• Bromwich, D. H., and S. H. Wang, 2005: Evaluation of the NCEP–NCAR and ECMWF 15- and 40-yr reanalyses using rawinsonde data from two independent Arctic field experiments. Mon. Wea. Rev., 133 , 35623578.

    • Search Google Scholar
    • Export Citation
  • Brunner, R. D., A. H. Lynch, J. Pardikes, E. N. Cassano, L. Lestak, and J. Vogel, 2004: An Arctic disaster and its policy implications. Arctic, 57 , 336346.

    • Search Google Scholar
    • Export Citation
  • Cassano, E. N., A. H. Lynch, J. J. Cassano, and M. R. Koslow, 2006: Classification of synoptic patterns in the western Arctic associated with high wind events and temperature trends at Barrow, Alaska. Climate Res., 30 , 8397.

    • Search Google Scholar
    • Export Citation
  • Cassano, J. J., J. E. Box, D. H. Bromwich, L. Li, and K. Steffen, 2001: Evaluation of Polar MM5 simulations of Greenland’s atmospheric circulation. J. Geophys. Res., 106 , 3386733890.

    • Search Google Scholar
    • Export Citation
  • Comiso, J., 1999: Bootstrap sea ice concentrations for NIMBUS-7 SMMR and DMSP SSM/I, June to September 2001. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/nsidc-0079.html.].

  • Déqué, M., 2003: Continuous variables. Forecast Verification: A Practitioner’s Guide in Atmospheric Science, I. T. Jolliffe and D.B. Stephenson, Eds., John Wiley & Sons, Ltd., 97–119.

    • Search Google Scholar
    • Export Citation
  • Drobot, S. D., and J. A. Maslanik, 2003: Interannual variability in summer Beaufort sea ice conditions: Relationship to spring and summer surface and atmospheric variability. J. Geophys. Res., 108 .3233, doi:10.1029/2002JC001537.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1993: A nonhydrostatic version of the Penn State-NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121 , 14931513.

    • Search Google Scholar
    • Export Citation
  • Flather, R. A., 1976: A tidal model of the northwest European continental shelf. Mem. Soc. Roy. Sci. Liege Ser., 6 , 10. 141164.

  • Grell, G. A., J. Dudhia, and D. R. Stauffer, 1994: A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 122 pp.

  • Hume, J. D., and M. Schalk, 1967: Shoreline processes near Barrow, Alaska: A comparison of the normal and the catastrophic. Arctic, 20 , 86103.

    • Search Google Scholar
    • Export Citation
  • Lestak, L. R., W. F. Manley, and J. A. Maslanik, 2003: Point Barrow, Alaska, and vicinity bathymetry. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://arcss.colorado.edu/data/arcss031.html.].

  • Lestak, L. R., A. H. Lynch, E. N. Cassano, L. Xie, S. Bao, W. F. Manley, J. A. Maslanik, and M. Peng, 2006: CEMEPS flood model output, bathymetry, topography, wind and related GIS layers for Barrow, Alaska. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/arcss302.html.].

  • Lynch, A. H., and R. D. Brunner, 2007: The importance of context in climate change impacts assessment: Lessons from Barrow, Alaska. Climatic Change, 82 , 93111.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., S. McIlwaine, J. Beringer, and G. B. Bonan, 2001: An investigation of the sensitivity of a land surface model to climate change using a reduced form model. Climate Dyn., 17 , 643652.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., E. N. Cassano, J. J. Cassano, and L. R. Lestak, 2003: Case studies of high wind events in Barrow, Alaska: Climatological context and development processes. Mon. Wea. Rev., 131 , 719732.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., and Coauthors, 2004a: Barrow climatic and environmental conditions and variations—A compendium (Tech. ed.). CIRES Rep., 124 pp.

  • Lynch, A. H., J. A. Curry, R. D. Brunner, and J. A. Maslanik, 2004b: Towards an integrated assessment of the impacts of extreme wind events on Barrow, Alaska. Bull. Amer. Meteor. Soc., 85 , 209221.

    • Search Google Scholar
    • Export Citation
  • Manley, W. F., L. R. Lestak, C. E. Tweedie, and J. A. Maslanik, 2005: High-resolution radar imagery, digital elevation models, and value-added GIS layers for collaborative research of environmental change at Barrow, Alaska. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/docs/arcss/arcss302/.].

  • Marquardt, D. W., 1963: An algorithm for least-squares estimation of nonlinear parameters. J. Soc. Ind. Appl. Math, 11 , 431441.

  • Mellor, G. L., 1996: User’s guide for a three dimensional, primitive equation, numerical ocean model. Princeton University, Princeton, NJ, 38 pp.

  • Murphy, A. H., and E. S. Epstein, 1989: Skill scores and correlation coefficients in model verification. Mon. Wea. Rev., 117 , 572581.

    • Search Google Scholar
    • Export Citation
  • National Ocean Service, 1997: Coastal bathymetry of the Bering, Chuckhi, and Beaufort Seas, Washington, D.C. NOAA, digital media. [Available online at http://alaska.usgs.gov/science/biology/walrus/index.html.].

  • Peng, M., L. Xie, and L. J. Pietrafesa, 2004: A numerical study of storm surge and inundation in the Croatan–Albemarle–Pamlico Estuary System. Estuarine Coastal Shelf Sci., 59 , 121137.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. O., and J. A. Curry, 1997: Role of radiative transfer in the modeled mesoscale development of summertime arctic stratus. J. Geophys. Res., 102 , 1386113872.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. O., J. A. Curry, and A. H. Lynch, 1999: Modeling clouds and radiation for the November 1997 period of SHEBA using a column climate model. J. Geophys. Res., 104 , 66616678.

    • Search Google Scholar
    • Export Citation
  • Rinke, A., and Coauthors, 2006: Evaluation of an ensemble of Arctic regional climate models: Spatiotemporal fields during the SHEBA year. Climate Dyn., 26 , 459472.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Commerce, 2001: 2-minute Gridded Global Relief Data (ETOPO2). NOAA/NGDC, digital media. [Available online at http://www.ngdc.noaa.gov/mgg/fliers/01mgg04.html.].

  • USGS, 1997: Chukchi Sea Bathymetry, Anchorage, AK. Biological Science Office, Alaska Science Center, Anchorage, AK, digital media. [Available online at http://www.absc.usgs.gov/research/walrus/bering/bathy/.].

  • USGS, 1999: National Elevation Dataset. EROS Data Center, U.S. Geological Survey, Sioux Falls, SD, digital media. [Available online at http://ned.usgs.gov/Ned/ned.html.].

  • Walsh, J. E., and W. L. Chapman, 2001: Arctic and Southern Ocean sea ice concentrations. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/g00799.html.].

  • Xie, L., and L. J. Pietrafesa, 1999: System-wide modeling of wind and density driven circulation in Croatane-Albemarlee-Pamlico Estuary system. Part 1: Model configuration and testing. J. Coastal Res., 15 , 11631177.

    • Search Google Scholar
    • Export Citation
  • Xie, L., L. J. Pietrafesa, and M. Peng, 2004: Incorporation of a mass-conserving inundation scheme into a three-dimensional storm surge model. J. Coastal Res., 20 , 12091223.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Three model domains were nested to produce the CEMEPS simulations shown here.

  • View in gallery

    Comparison of PMM5-simulated 10-m wind speed (m s−1, black) and direction (°, gray) at the closest grid point to Barrow to the National Weather Service observations at Barrow for the (a) October 1963 case, (b) September 1986 case, and (c) July 2003b case. The solid line shows PMM5 winds and the dashed line shows observed winds.

  • View in gallery

    Comparison of PMM5-simulated 10-m wind speed (m s−1, black) and sea level pressure (hPa, gray) at the closest grid point to Barrow to the National Weather Service observations at Barrow for the November 1950 case. The solid line shows PMM5 winds and the dashed line shows observed winds.

  • View in gallery

    Maximum flood debris line of the October 1963 storm as mapped by HS67 (white line) and as simulated by the PMM5 and CEMEPS (black fill). Note that modeled inundation covered most of the spit, which coincides with the October event. Inundation follows the HS67 flood debris line near the Dew Line station and ARL. Topography has changed between 1963 and 2002 at the landfill, at Isatquaq Lagoon (the bridge road is blocking flooding into the fresh water source), and Dredge Harbor. Note that there is no flooding over the beach berm near ARL. Flooding near the town of Barrow and Browerville stays close to the shoreline because of the higher topography in that area.

  • View in gallery

    Time series of flooding as simulated by the PMM5 and CEMEPS for the October 1963 storm.

  • View in gallery

    Comparison between the tide gauge and simulated flooding for the September 2003 storm.

  • View in gallery

    Scatterplot between the flooding simulated by the physical model work flow (PMM5 and CEMEPS) and the RFM having the highest correlation in Table 4. Eighteen storms applied in the fit are plotted with dots and three storms not used in the fit are plotted with squares.

  • View in gallery

    Single perturbation sensitivities in the flooding of the RFM having the highest correlation in Table 4. Storms are plotted using the numbering from Table 1.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 169 80 4
PDF Downloads 114 62 1

A Factorial Analysis of Storm Surge Flooding in Barrow, Alaska

View More View Less
  • 1 School of Geography and Environmental Science, Monash University, Clayton, Victoria, Australia
  • | 2 Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado
  • | 3 School of Geography and Environmental Science, Monash University, Clayton, Victoria, Australia
  • | 4 Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, Colorado
  • | 5 Coastal Fluid Dynamics Laboratory, North Carolina State University at Raleigh, Raleigh, North Carolina
Full access

Abstract

This paper describes work to improve the understanding of the broad range of factors affecting the occurrence of flooding in Barrow, Alaska, using as a basis the series of extreme events that have affected the community over the past 50 years. A numerical weather prediction model and a storm surge inundation model have been applied to the 21 case studies identified in National Weather Service data as high wind events. Based on this simulation work flow, a reduced-form model that adequately describes the flooding response has been developed. Specifically, it was found that when wind is forecast to be greater than 13 m s−1 (30 mph) for at least 20 h, this is the most accurate predictor of the possibility of damaging flood. It was found that wind direction, the magnitude of fetch to the sea ice edge (when present), and maximum wind speed were in contrast relatively small contributors to the likelihood of flooding.

Corresponding author address: Amanda H. Lynch, School of Geography and Environmental Science, Monash University, Clayton, VIC 3800, Australia. Email: amanda.lynch@arts.monash.edu.au

Abstract

This paper describes work to improve the understanding of the broad range of factors affecting the occurrence of flooding in Barrow, Alaska, using as a basis the series of extreme events that have affected the community over the past 50 years. A numerical weather prediction model and a storm surge inundation model have been applied to the 21 case studies identified in National Weather Service data as high wind events. Based on this simulation work flow, a reduced-form model that adequately describes the flooding response has been developed. Specifically, it was found that when wind is forecast to be greater than 13 m s−1 (30 mph) for at least 20 h, this is the most accurate predictor of the possibility of damaging flood. It was found that wind direction, the magnitude of fetch to the sea ice edge (when present), and maximum wind speed were in contrast relatively small contributors to the likelihood of flooding.

Corresponding author address: Amanda H. Lynch, School of Geography and Environmental Science, Monash University, Clayton, VIC 3800, Australia. Email: amanda.lynch@arts.monash.edu.au

1. Introduction

For the last two decades, the people of Barrow, Alaska, have been working to reduce their vulnerability to flooding arising from extreme storms. The motivating event was the loss of cultural artifacts to the storms of 12 and 20 September 1986 that first exposed artifacts, including the remains of ancestors, and then washed them out to sea. The longer-term concern is that climate change has increased Barrow’s vulnerabilities and damages in the recent past and may continue to do so in the future. The return period for intense cyclones is one important measure of the changing climate around Barrow. However, our work has shown that the return periods for the high wind events associated with these cyclones may be changing with time in a nonlinear way (Lynch and Brunner 2007; Lynch et al. 2004a). The instability of the return period suggests that this quantity cannot be reasonably projected into the future with any confidence. Thus major uncertainties are inevitable. It would be prudent to take them into account in planning responses to flooding, and the research presented here is one contribution to those responses.

A confounding factor in the development of improved forecasting of individual flooding events in Barrow is the paucity of verification data. The meteorological data are limited: there is only one National Weather Service station in the vicinity, in Barrow itself, and some corroboration is available from a nearby National Oceanic and Atmospheric Administration (NOAA) Clean Air site. The upper-air data from the Barrow station has had quality control issues in the past [J. Curry, Georgia Institute of Technology, Atmospheric Radiation Measurement External Dataset Descriptions, 2006, personal communication; more information available online at http://www.arm.gov/xds/static/nwsupa.stm (accessed 13 December 2006)]. More importantly, there is no tide gauge, because of the presence of seasonal ice cover along the coast. This presents a problem not just to studies such as these, which seek to support the community in its goal of reducing its vulnerability to flood, but also to efforts to implement operational forecasting of storm surge to Barrow [the Sea, Lake, and Overland Surges from Hurricanes (SLOSH) model is currently run unverified in Barrow by the National Oceanic and Atmospheric Administration; more information available online at http://www.weather.gov/mdl/etsurge]. Nevertheless, the community vulnerability remains. The intent of the study described herein is to illustrate a methodology that may fill the gap until better validation measures are available for operational implementation. Furthermore, the methodology allows some perspective regarding the large-scale drivers of events that can aid in long-term planning of coastal protection. Results from this project have been used in this context by the Army Corps of Engineers as part of a feasibility study [Brunner et al. 2004; F. Brooks, Army Corps of Engineers (ACoE), 2005, personal communication].

A common feature of many damaging storms is the presence of a large open-water fetch, which limits the damping by ice of waves and wind-driven storm surge. Hence, the season of greatest potential danger from flooding is late summer and early autumn, when the sea ice extent is a minimum and the open-water fetch is the greatest. Autumn storms have historically been most feared and prepared for by Barrow residents. The primary focus of this study, then, is high wind events that occurred during the time of year in which open water is possible near Barrow: the months from July to November (Table 1). For most of the case studies, the criteria for choosing a high wind event were based on National Weather Service observations at Barrow. Days with a peak wind of 24 m s−1 (55 mph), a 1- or 2-min sustained wind greater than 20 m s−1 (45 mph), or daily average wind speed greater than 13 m s−1 (30 mph) were criteria that could be individually or severally satisfied, and were also the same criteria used for selecting high wind events in Lynch et al. (2004b). These different threshold criteria were used because the wind measurement protocol changed throughout the observational record. Sustained and peak winds were not available during the entire time period, but were used as an additional criterion to identify high wind events when available. Although the thresholds are somewhat arbitrary, these criteria successfully capture the strongest high wind events to impact Barrow during the past 60 years. In addition to these cases, other events were identified through meetings and discussions with local residents in Barrow. Though these cases did not meet the thresholds above, they caused significant impacts in the community (e.g., flooding, erosion, structural damages) and thus were important to include in the analysis. For example, in September 2003, the wind speeds at Barrow did not meet the thresholds above, yet it was a storm that had a substantial impact on the community (T. Brower III, Environmental Program Manager, Native Village of Barrow, 2003, personal communication). There were moderately high winds over a 5-day period, which caused erosion along the coast and road closures in Barrow itself. A second case that had caused significant damage but that did not satisfy the threshold criteria occurred in October 1954. This storm caused water to wash into the Arctic Research Laboratory camp, and debris was thrown onto the beach (Hume and Schalk 1967). Finally, a storm event that occurred during midwinter was included in the case studies. This record-high wind event occurred on 22 February 1989 and was included as a test case to determine the impacts of such a storm occurring during an ice-free time of year. The storm did substantial wind damage and resulted in a rather severe “ice push” (i.e., wind-driven surge of sea ice onto shore).

As noted, the storm events examined in this study are characterized by very sparse data, particularly for the earlier events. To obtain a uniform depiction of all events, a modeling approach was used, whereby each storm was simulated by a mesoscale atmospheric model and a storm surge model. Validation was performed at each stage to the extent possible. Based on this information, heuristics were developed that characterize the likely flooding extent, using a reduced-form model that mimics the behavior of the more complex physical models. The methodology is described in detail in section 2, and the performance of the atmospheric and inundation models is analyzed in sections 3 and 4, respectively. The resulting reduced-form model is presented in section 5 and conclusions, limitations and future work are presented in section 6.

2. Methodology

The component models and experimental design composing this work flow are described briefly in this section.

a. Mesoscale atmospheric model

The Polar MM5 (PMM5; Cassano et al. 2001) is used to produce the atmospheric simulations for the events listed in Table 1. The Polar MM5 is a nonhydrostatic mesoscale atmospheric model based on the fifth-generation Pennsylvania State University–National Center for Atmospheric Research (PSU–NCAR) Mesoscale Model (MM5; Dudhia 1993; Grell et al. 1994), modified to better represent processes occurring in the polar regions. These modifications are described in detail in Cassano et al. (2001) and Lynch et al. (2003) for version 3.4. However, the newer version 3.6.2 was used for the storm events not simulated for Lynch et al. (2003) (i.e., all events except August 2000 and October 1963). Version 3.6.2 incorporates some improvements in longwave radiation computations that are important for cold-climate simulations (Pinto and Curry 1997; Pinto et al. 1999), but tests indicate that the simulations do not diverge significantly from those conducted using version 3.4.

As for Lynch et al. (2003), the model domain is centered at 68°N latitude and 172°W longitude and has a horizontal extent of 2820 × 2370 km, with a horizontal grid spacing of 30 km and 23 vertical levels (see Fig. 2 in Lynch et al. 2003). The lowest sigma level is located at a nominal height of 38 m above mean sea level and the model top is specified at the 100-hPa constant pressure level. The initial and boundary conditions for the model atmosphere are created using European Centre for Medium-Range Weather Forecasts (ECMWF) Tropical Ocean Global Atmosphere (TOGA) operational analyses for the cases from September 1986 onward, and the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis product for cases prior to September 1986. Bromwich and Wang (2005) found that for the region in question, neither reanalysis shows evidence of significant bias. The sea ice concentration and extent are specified using data from the National Snow and Ice Data Center (NSIDC). For the cases from September 1986 onward, the ice concentrations are derived from the Defense Meteorological Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I), using the bootstrap algorithm (Comiso 1999). For cases prior to September 1986, sea ice concentration and extent were obtained from the Arctic and Southern Ocean sea ice concentration dataset developed by Walsh and Chapman (2001).

b. Storm surge model

Barrow inundation was modeled using the North Carolina State University Coastal and Estuary Marine and Environment Prediction System (CEMEPS; Peng et al. 2004; Xie et al. 2004; Xie and Pietrafesa 1999). The model includes a three-dimensional storm surge component, a linear wave model, and an inundation component.

The storm surge component is based on the Princeton Ocean Model (POM; Mellor 1996), a three-dimensional ocean model that includes a second-moment turbulence closure parameterization and a free surface with explicit prediction of sea level change. Atmospheric forcing for the model is provided, in this application, by the PMM5 simulations, and includes the application of surface wind stress and a pressure perturbation. A radiation open boundary condition (Flather 1976) for the outer domain edges that do not comprise land permits surface waves to propagate out of the model domain.

The inundation/drying scheme is described in detail in Xie et al. (2004). This scheme uses two criteria to determine the incidence of inundation: the sea level height relative to an adjacent land grid point and the flow velocity at that location. The inundation velocity is computed from the surface current derived from the three-dimensional ocean model. Draining at a flooded land grid point occurs when the water depth falls below a preset threshold value, because of the complexity of determining a unique drainage pattern in the highly complex terrain. An innovative aspect of the model is that a mass balance constraint is imposed on the inundation process. While conservation of water mass may not be important for sea level on regions that are connected to an open ocean (as in this application), it cannot be ignored in a closed or nearly closed shallow-water system.

In this study, the storm surge and inundation modeling system was configured for the Barrow coastal zone and its adjacent shelf using three nested model domains (Fig. 1). The grid spacing of the domains was 800, 200, and 50 m, respectively. The domains were derived from an interpolated bathymetry grid and digital terrain model (DTM), which was originally created at 50- and 200-m resolution using a variety of digital datasets (Lestak et al. 2006) including an Interferometric Synthetic Aperture Radar (IFSAR) DTM (Manley et al. 2005), the U.S. Geological Service (USGS) National Elevation Dataset (NED; USGS 1999), the NOAA bathymetry chart (Lestak et al. 2003), points from the NOAA 2-minute gridded elevations/bathymetry for the world (ETOPO2) dataset (U.S. Department of Commerce 2001), and the USGS and National Ocean Service (NOS) Bering and Chukchi Sea Bathymetry Datasets (USGS 1997; National Ocean Service 1997). These datasets were first produced in the Universal Transverse Mercator (UTM) map projection and then projected onto a latitude–longitude grid for use by CEMEPS. The effects of sea ice were not included in these flooding simulations, although they were included in the Polar MM5 simulations.

c. Reduced-form model

The technique applied in this study to create a reduced-form model (RFM) in order to generate a Barrow flooding heuristic follows that of Lynch et al. (2001). The reduced-form model was chosen from a set of distributions consisting of first-, second-, and third-degree polynomial, exponential, Gaussian, and Gamma curves. Multidimensional nonlinear least squares fitting was applied using the Levenberg–Marquardt solver (Marquardt 1963) to determine the models of best fit. The combination of Gaussian and third-degree polynomial forms was chosen for the subsequent analysis, because it produced the best overall correlation to the flooding simulated by the physical model work flow. The successful model to be fitted is, then, the sum of NM Gaussians and M = [0, N] third-degree polynomials:
i1520-0493-136-3-898-e1
where the vector x is the perturbation vector and the parameters a are determined using the Levenberg–Marquardt solver. Here ŷ is the estimated response metric, which is flooding in Barrow simulated by the physical model work flow.

Out of a total of 21 events, 18 were chosen for the construction of the reduced-form model based on Eq. (1), with three omitted to test the validity of resulting model (see Table 3). Because Eq. (1) requires three fitting parameters per perturbation variable, one can have up to N = 6 variables in the reduced-form model. Eight candidates for perturbation variables were defined (Table 2), resulting in 8!/N!(8 − N)! RFMs for N = [1, 6]. Perturbation variables were based on PMM5 model output and weather station observations in Barrow. RFMs based on every combination of perturbation variables were calculated and compared with the flooding simulated by the physical model (Table 4). Results are described in detail in section 5.

3. Simulation of high wind events

The atmospheric model simulations were typically 1 week to 10 days long with two exceptions: September 1986 and July 2003b (see Table 1). There were two high wind events that occurred about 1 week apart for each of these cases, so the model simulation was long enough to capture both events in the same experiment. In each case, the model simulations began 3 days before Barrow station observations indicated that high winds affected Barrow. The model results generally compared well with observations (e.g., October 1963; Fig. 2a), although in some cases simulations were not as skillful as expected (e.g., September 1986; Fig. 2b). For this latter case, the Polar MM5 failed to capture the second high wind peak that occurred with the passage of a second cyclone 8 days after the first. This is perhaps not surprising, as significant skill would not be expected after 11 days of simulation. On the other hand, the July 2003b case maintains a high level of skill after more than 15 days of simulation (Fig. 2c).

Table 1 shows a summary of forecast statistics for the case study high wind events. These assessments of quality are based solely on the Barrow station observations, which is the only quality-controlled data source available for the entire time period, and also represents the location of impact, being the primary population center for this time period. However, other more detailed assessments have been performed for individual cases where more data is available (e.g., Lynch et al. 2003). Shown in Table 1 are biases for sea level pressure and 10-m wind speed calculated as
i1520-0493-136-3-898-e2
where obs is the observation at the Barrow National Weather Service station, MM5 is the value at the closest grid point to Barrow, and n is the total number of time periods available for the calculation. Of course, it is possible that compensating positive and negative values would result in a small bias, and in fact such was the case for the July 2003b case. However, generally this compensation was not evident. The sea level pressure biases ranged from −3.24 to +3.67 hPa with 13 of the 21 cases having a positive bias. The bias in the wind speeds ranged from −2.82 to 1.39 m s−1, with 14 of the 21 cases having a negative bias. The range of results confirms previous analyses that the PMM5 exhibits adequate performance with no systematic problems (e.g., Cassano et al. 2001; Rinke et al. 2006).
Perhaps a more representative measure than an assessment of bias, the root-mean-square error was calculated:
i1520-0493-136-3-898-e3
The sea level pressure RMSE ranges from 1.6 hPa for the October 2002 case to a maximum of 5.53 hPa for the October 1954 case, apart from an outlier of 9.51 hPa for the November 1950 case. Around half the cases have an RMSE under 3 hPa (Table 1). For the wind speed RMSE, the values ranged from 1.32 m s−1 for the November 2000 case to 3.76 m s−1 for the September 1986 case, with around half exhibiting an RMSE of under 2.5 m s−1. Again, November 1950 was an outlier with an RMSE of 5.17 m s−1.
The mean-square error skill score (MSESS) was calculated as an additional measure of model skill. This statistic compares the model output to a lower skill forecast such as climatology or persistence (Déqué 2003). In this application, the calculation follows Murphy and Epstein (1989):
i1520-0493-136-3-898-eq1
where clim is the monthly climatological value of sea level pressure or wind speed. A value of MSESS equal to 1 is the maximum value possible and represents a perfect forecast, a 0 represents a model forecast equivalent to a climatological forecast, and a negative number represents a model forecast accuracy less than that of the climatological forecast (Déqué 2003). The MSESS values (Table 1) for the SLP range from 0.51 for October 1977 to 0.98 for both September 2003 and October 2002, aside from an outlier of −0.01 for November 1950. For the wind speeds, the MSESS values ranged from a low of 0.08 for September 1986 to a maximum of 0.86 for both August 1950 and October 1998. Hence, apart from November 1950 (discussed in more detail below), the model produced forecasts better than climatology, which is a minimum standard required.

Finally, the correlations between the time series of observations and simulated values were calculated. Most of the correlations were quite high, with 18 of the 21 sea level pressure correlations above 0.9. However, the November 1950 case had a sea level pressure correlation of only 0.58, consistent with the high RMSE and negative MSESS for this case. Figure 3 shows the observational time series for this event compared with the simulated values. Like the September 1986, October 2002, and July 2003b cases, this event consisted of a series of high wind events occurring in Barrow in quite rapid succession. The simulation was able to capture the passage of the second cyclone (labeled B in the figure) fairly well, but the model fails to capture both the intensity of the first pressure minimum on the 27th (labeled A) and the third pressure minimum on 1 December (labeled C) is missed completely.

The wind speed correlations across the case studies were somewhat lower, as expected, with 12 of the 21 cases exhibiting correlations greater than 0.8. The September 1986 case had the poorest performance in this regard, because of the failure to capture the passage of the second event (Fig. 2b). The September 2003 case was also problematic, along with the November 1950 case (Fig. 3). The effect of these failures on the simulation of flooding will be assessed in section 5.

4. Simulation of flood events

The primary challenge of the analysis of flood events in Barrow, as noted earlier, is the absence of data on the occurrence of flood events or the extents of documented floods. One case that has data available is the October 1963 flood, the storm that stands out in living memory as the most severe and damaging storm to have affected Barrow. This storm has been described in detail in Lynch et al. (2003, 2004b) and hence only factors pertinent to the simulation of flooding will be addressed here. While the October 1963 cyclone was certainly very intense, it did not produce record winds at Barrow. However, the combination of sustained westerly winds and open water were thought to have been responsible for the extensive flooding (Hume and Schalk 1967, hereafter referred to as HS67). The progress of the flood was documented in an Arctic Research Laboratory (ARL) annual report, which in the absence of the tide gauge noted the times at which certain key landmarks were inundated. In this report, it was noted that while strong winds were evident from midnight [Alaskan Standard Time (AST), or 0900 UTC], water did not start to rise until early morning. By around 1000 AST (1900 UTC) water covered the ARL runway (Fig. 1), and the freshwater lake (Imikpuk Lake) had been contaminated with seawater. Water entered the ARL “camp” by 1300 AST (2200 UTC) that afternoon, with at least 0.3 m of water evident in the main street. The peak occurred between 1 and 3 h later, with water as deep as 1 m in some locations and currents very strong. HS67 estimated the storm tide surge at the peak of the storm to have been as large as 3.5 m based on the reported water depth and beach elevation, in comparison to the estimated surge of 2.5–3 m for the October 1954 storm. Subjective estimates of wave heights that would have added to this surge are typically unreliable; HS67 assume a likely wave height of around 3 m. A final flood extent was mapped by HS67 based on debris and still-water lines, and is reproduced in Fig. 4.

The modeled flood extent (Fig. 4) shows a good correspondence to the HS67 estimate, although the details of timing are not as well reproduced. At 1000 AST (1900 UTC), simulated water flow was slowly moving toward camp, covering portions of the coast road and portions of the ARL runway. At this time, simulated inundation was less than actual storm flooding, and Imikpuk Lake was not breached until 1300 AST (2200 UTC) in the simulation. By 1400 AST most of the main street was flooded by the model. The simulated flood waters followed the road into camp, which corresponded well to descriptions in the ARL report. Modeled flood water depth was between 0.5 and 1 m at this time. During the actual event, flood waters had completely receded by 1830 AST, but in the simulation, maximum extent of flood waters was reached at 1800 AST (Fig. 5), and by 1700 AST (0400 UTC 4 October) water levels began to drop as flood water equalized with Elson Lagoon. Flood water in the simulation remained in the ARL camp for the duration of the model run [until 1400 AST (2300 UTC) 4 October].

In summary, while the simulated flood lags the actual flood by some hours, the main characteristics of the composite flood area were adequately reproduced. The time lag in the simulated onset of peak flooding and subsequent retreat was likely because the initial ocean currents and water level were unknown, and hence the simulation commenced from an “at rest” ocean state with zero water level. Furthermore, during the actual event, storm water currents created a breech into Elson Lagoon, thus allowing flood waters to drain into Elson Lagoon. Such a modification of the landscape was explicitly excluded from the simulation. Finally, the atmospheric model simulation was conducted at a grid resolution of 30 km, and hence the local wind variations of potential importance to the details of the flood simulation could not be reproduced. The good correspondence to the observed flood extent, then, was achieved despite a high level of uncertainty in initial sea state, fine (and changing) geographic features, and local winds.

Turning to the other flood simulations, we present a summary in Table 3 of the maximum sea level height simulated in each case (variable flood, right-hand column) and an indication of the observed flooding (variable observed flood). In comparing these variables, we define a false positive as an instance in which the model work flow predicted a flood and a flood was not observed to occur. A false negative is an instance in which the model work flow predicted no flood and a flood did in fact occur. In the forecast and warning sense, a false negative is more serious (and potentially dangerous) than a false positive. Noteworthy in this summary is that the model simulates three apparently false positives (only one of which involved significant fetch) and no false negatives, which supports the use of these simulations as a form of forecast guidance. The storm of September 1970 is listed as unknown because in interviews, references to this event, and specifically whether a flood occurred, were ambiguous. The maximum flood simulated was November 1966—this is discussed in more detail below. As noted, the maximum simulated flood depth in October 1963 was almost 1 m, which did not include wave setup effects. This is the only storm for which we have good data regarding total flooded area, but there is some information available for other floods in this list. For example, as noted earlier, HS67 mention a flooding event in October 1954—in this case the flood simulated by the model is much lower and less extensive than the anecdotal reports of this severe storm. Similarly, the flooding reported in September 1986 was probably more extensive than the simulation indicates. In this case, as noted in the previous section, the MM5 simulation failed to capture the second wind peak. In addition, considerable erosion was evident in photographs taken at the time, including some catastrophic shoreline collapse. This suggests that wave action played a large part in these storms, and such wave action is omitted from the model. Several of the other cases (e.g., September 1970 and October 1993) demonstrated simulated flooding where none has been reported, but because of the lack of data floods on these dates cannot be ruled out. No flood was simulated for the November 1950 case, which was poorly captured by the MM5 simulation. However, in this case the ice was in and hence flooding was unlikely in any case. Certainly, no flood was reported on this date. There is more information for the more recent flood events. The August 2000 storm was a significant event, but the flooding simulated by the model was probably a slight overestimate based on contemporary reports. In October 2002, a storm occurred that resulted in the washing out of the coastal road, but the lagoons were not flooded as the model simulation suggests. Hence, many of these latter cases appear to be overestimates.

In 2003, Evans–Hamilton Ltd. was contracted by the Army Corps of Engineers to collect data on sea level height and waves at two sites along the coast at Barrow. Data was collected for the time periods: 12 August–30 September 2003 and 16 July–13 September 2004. The earlier period encompasses the storm in September 2003 (Fig. 6). The sea level height derived from the gauge includes wave height in addition to storm surge, and hence is plotted on a different scale from the storm surge–only model output. For this reason, a direct comparison of magnitudes is not possible. Figure 6 does indicate, however, a significant surge in response to the wind forcing. Like the October 1963 storm, a delayed response is evident, which is associated with the initialization of the model as noted above. Furthermore, the duration of the event is too short, which is associated with shortcomings in the wind simulation of the PMM5 (Table 1)—specifically the mesoscale model was unable to maintain a stationary system for the two day period that was observed. Clearly, the effect of waves and wave setup is a crucial component in the simulation of these events.

One storm demonstrated a more severe flood than October 1963 in the simulations listed in Table 3; this was the November 1966 case. This case should be considered a hypothetical experiment, because the sea ice was fast in to the coast during this storm, and hence protected the community from storm surge inundation. This storm serves as a particularly interesting hypothetical case because sea ice set up in recent years has trended rapidly later in the year. The average time of freeze-up in the 1960s, according to HS67 and based on observations from the U.S. Hydrographic Office, was between 1 and 5 October. In comparison, the average freeze-up in the 26-yr record since 1979 is 17 October and the linear trend from 1979 to 2004 is 14.8 days decade−1 (p value = 0.003). While the satellite record does suffer somewhat from coastal contamination, the trend toward a later freeze-up date is consistent with a reduction in multiyear sea ice extent and changes in atmospheric circulation as noted in Drobot and Maslanik (2003).

Interestingly, this case was very similar to the February 1989 case, in which very little flooding was simulated. (Note that this is also a hypothetical experiment—the ice is always in during February.) In both storms, the low pressure system moved north from the Aleutian Islands through the Bering Strait, to the north of Barrow. The average daily wind speed was 16.1 m s−1 in 1989 and 15.2 m s−1 in 1966, and the 1-min sustained wind was 24.6 m s−1 in 1989 and 24.1 m s−1 in 1966. In both cases, the prevailing wind direction for the day was from the southwest. In the next section the factors that may explain why the November 1966 case, with slightly lower winds, produced substantial modeled flooding, while the February 1989 produced almost none, will be explored.

5. Reduced-form modeling

As demonstrated in the above discussion, any analysis of the necessary and sufficient factors to result in a damaging flood is hampered by a severe lack of data on the one hand and sometimes problematic simulation quality on the other. Nevertheless, better information that can inform the prediction of damaging and dangerous flooding in Barrow is certainly needed (Lynch et al. 2004a, b; Lynch and Brunner 2007). In this context, by including a broad range of cases, it should be possible to extract some general heuristics for flood occurrence in Barrow that can then be used as additional guidance by weather forecasters.

The model-simulated flooding at Barrow was the response metric to be characterized in this exercise, as a proxy for the unavailable observed flooding. This metric was defined as the maximum sea level height in any one grid point of the 81 closest grid points around Barrow during the storm. This choice was found to have a good subjective relationship to the reports of observed flooding that were available, within the limits of the simulation quality. As demonstrated in the previous section, CEMEPS suffers from little systematic bias, except in the speed of onset of flooding, and hence we feel this is a useful exercise. The variables that were proposed as possible forcing of this response metric are listed in Table 2 and include both observed and modeled variables. The observed variables included the Barrow station wind speed and fetch to the sea ice edge derived from satellite data. Both of these variables are routinely available to forecasters. In addition, model-derived variables associated with simulated cyclone intensity and track act as a proxy for forecast products available from NCEP.

The forcing variables and resultant flooding for a selected 18 of the 21 cases were used in Eq. (1) to determine the appropriate values of the parameters a for a given reduced-form model. This procedure was performed by excluding a number of different sets of three, including both randomly selected sets and the set of the three poorest performing wind simulations as defined by the correlation in Table 1 (November 1950, September 1986, and September 2003). The differences between the results based on these different selections were insignificant, and the results presented here use the latter selection (see also Table 3). The flooding predicted by the RFM was then compared with the flooding predicted by the physical model work flow, and the correlations for all 21 storms are listed in Table 4 for a variety of forcing variable combinations. All possible combinations were tested in order to understand the minimum requirements for predicting a flood event. Only those combinations that demonstrated with a probability higher than 25% to have the highest correlation within its group are listed, and given the number of cases, only combinations of a maximum of three variables are possible to avoid overfitting.

When one forcing variable is used, there is only one choice that produces a significantly high correlation, and this is the time period (h) over which the simulated wind speed exceeded 13 m s−1 (30 mph; WGt1). This is consistent with subjective expectations, because WGt1 is proportional to the energy of a storm, which is the power delivered by the wind (roughly proportional to the cube of the wind speed) integrated over WGt1. Therefore WGt1 contains information regarding the period when the storm was strong enough to cause flooding, assuming the threshold value was suitably chosen. WGt1 was also estimated based on a higher threshold of 20 m s−1 (45 mph), but simulated winds exceeded this limit during three storms only, a population too small for reduced-form model construction. One may note that WGt1 is included in all of the fits producing high correlations in Table 4, which signifies its importance as a variable associated with flooding. The second best fit in the one variable RFMs was obtained by using Dir, with a correlation of 0.70, followed by WMaxObs, (correlation of 0.67).

Two to three forcing variable combinations produce a set of three reduced-form models, listed in Table 4, which emulate the physically modeled flooding reasonably well. Apart from WGt1, the variables with the greatest predictive value are WMaxMM5 (a variable in two combinations), Fetch, and SLP. The combination producing the highest correlation includes three variables: WGt1, WMaxMM5, and Fetch. In this reduced-form model, WGt1 explains 87% of the variance, which means that the remaining variables simply “fine-tune” the fit.

Overall the best RFM produces reasonable flooding in comparison with the physical model work flow, and also for the three storms not included in the fit (Fig. 7). The reduced-form model behavior shown in Figure 7 indicates that the reduced-form model performs best (i.e., best reproduces the behavior of the physical model work flow) when the forcing is large, and relatively poorly when the forcing is small. Hence, the RFM has a tendency to produce false positives, although these present flooding occurrences of magnitude less than 10 cm. Hence, this reduced-form model is deemed suitable for the generation of a Barrow flood heuristic.

To examine the behavior of this RFM in more detail, single variable sensitivities are presented in Fig. 8. These plots project the three-dimensional response surface onto a single dimension, and hence cannot display the total fit developed by the RFM, but they do provide information on the response to particular forcing variables. Certain perturbation variables explain flooding during certain storms apparently better than other variables. For example, WMaxMM5 explains flooding in February 1989 and August 2000 (numbers 10 and 15 in Table 1 and in Fig. 8), while WGt1 explains flooding of between 1.0 and 1.2 m in the most severe cases of October 1963 and November 1966, and also in October 1993 (cases 5, 6, and 12). Fetch explains flooding well in October 1954, September 1970, and October 2002 events (cases 3, 7, and 17). The spread of cases in the WGt1 space suggests that there is indeed an optimal time period for high winds to generate extensive flooding, and hence a useful heuristic that is generated is that flooding can be expected if winds a forecast to exceed 13 m s−1 (30 mph) for around 20 h when the ice is out.

6. Conclusions

In this paper we have analyzed the factors leading to damaging floods during storms in Barrow, Alaska, in the context of very poor data availability. Nevertheless, the models used here have been demonstrated to exhibit sufficient skill such that it is possible to gain some insights into the factors leading to these floods. The PMM5 model exhibited high simulation quality in all but a handful of storms, and compared well to the available data at the Barrow National Weather Service site. A more detailed analysis was conducted for two of these storms in Lynch et al. (2003). The CEMEPS model has not previously been deployed for such (relatively speaking) weak midlatitude systems, and from the information available performed remarkably well. By combining these two models into a work flow it has been possible to create an analysis that links forecast weather to the likelihood of a flood.

Specifically, it has been possible to demonstrate that a forecast for winds exceeding 13 m s−1 (30 mph) for at least 20 h is the most significant predictor of a severe flood, as long as there is relatively open water close to shore. It is significant that forecast winds, available in advance of a storm, appear to contain significant information. We found that wind direction was of relatively less importance. This latter finding is not because the westerly winds (commonly believed to be necessary for flooding events) are the strongest winds nor are westerly winds prevalent during the open water season (see, e.g., Cassano et al. 2006). Hence, we suggest that wind direction be given less weight than duration of strong winds in assessing a forecast.

A major limitation of the present study is the inability to make a quantitative assessment of the importance of fetch. While it is clearly evident that open water is a necessary precondition for flooding to occur, the fact that the CEMEPS model was not equipped to deal with this factor resulted in several false positives that we have, for the purposes of this study, classed as hypothetical experiments. This was necessary in order to create enough cases to enable the creation of the RFM. Because of this, it is not possible to create an alternative RFM that omits these hypothetical cases and hence allows a quantitative assessment of fetch. It is our intention instead to consider the substantial challenge of incorporating the effects of sea ice into the storm surge inundation model.

Acknowledgments

This work would not have been possible without the insights and inspiration provided by Anne Jensen of the Ukpeagvik Inupiat Corporation, Glenn Sheehan of the Barrow Arctic Science Consortium, Gina Sturm of the NWS station in Barrow, Alaska, and Ron Brunner of the University of Colorado, Boulder, Colorado. Data and engineering information provided by Deirdre Ginter at the U.S. Army Corps of Engineers in Anchorage (CEPOA-EN-CW-HH) and Kari Sauers at Evans–Hamilton Inc. has been very useful. Sheldon Drobot and Jim Maslanik provided the information on fetch and trends in ice seasons. Three anonymous reviewers provided valuable advice on refinements to the approach and the manuscript. Finally, the importance of the participation, support, and interest of the people of Barrow cannot be overestimated. This work has been sponsored by the Australian Research Council though Grant FF0348550, the National Science Foundation though Grant NSF OPP-0100120, and the National Oceanic and Atmospheric Administration through Grant NA16RP2543.

REFERENCES

  • Bromwich, D. H., and S. H. Wang, 2005: Evaluation of the NCEP–NCAR and ECMWF 15- and 40-yr reanalyses using rawinsonde data from two independent Arctic field experiments. Mon. Wea. Rev., 133 , 35623578.

    • Search Google Scholar
    • Export Citation
  • Brunner, R. D., A. H. Lynch, J. Pardikes, E. N. Cassano, L. Lestak, and J. Vogel, 2004: An Arctic disaster and its policy implications. Arctic, 57 , 336346.

    • Search Google Scholar
    • Export Citation
  • Cassano, E. N., A. H. Lynch, J. J. Cassano, and M. R. Koslow, 2006: Classification of synoptic patterns in the western Arctic associated with high wind events and temperature trends at Barrow, Alaska. Climate Res., 30 , 8397.

    • Search Google Scholar
    • Export Citation
  • Cassano, J. J., J. E. Box, D. H. Bromwich, L. Li, and K. Steffen, 2001: Evaluation of Polar MM5 simulations of Greenland’s atmospheric circulation. J. Geophys. Res., 106 , 3386733890.

    • Search Google Scholar
    • Export Citation
  • Comiso, J., 1999: Bootstrap sea ice concentrations for NIMBUS-7 SMMR and DMSP SSM/I, June to September 2001. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/nsidc-0079.html.].

  • Déqué, M., 2003: Continuous variables. Forecast Verification: A Practitioner’s Guide in Atmospheric Science, I. T. Jolliffe and D.B. Stephenson, Eds., John Wiley & Sons, Ltd., 97–119.

    • Search Google Scholar
    • Export Citation
  • Drobot, S. D., and J. A. Maslanik, 2003: Interannual variability in summer Beaufort sea ice conditions: Relationship to spring and summer surface and atmospheric variability. J. Geophys. Res., 108 .3233, doi:10.1029/2002JC001537.

    • Search Google Scholar
    • Export Citation
  • Dudhia, J., 1993: A nonhydrostatic version of the Penn State-NCAR mesoscale model: Validation tests and simulation of an Atlantic cyclone and cold front. Mon. Wea. Rev., 121 , 14931513.

    • Search Google Scholar
    • Export Citation
  • Flather, R. A., 1976: A tidal model of the northwest European continental shelf. Mem. Soc. Roy. Sci. Liege Ser., 6 , 10. 141164.

  • Grell, G. A., J. Dudhia, and D. R. Stauffer, 1994: A description of the fifth-generation Penn State/NCAR mesoscale model (MM5). NCAR Tech. Note NCAR/TN-398+STR, 122 pp.

  • Hume, J. D., and M. Schalk, 1967: Shoreline processes near Barrow, Alaska: A comparison of the normal and the catastrophic. Arctic, 20 , 86103.

    • Search Google Scholar
    • Export Citation
  • Lestak, L. R., W. F. Manley, and J. A. Maslanik, 2003: Point Barrow, Alaska, and vicinity bathymetry. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://arcss.colorado.edu/data/arcss031.html.].

  • Lestak, L. R., A. H. Lynch, E. N. Cassano, L. Xie, S. Bao, W. F. Manley, J. A. Maslanik, and M. Peng, 2006: CEMEPS flood model output, bathymetry, topography, wind and related GIS layers for Barrow, Alaska. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/arcss302.html.].

  • Lynch, A. H., and R. D. Brunner, 2007: The importance of context in climate change impacts assessment: Lessons from Barrow, Alaska. Climatic Change, 82 , 93111.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., S. McIlwaine, J. Beringer, and G. B. Bonan, 2001: An investigation of the sensitivity of a land surface model to climate change using a reduced form model. Climate Dyn., 17 , 643652.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., E. N. Cassano, J. J. Cassano, and L. R. Lestak, 2003: Case studies of high wind events in Barrow, Alaska: Climatological context and development processes. Mon. Wea. Rev., 131 , 719732.

    • Search Google Scholar
    • Export Citation
  • Lynch, A. H., and Coauthors, 2004a: Barrow climatic and environmental conditions and variations—A compendium (Tech. ed.). CIRES Rep., 124 pp.

  • Lynch, A. H., J. A. Curry, R. D. Brunner, and J. A. Maslanik, 2004b: Towards an integrated assessment of the impacts of extreme wind events on Barrow, Alaska. Bull. Amer. Meteor. Soc., 85 , 209221.

    • Search Google Scholar
    • Export Citation
  • Manley, W. F., L. R. Lestak, C. E. Tweedie, and J. A. Maslanik, 2005: High-resolution radar imagery, digital elevation models, and value-added GIS layers for collaborative research of environmental change at Barrow, Alaska. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://nsidc.org/data/docs/arcss/arcss302/.].

  • Marquardt, D. W., 1963: An algorithm for least-squares estimation of nonlinear parameters. J. Soc. Ind. Appl. Math, 11 , 431441.

  • Mellor, G. L., 1996: User’s guide for a three dimensional, primitive equation, numerical ocean model. Princeton University, Princeton, NJ, 38 pp.

  • Murphy, A. H., and E. S. Epstein, 1989: Skill scores and correlation coefficients in model verification. Mon. Wea. Rev., 117 , 572581.

    • Search Google Scholar
    • Export Citation
  • National Ocean Service, 1997: Coastal bathymetry of the Bering, Chuckhi, and Beaufort Seas, Washington, D.C. NOAA, digital media. [Available online at http://alaska.usgs.gov/science/biology/walrus/index.html.].

  • Peng, M., L. Xie, and L. J. Pietrafesa, 2004: A numerical study of storm surge and inundation in the Croatan–Albemarle–Pamlico Estuary System. Estuarine Coastal Shelf Sci., 59 , 121137.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. O., and J. A. Curry, 1997: Role of radiative transfer in the modeled mesoscale development of summertime arctic stratus. J. Geophys. Res., 102 , 1386113872.

    • Search Google Scholar
    • Export Citation
  • Pinto, J. O., J. A. Curry, and A. H. Lynch, 1999: Modeling clouds and radiation for the November 1997 period of SHEBA using a column climate model. J. Geophys. Res., 104 , 66616678.

    • Search Google Scholar
    • Export Citation
  • Rinke, A., and Coauthors, 2006: Evaluation of an ensemble of Arctic regional climate models: Spatiotemporal fields during the SHEBA year. Climate Dyn., 26 , 459472.

    • Search Google Scholar
    • Export Citation
  • U.S. Department of Commerce, 2001: 2-minute Gridded Global Relief Data (ETOPO2). NOAA/NGDC, digital media. [Available online at http://www.ngdc.noaa.gov/mgg/fliers/01mgg04.html.].

  • USGS, 1997: Chukchi Sea Bathymetry, Anchorage, AK. Biological Science Office, Alaska Science Center, Anchorage, AK, digital media. [Available online at http://www.absc.usgs.gov/research/walrus/bering/bathy/.].

  • USGS, 1999: National Elevation Dataset. EROS Data Center, U.S. Geological Survey, Sioux Falls, SD, digital media. [Available online at http://ned.usgs.gov/Ned/ned.html.].

  • Walsh, J. E., and W. L. Chapman, 2001: Arctic and Southern Ocean sea ice concentrations. National Snow and Ice Data Center, Boulder, CO, digital media. [Available online at http://www.nsidc.org/data/g00799.html.].

  • Xie, L., and L. J. Pietrafesa, 1999: System-wide modeling of wind and density driven circulation in Croatane-Albemarlee-Pamlico Estuary system. Part 1: Model configuration and testing. J. Coastal Res., 15 , 11631177.

    • Search Google Scholar
    • Export Citation
  • Xie, L., L. J. Pietrafesa, and M. Peng, 2004: Incorporation of a mass-conserving inundation scheme into a three-dimensional storm surge model. J. Coastal Res., 20 , 12091223.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Three model domains were nested to produce the CEMEPS simulations shown here.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 2.
Fig. 2.

Comparison of PMM5-simulated 10-m wind speed (m s−1, black) and direction (°, gray) at the closest grid point to Barrow to the National Weather Service observations at Barrow for the (a) October 1963 case, (b) September 1986 case, and (c) July 2003b case. The solid line shows PMM5 winds and the dashed line shows observed winds.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 3.
Fig. 3.

Comparison of PMM5-simulated 10-m wind speed (m s−1, black) and sea level pressure (hPa, gray) at the closest grid point to Barrow to the National Weather Service observations at Barrow for the November 1950 case. The solid line shows PMM5 winds and the dashed line shows observed winds.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 4.
Fig. 4.

Maximum flood debris line of the October 1963 storm as mapped by HS67 (white line) and as simulated by the PMM5 and CEMEPS (black fill). Note that modeled inundation covered most of the spit, which coincides with the October event. Inundation follows the HS67 flood debris line near the Dew Line station and ARL. Topography has changed between 1963 and 2002 at the landfill, at Isatquaq Lagoon (the bridge road is blocking flooding into the fresh water source), and Dredge Harbor. Note that there is no flooding over the beach berm near ARL. Flooding near the town of Barrow and Browerville stays close to the shoreline because of the higher topography in that area.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 5.
Fig. 5.

Time series of flooding as simulated by the PMM5 and CEMEPS for the October 1963 storm.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 6.
Fig. 6.

Comparison between the tide gauge and simulated flooding for the September 2003 storm.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 7.
Fig. 7.

Scatterplot between the flooding simulated by the physical model work flow (PMM5 and CEMEPS) and the RFM having the highest correlation in Table 4. Eighteen storms applied in the fit are plotted with dots and three storms not used in the fit are plotted with squares.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Fig. 8.
Fig. 8.

Single perturbation sensitivities in the flooding of the RFM having the highest correlation in Table 4. Storms are plotted using the numbering from Table 1.

Citation: Monthly Weather Review 136, 3; 10.1175/2007MWR2121.1

Table 1.

Forecast quality summary. This table shows the sea level pressure (SLP) and 10-m wind speed bias, RMSE, MSESS, and correlations between the time series of observations and value at the closest grid point in MM5 (r). Here n is the number of time steps used in the analysis and the simulation duration is indicated in the form of hour (hh).day (dd).month (mm)–hh.dd.mm.

Table 1.
Table 2.

Definitions of forcing variables derived for each storm.

Table 2.
Table 3.

Forcing variable values and resulting flooding (m) simulated by the physical models (PMM5 and CEMEPS) for each storm.

Table 3.
Table 4.

Correlation coefficients (r) between the flooding simulated by the physical models (PMM5 and CEMEPS) and simulated by the RFM based on 21 storm cases. All sets of forcing variables with probability more than 25% to have the highest correlation within its RFM group are listed. A maximum of three forcing variables is permitted by the number of flooding cases in this application.

Table 4.
Save