Andrea Molod Global Modeling and Assimilation Office andrea.molod@nasa

1. Localized Characterization of Winter Storms in the New York Metropolitan Area and Statistical Prediction of Associated Storm Surge Andrea Molod Global Modeling and Assimilation Office andrea.molod@nasa.gov With: Haydee Salmun, Frank Buonaiuto, K. Wisniewska (student) - Hunter College of CUNY NASA Goddard Space Flight Center GLOBAL MODELING AND ASSIMILATION OFFICE

2. Outline • Context and motivation • Identifying and characterizing winter storms • Storm impacts - beach erosion and storm surge • Statistical prediction of surge

3. Alberta East Coast Nevada Colorado FL/Bahamas Gulf Context and Motivation • East Coast Cool-weather Storms (ECCS) – storms that impact the northeast • “Nor’easter” (“Hatteras Bomb”) • “Alberta Clipper” • “Gulf Low” • “Colorado Low” • “Chatanooga Choo Choo” • ~1000 km dia. • warm and cold fronts • major weather producers Regions of Cyclogenesis (Davis and FitzGerald, 2004)

4. Context and Motivation Kocin and Uccellini 2004: Estimated area (X 103 mi2) and population (in millions, from the 1999 census) affected by snowfall accumulations of 10 in. (25 cm) and greater Feb 27 – March 7, 1717 –series of storms January 28-? ,1772 - 3 ft in Baltimore – (Washington-Jefferson Storm) February 2-4, 1886 March 27-28, 1891 March 15-18, 1892 February 8, 1899 February 11-13, 1899 February 16-18, 1900 December 22-23, 1908 January 27-28, 1922 - (Knickerbocher Storm) January 29-30, 1930 December 17, 1932 February 7, 1936 March 28-29, 1942 1950 Nov. 25–27 18–19 Mar 1956 28,588 32.8 14–17 Feb 1958 126,004 53.8 18–21 Mar 1958 62,103 40.7 2–5 Mar 1960 133,734 53.9 11–13 Dec 1960 74,528 48.0 18–21 Jan 1961 62,260 43.0 2–5 Feb 1961 112,171 50.3 March 5-7, 1962 11–14 Jan 1964 110,258 48.8 29–31 Jan 1966 122,452 23.8 23–25 Dec 1966 83,389 18.1 5–7 Feb 1967 50,896 44.8 8–10 Feb 1969 66,440 31.2 22–28 Feb 1969 48,370 10.3 25–28 Dec 1969 131,351 25.0 18–20 Feb 1972 140,869 24.5 1977Jan. 28–29 19–21 Jan 1978 161,583 50.9 5–7 Feb 1978 120,490 47.6 17–19 Feb 1979 56,923 31.5 6–7 Apr 1982 76,839 22.5 10–12 Feb 1983 111,129 51.4 21–23 Jan 1987 132,772 34.9 25–26 Jan 1987 38,008 11.5 22–23 Feb 1987 28,276 16.6 November 11, 1987 12–14 Mar 1993 212,594 59.9 8–12 Feb 1994 54,951 39.0 2–4 Feb 1995 97,971 29.9 6–8 Jan 1996 137,918 56.6 31 Mar–1 Apr 1997 32,021 13.0 24–26 Jan 2000 59,567 19.7 30–31 Dec 2000 56,484 28.0 February 15-18, 2003 Feb 11-13, 2006 Dec 16-20 2009 Feb 5-10 2010 Feb 9-10 2010 Knickerbocker Snowstorm (NOAA Photo Library). • Blizzard (Mar 13 – 14, 1993) • Coastal Flooding (Jan 04 – 5, 1994) • Ice Storm (Mar 03 – 5, 1994) • High Winds (Oct 06, 1995) • Flooding (Oct 19 – 20, 1996) • Heavy Snow (Feb 14 – 18, 1996) • Heavy Rain (Oct 08 – 09, 1996) • Extreme Wind Chill (Jan 25 – 27, 2000) • Freeze (Oct 16 – 17, 2002) http://www.erh.noaa.gov/lwx/Historic_Events/snohist.htm

5. Characterization of East Coast Cool-Weather Storms • Coastal damage reports (Mather, 1964) • Surface weather maps (Colucci, 1976, Hayden, 1981) • Records of wave heights (Carter & Draper, 1988, Davis, et al., 1993) • Water level data (Zhang, et al., 2000) • Reanalysis (NCEP/NCAR) data (Hirsch et al., 2001, DeGaetano, 2008) • -------> Present Study: • Meteorological data from NOAA Buoys (NDBC archive): • Advantages: Spatially focused, relevant for assessing local impacts • Disadvantages: Eulerian approach (not always “seeing” storm center)

6. Identification and Characterization of ECCSs Data Buoys: atmospheric data (air temperature, pressure and winds) and water surface data (wave field) Precipitation: Shirley, MacArthur Airport & Montauk, LI. (NCDC Surface Data) Beach erosion: Tiana, Shinnecock, East Hampton, LI.

7. Identification of ECCSs - Method • Compute multi-year average and standard deviation of sea level pressure. • Storm is identified as an event when the pressure is less than two standard deviations below the mean, the minimum lasts for more than 4 hours and is separated from another event by at least 20 hours. • Exclude tropical storms from record using data from National Hurricane Center lists of hurricanes and tropical storms. Confirm storm event using daily weather maps • Classify storms - 3 levels of intensity - using pressure tendencies.

8. Identification of ECCSs - Method NDBC 44004 – 195 ECCSs ‘91-’07 40 20 0 -20 -40 90 95 ‘00 ‘05 NDBC 44025 – 206 ECCSs ’91-’07 40 20 0 -20 -40 90 95 ‘00 ‘05 Time Series of Pressure Anomaly Possible Storm Events - Tropical or Extratropical: Tropical eliminated from National Hurricane Center data --> average of 11 - 12 storms per year - very good agreement with previous work (Hirsch, et al., 2001)

9. Characterization of ECCSs – Intensity Scale • Categorize according to “how quickly the low deepens”, (Pressure Tendency) • Compute mean pressure tendency and standard deviation • Three storm intensity classes – based on terciles of standard deviation

10. Typical ECCSs of Different Intensity Classes Approximate location of NBDC Sta. 44025 is indicated by a X on the maps x x x x x x Level 3 Level 1 Level 2

11. a) b) Station 44025 c) d) Station 44004 Annual and Interannual Variability Similar annual distribution of storm counts - max in Jan, min in summer months Differences between interannual distributions (more variability at 44004) – record too short at 44025 to characterize ENSO or NAO related variability

12. Characterization of ECCSs – Storm Averages *indicates that data are for the period 2001-2007

13. Local Impacts – Coastal Erosion and Storm Surge Storm Waves 10/21/04 – 10/27/04 Hs = 3.8 to 4 m

14. Local Impacts – Coastal Erosion Tiana Beach Westhampton Beach Sand Transport SUNYSB COAST institute beach profile measurements

15. Local Impacts – Coastal Erosion Survey Period: September 23, 2004 – November 23, 2004 (Measurements at irregular intervals -- several weeks to several months)

16. Local Impacts – Coastal erosion

17. Local Impacts – Coastal Erosion • Morphology changes do not match intensity scale: WHY? • Surveys not timed with arrival/departure of storm • Water levels (astronomical tides, surge) not considered • Impact of background barrier morphology not known

18. Local Impacts of ECCSs – Storm Surge November 1950 La Guardia Airport FDR Drive December 1992 Ref: Bloomfield, J., M. Smith and N. Thompson, 1999. Hot Nights in the City. Environmental Defense Fund, NY.

19. Local Impacts: Storm Surge Storm Surge = observed water level – predicted astronomical tide Relationship between intensity and surge is robust

20. Local Impacts: Storm Surge • Statistical Models based on obs and/or relationships derived from simplified theory: • Tancredo (1958) - Regression equation relating significant wave height and storm surge, significant wave heights derived from simplified theory that relate wave heights to winds. • Harris (1962) - regression equation based on a “meteorological factor”, chosen based on linearized two-dimensional hydrodynamic equations. • Pore et al. (1974) - Regression equation (for New York) based on atmospheric pressure at eight grid points with time lags ranging from 0 to 6 hours. • DeGaetano (2008) - extreme surge events based on storm strength (criterion based on meteorological conditions) • Dynamical models: • ET-SURGE - NOAA’s forecast model for extratropical storm surge (Ji et al., 2010). Operational forecasts include a simple bias correction. • ADCIRC - Advance Circulation Model for Coastal Ocean Hydrodynamics of Luettich et al. (1992). Uses steady state, barotropic equations, grid designed for simulation of flooding during storm events (Colle et al., 2008). Run at SUNYSB. • ECOM - Estuarine Coastal and Ocean Model of Blumberg and Mellor (1987). Run by the New York Harbor Observing and Prediction System.

21. Example Dynamical model Sequence (ADCIRC) • MM5 meteorological fields drive WWIII (NOAA) • WWIII and MM5 fields drive SWAN (Delft Univ. Tech) • SWAN drives ADCIRC WAVEWATCH III

22. Local Impacts: Storm Surge (Colle et al., 2008) --> National Weather Service (NWS)’s extratropical storm related flood warnings during 2002-2006 had false-alarm rate of 85% Modern NWP forecast lead times are >10 days or longer (16 days from NCEP) - would like storm surge forecasts at longer lead times. Regional downscaling of climate forecasts (IPCC AR4) raises the possibility of driving climate forecasts of storm surge. All motivation for the development of a new statistically based model based on NDBC buoy measurements and storm surge measurements at The Battery, NY.

23. Statistical Model • Regression analysis for “storm maximum storm surge” • Used measurements at NDBC station 44004 and 44025 as predictors. Tested linear and non-linear combinations of: • Minimum pressure, pressure tendency, wind speed, wind gustiness, significant wave height, storm duration. • Found for predictions of “storm maximum storm surge”: • SSMAX based linearly on significant wave height alone is statistically indistinguishable from any other combination at both buoys • Using data from the closer buoy (44025) produces a statistically better prediction of SSMAX.

24. Statistical Model ; (meters) with RMS error 0.145 m. SSMAX = “storm maximum” storm surge, is the maximum value of storm surge reached during the storm period H = storm composite significant wave height, is the average of the ‘top third’ largest significant wave heights during the storm -> Evaluate predictive capability of this statistical model

25. Statistical Model - Data used for Evaluation Retrospective forecasts (NAM-WRF & NOMADS) of sea level pressure at location of NDBC Station 44025 for the period Feb 2005 - Dec 2008 Retrospective forecasts of significant wave heights from NOAA’s WAVEWATCH IIITM output at location of NDBC Station 44025 NOAA ET-SURGE standard forecast of surge values at The Battery, and NOAA operational forecasts Observed storm surge calculated from water level data at The Battery for the period 1959 – 2008 obtained from NOAA Results based on 41 storms The Battery NDBC Station 44025 Locations of The Battery and NDBC station 44025

26. Statistical Model - Method used for Evaluation Sea Level Pressure at Buoy 44025 Threshold algorithm to identify storms List of Storms Wind, wave data at Buoy 44025 Compute storm composite values Storm surge (computed) at The Battery Statistical Model for Storm-max Surge at The Battery Perform regression analysis Verify storm event in forecast NAM/NOMADS forecasted Sea Level Pressure List of Test Storms WWIII Forecasted fields at Buoy 44025 Compute storm composite values Storm surge (computed) at The Battery NOAA Forecasted Storm surge at The Battery Observed, NOAA, Statistical Storm-max surge values Use regression to estimate storm-max surge

27. Statistical Model - Evaluation • statistical forecast (STAT FCST) • NOAA ET-SURGE output (NOAA ET) • NOAA operational forecast (NOAA ETANOM) • observations at The Battery (OBS) Notes: NOAA ET-SURGE predictions are from direct (archived) output from ET-SURGE. NOAA’s operational forecast (NOAA ETANOM) consists of the ET-SURGE output and an error correction (not in archived data), computed as the 5-day running mean of the previous days’ errors of ET-SURGE output. Variability in the anomaly correction is an indication of the variability in the error of ET-SURGE model output (range of anomaly: -0.13 m – 0.28 m)

28. Values of OBS SSMAX are always positive • SSMAX estimates using STAT FCST are always positive • OBS SSMAX range: 0.1 m – 0.92 m • Range of SSMAX from STAT FCST: 0.17 m – 0.83 m

29. SSMAX from ET-SURGE model output can be negative (4 cases) • SSMAX from ET-SURGE operational fcst negative only once • Range of SSMAX from ET-SURGE model output: -0.28 m – 0.72 m • Range of SSMAX from ET-SURGE operational fcst: -0.19 m – 0.84 m

30. Differences among the different estimates of SSMAX • SSMAX from NOAA-ET is underpredicted by 0.25 m (  =0.11 m) • SSMAX from NOAA-ETANOM is underpredicted by 0.14 m (  =0.11 m) • Mean error of SSMAX from STAT FCST is 0.05 m (  =0.15 m)

31. Statistical Model - Results of Evaluation • From statistical analysis for 12-hr lead time: • error in NOAA-ET SSMAX is greater than the error in STAT FCST SSMAX at > 95% significance level • error in SSMAX from NOAA-ETANOM is statistically indistinguishable from error in STAT FCST of SSMAX

32. Statistical Model - Results of Evaluation • Characterization of the STAT FCST error: • Can distinguish between error in STAT FCST due to failure of the regression relation and error in STAT FCST due to wave height forecast errors -- use standard deviation of the regression as the metric. • * STAT FCST tends to slightly underpredict or overpredict on average • underpredictions --> errors in forecasted significant wave heights • overpredictions --> failure of the regression relation • * NOAA forecasts, both with and without anomaly, tend to underpredict

33. In Conclusion: • we established that the statistical method for predicting “storm-maximum” storm surge (SSMAX) is robust • in two thirds or more of the cases, using predicted values of wave height does not have a negative impact on the statistical estimate of SSMAX • evaluation of 12-, 24- and 48-hr lead time STAT FCST predictions showed it to outperform NOAA’s ET-SURGE model forecasts and be equivalent in skill to NOAA’s (bias-corrected) operational forecast • the lead time of NOAA’s operational forecast is limited due to the form of the bias correction – STAT FCST does not have this limitation and we propose that it could be used in conjunction with ET-SURGE forecasts. Salmun, H., A. Molod, F. Buonaiuto, K. Wisniewska and K. Clarke (2009): East Coast Cool-weather Storms in the New York Metropolitan Region. Journal of Applied Meteorology and Climatology, 48, 11 2320-2330. Salmun, H., A. Molod, K. Wisniewska and F. Buonaiuto (2010): Statistical Prediction of the Storm Surge Associated with Cool Weather Storms at The Battery, New York. Journal of Applied Meteorology and Climatology; In Press

34. Future Work Expand spatial scope – how robust spatially is our regression? • Evaluation of validity of regression coefficients for predicting SSMAX at The Battery using wave heights from other nearby buoys • Evaluation of validity of regression coefficients for predicting SSMAX at other water level measurement locations

35. Future Work - Characterization of STAT FCST Error Typical sea level pressure fields from NASA’s MERRA re-analysis (~ 0.5° resolution) showing typical patterns of storms for which error in STAT FCST SSMAX are small (top – stronger storms with centers passing directly over area) and those for which errors are large (bottom – weaker storms, centers passing farther away x x x approximate location of NDBC station 44025 approximate location of storm’s center Also: re-evaluate role of other predictors (we were surprised by what was NOT important as a predictor)