sample points, is
segregated by the pre-storm and post-storm elevations of the 122 samples, as determined
from ATM LIDAR data at the sample point locations. The figure clearly illustrates the
generally lower elevations for the samples in regimes 1-3 (Swash, Collision, and
Overwash), with the slight increase elevation noted for Inundation regime (4) samples.
Summary statistics were calculated for the samples as well. The mean elevation
for all samples in the pre-storm LIDAR data was 3.52 m. The mean elevation of the same
samples in the post-storm LIDAR data was 3.07 m. The mean change for all samples in
the elevation of D-high after the storms was -0.213 m. A two-sample KolmogorovSmirnov goodness of fit test (KSGF) was conducted on the sample elevation data for
analysis of the pre-storm and post-storm distributions of elevation. The results of the
KSGF indicated that the distribution of elevation values before the storms was different
from the distribution of elevation values post-storm (p-value = 0.04). The likely cause of
the significant difference in distributions is the erosion of the sample points by wave and
surge action during the storms.
Continued CIAT Development Strategies
The CIAT has been shown to be relatively accurate in predicting the estimated
impact of coastal storm on beach dunes and barrier islands, through both qualitative and
quantitative assessments of model hindcasts. The two nor’easters which struck
Assateague Island in 1998 were evaluated separately, with the impact of each storm not
factored into the effects from the other storm. This is due to the fact that no LIDAR
surveys were completed between the two storms. As a result, several improvements to the
CIAT model are expected to be addressed in future versions.
One improvement will be to address duration modeling capabilities, as the CIAT
is currently designed to deliver a “snapshot” model prediction of coastal impacts.
Research in this topic is currently underway at the USGS, and other coastal research
centers, and a component which incorporates the results of this work can be added in
modular format to future CIAT versions. Another improvement is a module that
incorporates wave and/or beach orientation. The CIAT is currently designed to address
shore-perpendicular waves, which may not accurately reflect the interaction of the
shoreline and wave direction during storms. Here as well, current research results will
shape the sophistication of the modular component added to incorporate this parameter.
Finally, the 3D visualization of model results, and coastal change in general, is an
important capability associated with the CIAT. In its current version, the CIAT is capable
of displaying model results and ancillary data in 3D through the use of ESRI’s ArcScene,
a basic 3D viewer associated with ArcGIS. Future versions will be adapted for use with
ArcGIS 9.x and greater, allowing for the use of the ArcGlobe product or other 3D
visualization programs to be integrated with the CIAT.
Summary of CIAT Capabilities
The CIAT can be used, in a GIS-based environment, and in conjunction with
processed coastal dune elevation and beach slope data, to predict the impact of coastal
storms on barrier islands and beach dunes. Although there are some limitations to the
product, it represents an effective first step at combining coastal hazard research,
remotely-sensed data, 3D visualization capability, and GIS mapping technology to
determine acute coastal change. Future versions will include updated 3D visualization
capability, enhanced directional and temporal complexity within the model calculations,
and operational-use possibilities.
Acknowledgements
I would like to thank Laurinda Travers and Betsy Boynton for assistance with
figures, especially Figure 1; Robert Morton and Abby Sallenger for mentorship with the
research; John Brock and his team of LIDAR engineers for assistance with the EAARL
LIDAR data; Hillary Stockdon for wave data; and Kristy Guy for assistance with VBA
programming.
References
Bossak, B. H., R. A. Morton, A. H. Sallenger, Jr., 2005: A GIS-based system for
predicting impacts from coastal storms – The Coastal Impact Assessment Tool
(CIAT), Version 1.0, User’s Manual. USGS Open-File Report 2005-1260, 45 pp.
Brock, J. C., C. W. Wright, A. H. Sallenger, Jr., W. B. Krabill, and R. N. Swift, 2002:
Basis and Methods of NASA Airborne Topographic Mapper Lidar Surveys for
Coastal Studies. J. Coastal Research, 18 (1), 1-13.
Brock, J. A. Sallenger, W. Krabill, R. Swift, S. Manizade, A. Meredith, M. Jansen, and
D. Eslinger, 1999: Aircraft laser altimetry for coastal process studies.
Proceedings, Coastal Sediments ’99, 2414-2428.
Elko, N. A., A. H. Sallenger, Jr., K. Guy, H. F. Stockdon, and K. L. M. Morgan, 2002:
Barrier Island Elevations Relevant to Potential Storm Impacts. USGS Open-File
Report 02-287, 6 pp.
Holman, R. A., 1986: Extreme values for wave runup on a natural beach. Coastal
Engineering, 9, 527-544.
Morton, R. A., K. K. Guy, H. W. Hill, and T. Pascoe, 2003: Regional morphological
responses to the March 1962 Ash Wednesday storm. Proceedings, Coastal
Sediments ’03, 11 pp.
Morton, R. A., and A. H. Sallenger, Jr., 2003: Morphological Impacts of Extreme Storms
on Sandy Beaches and Barriers. J. Coastal Research, 19 (3), 560-573.
Norman, G. R., and D. L. Streiner, 2003: PDQ Statistics. BC Decker, Inc., Hamilton,
Ontario, Canada, 218 pp.
Sallenger, Jr., A. H., 2000: Storm Impact Scale for Barrier Islands. J. Coastal Research,
16 (3), 890-895.
Sallenger, Jr., A. H., P. Howd, J. Brock, W. B. Krabill, R. N. Swift, S. Manizade, and M.
Duffy, 1999: Scaling Winter Storm Impacts on Assateague Island, MD, VA.
Proceedings, Coastal Sediments ’99, 1814-1825.
Stockdon, H. F., R. A. Holman, P. A. Howd, and A. H. Sallenger, Jr., (In Review):
Empirical parameterization of setup, swash, and runup. 42 pp.
Stockdon, H. F., A. H. Sallenger, Jr., J. H. List, and R. A. Holman, 2002: Estimation of
Shoreline Position and Change using Airborne Topographic Lidar Data. J.
Coastal Research, 18 (3), 502-513.
Figure 1. The tracks of hurricanes Charley, Frances, Ivan, and Jeanne during the 2004
hurricane season. (Graphic: Laurinda Travers, USGS)
Figure 2. Coastal erosion from hurricanes Frances and Jeanne
along Florida’s east coast. (Images: USGS)
Figure 3. The four regimes of the Sallenger Storm Impact Scale (SSIS) utilized in the
CIAT. (Graphics: Betsy Boynton and Laurinda Travers, USGS)
Figure 4. Coastal impact forecast issued by the USGS in advance of Hurricane Ivan’s
landfall in 2004. (Graphics: Kristy Guy, USGS)
Figure 5. Assateage Island, Maryland/Virginia. Bottom image is false-color from the
Landsat 7 ETM sensor.
Figure 6. EAARL LIDAR. A) EAARL is a green wavelength laser with a swath width of
240 m, spatial resolution of 1 m, and vertical accuracy of approximately 10-15 cm. B)
EAARL can resolve an entire waveform, allowing for canopy or bare earth processing.
EAARL is also water penetrating to a depth of 1-secchi-disk. (Graphics: Betsy Boynton,
USGS)
Figure 7. Beach slopes were resolved by calculating the slope
between the first two lines of pixels (5 m) landward of MHW.
Figure 8. Transects every 50 m and spatial profiles along these transects were used to
digitize D-high lines along Assateague.
Figure 9. Deepwater wave heights recorded during two nor’easters near Assateague
Island, MD/VA.
Figure 10. Spatial profiles of pre- (blue) and post- (red) nor’easter beach dunes on
Assateague Island, MD/VA. Top profile represents an Overwash regime, with landward
displacement of the beach dune. Bottom profile represents a Collision regime, with dune
erosion, retreat, and sediment repositioning offshore.
Figure 11. Overwash regime predictions, produced as model output from the CIAT, for
the January (left) and February (right) nor’easters that took place on Assateague Island in
1998. Region of interest is located on the north end of the island.
Figure 12. Model predictions, produced by the CIAT, for the January (left) and February
(right) nor’easters that impacted Assateague Island in 1998. Region of interest is located
in the central portion of the island.
Figure 13. Model predictions, produced by the CIAT, of the January (left) and February
(right) nor’easters that impacted Assateague Island in 1998. Region of interest is located
at the southern end of the island.
Figure 14. Boxplot of the displacement in D-high location related to predicted SSIS
value from the 122 samples selected following the CIAT predictions of the 1998
nor’easters on Assateague Island. Median values are represented by solid diamond
symbols, while outliers are represented by the solid triangles. The least variability occurs
in the Swash regime, with the greatest variability present in the Inundation regime.
Figure 15. Boxplot of the difference in D-high elevation for 122 samples selected after
the CIAT model run on the 1998 Assateague nor’easter impacts. The greatest variability
in D-high elevation changes occurred in the predicted Collision regime locations, with
accretion occurring in the predicted Inundation regime locations.
Figure 16. Boxplot of the change in elevation of the 122 sample points selected
following the 1998 Assateague nor’easter model run by the CIAT. The median values for
all predicted SSIS regimes demonstrate erosion except for inundation samples which
show a very slight elevation increase due to sediment redeposition. The greatest
variability occurred in the predicted Collision regime samples.