1. Estimating Structural Reliability Under Hurricane Wind Hazard : Applications to Wood Structures Balaji Rajagopalan, Edward Ou, Ross Corotis and Dan Frangopol Department of Civil, Environmental and Architectural Engg. University of Colorado Boulder, CO Probabilistic Mechanics Conference Albuquerque, NM July 26-28

2. Acknowledgments Funding for this work was provided by NSF grant SGER (CMS-0335530) Discussions with Prof. Ellingwood, Dr. Simiu and Dr. McGuire are thankfully acknowledged

3. Motivation Insured losses in the US from “natural hazards” reached $22 billion in 1999 Second largest loss during 1990’s - $26 billion in 1992 due to Hurricane Andrew (in Florida and Louisiana)Topics (2000 - Munich) The U.S. House of Representatives, is working on bill H.R. 2020 - Hurricane, Tornado and Related Hazards Research Act, to promote : “inter-disciplinary research in understanding and mitigating windstorm related hazard impacts new methodologies for improved loss estimation and risk assessment”

4. Property Loss due to Hurricanes in the US

5. Motivation (contd..) (i) Often, structural reliability is estimated in isolation of realistic likelihood estimates of hurricane frequencies and magnitudes. (ii) Knowledge of year-to-year variability in occurrence and steering of hurricanes in the Atlantic basin is not incorporated in structural reliability estimation. (iii) The estimation of losses is purely empirical, based on the wind speed and no consideration of structural information. ( For example, a new structure and a 25 year old structure are assumed to have the same probability of failure for a given wind speed.) (iv) The life cycle cost of structures is also not considered  substantial misrepresentation of losses and consequently sub-optimal decision making.

6. Hurricane Tracks - 1997 1997 was strongest El Nino year  Fewer hurricanes

7. Hurricane Tracks - 2000 2000 was a strong La Nina year  more hurricanes

8. - La Nina conditions are almost a reverse of the El Nino conditions. - The ENSO phenomenon is irregular occurring every 3 ~ 8 years. - Impacts global weather and climate

9. An index of ENSO (based on Sea Surface Temperatures and Sea Level Pressures in the tropical Pacific Ocean) Notice the Evolution of Different El Nino and La Nina events

10. Global Impacts of ENSO

11. ENSO phenomena impacts climate over the US by modulating The winter time jet stream

12. Notice more Hurricanes during La Nina years and vice-versa Notice negative correlations between #of Atlantic Hurricanes and SSTs Over Eastern Tropsical Pacific  La Nina pattern

13. Motivation (contd..) (i)Clearly, large scale climate phenomenon (e.g., ENSO) has a significant impact the frequency and strength of hurricanes. (ii)Incorporating this information is key to realistic estimation of structural reliability (iii) Thus, need to develop a framework that will facilitate this.

14. Proposed Framework

15. Structural Reliability Estimation Steps: 1.Generate scenarios of maximum wind speeds conditioned on large-scale climate information. - i.e. simulate from conditional PDF f(wind speed | climate) “Load Scenarios” 2.Scenarios generated for different large-scale climate states (El Nino, La Nina) 3. Convert the maximum wind speed to 3-second gust (gust correction factor, Simiu, 1996) 4.“convolute” with fragility curves to estimate the failure probability – consequently the reliability 5.Considered 25 year time horizon, wooden structures

17. Data for wind scenario 1.Historical Hurricane track data from http://www.nhc.noaa.gov http://www.nhc.noaa.gov 2.Get the historical track for the region of interest (2deg X 2deg box over N. Carolina) 3.Estimate the annual maximum hurricane wind speed for the grid box (wind speed) 4. Climate information (e.g., El Nino index) is obtained from http://www.cdc.noaa.gov (climate index)http://www.cdc.noaa.gov 5.Simulate scenarios from the conditional PDF f(wind speed | climate)

18. Nonparametric Methods Kernel Estimators (properties well studied) Splines Multivariate Adaptive Regression Splines (MARS) K-Nearest Neighbor Bootstrap estimators Locally Weighted Polynomials http://civil.colorado.edu/~balajir/

19. Nonparametric Methods A functional (probability density, regression etc.) estimator is nonparametric if: It is “local” – estimate at a point depends only on a few neighbors around it. (effect of outliers is removed) No prior assumption of the underlying functional form – data driven

20. Basic Philosophy Find K-nearest neighbors to the desired point x Fit a polynomial (or weighted average) to the neighbors  recovers the underlying PDF (nonparametric density estimation) If the data is X and Y then fitting a polynomial to the neighbors  recovers the underlying relationship (nonparametric regression) Number of neighbors K and the order of polynomial p is obtained using GCV (Generalized Cross Validation) – K = N and p = 1  Linear modeling framework. Several variations to this are possible

21. Applications to date…. Monthly Streamflow Simulation Multivariate, Daily Weather Simulation Space and time disaggregation of monthly to daily streamflow Monte Carlo Sampling of Spatial Random Fields Probabilistic Sampling of Soil Stratigraphy from Cores Ensemble Forecasting of Hydroclimatic Time Series Downscaling of Climate Models Biological and Economic Time Series Exploration of Properties of Dynamical Systems Extension to Nearest Neighbor Block Bootstrapping -Yao and Tong

22. Logistic Map Example k-nearest neighborhoods A and B for x t =x* A and x* B respectively 4-state Markov Chain discretization

23. K-NN Local Polynomial

24. ENSO characterization Tropical Pacific Ocean Sea Surface Temperature based index called (NINO3 index) is used to characterize ENSO index value > 0.5 indicates El Nino years values < -0.5 are La Nina years

25. ENSO index Joint PDF of Max. Wind Speed and ENSO index La Nina Years El Nino Years All Years Neutral Years Histogram of #of Hurricane Occurrences over N. Carolina – With Respect to Large-scale Climate

26. ENSO index Wind Speed Joint PDF of Max. Wind Speed and ENSO index Notice non-Gaussian features

27. ENSO index Joint PDF of Max. Wind Speed and ENSO index Conditioned on ENSO index Value of –1 (solid line) (La Nina) 1(dashed line) (El Nino) Notice non-Gaussian features Conditional PDF of Max. wind speed

28. Joint PDF of Max. Wind Speed and ENSO index All Year Simulations CDFs from unconditional simulations CDF of - Historical data in (purple) -El Nino years in (red) -La Nina years in (blue)

29. Joint PDF of Max. Wind Speed and ENSO index CDFs of Wind Speeds conditioned on ENSO Red line is the historical CDF of El Nino years Blue line is the historical CDF of La Nina years Notice the differences at Lower speeds

33. Failure Due to Panel Uplift

34. Failure due to Roof-to-wall Separation

35. Gust Effect - Failure due to Panel Uplift

36. Summary Integrated (Interdisciplinary) framework to estimate infrastructure risk due to hurricane hazard is presented Nonparametric method is used to generate hurricane wind scenarios conditioned on large- scale climate state (El Nino, La Nina etc.) Large-scale climate state appears to impact the number of hurricanes, maximum wind speed and consequently, infrastructure risk (over N. Carolina)

37. Further Extensions –Extension to other types of structures (concrete, bridges etc.) –Investigate gust correction factors for hurricane winds –Study the impact of time-varying infrastructure risk estimation on the loss estimates –Incorporate other relevant climate information for Hurricane occurrence and steering (such as, North Atlantic Ocean and Atmospheric conditions) –Integrating life-cycle cost for optimal decision making on maintenance and replacement