Project Description

I Introduction

Natural hazards such as hurricanes lead to major disasters at sea and in the regions of landfall. These severe storms are characterized by intense precipitation, flooding and high winds (at a minimum 18 m/s). Major (or intense) hurricanes (i.e. those of category 3-5 on the Saffir-Simpson hurricane damage potential scale) in the Atlantic Ocean (Gray,1990) represent one of Nature’s most destructive forces, sometimes causing tremendous loss of life and damage to ecology, property, wetlands and coastal estuaries (Gray and Landsea, 1992; Pielke and Landsea, 1998). These intense storms can contain maximum sustained wind speed of greater than 50 m/s with an accompanying storm surge of more than 2.7m (Wilson, 1999). Clearly, the winds and the flooding associated with the hurricanes have a significant impact on structural safety/reliability in the regions that are prone to hurricane (e.g., the East coast of the United States).

There has been an increasing trend of economic losses from recurring natural hazards, in particular, from hurricanes and other climate induced hazards. The potential for even greater losses over time is also increasing due to development. A recent report from Munich Reinsurnce (“Topics 2000”) shows that insured losses (those protected by insurance) in the United States from natural hazards reached $22 billion in 1999, the second largest loss during the 1990’s to the $26 billion of 1992, the year Hurricane Andrew struck Florida and Louisiana. Furthermore, during the past 50 years, 250 great natural catastrophes in the world claimed the lives of over 1.4 million people and caused economic losses (insured and uninsured) of $960 billion. The increasing socio-economic costs and the rising potential are placing unprecedented strain on public and private resources.

During Fall 2001, following a two-day workshop titled “ Ten Most Wanted : The search for solutions to Reduce Recurring Losses from Natural Hazards”, ASCE and the Institute for Business & Home Safety (IBHS) with support from HUD and USGS prepared a very interesting and thoughtful report (IBHS, 2001). In this, they identify five main Climate Related Natural Hazards (CRNH) (the sixth being earthquakes) that pose substantial risks to infrastructure safety – hurricanes and severe windstorms, floods, hail, severe winter storms and wildfire. They also recommend that “holistic” solutions are the best ways to meet the challenge faced by communities throughout the nation as they seek to reduce infrastructure vulnerabilities. The U.S. House of Representatives, with substantial support from its members, is working on bill H.R. 2020 on similar lines, also known as the “Hurricane, Tornado and Related Hazards Research Act”; to promote inter-disciplinary research in understanding and mitigating windstorm related hazard impact in the next 10 years. Two of the seven research and development elements of this bill, specifically mentions the need for research in developing “new methodologies for improved loss estimation and risk assessment systems for predicting and evaluating damaging windstorm impacts and for identifying, evaluating, and reliably characterizing windstorm hazards”.

Recognizing this need, the GOALI research proposed here aims to develop an integrated framework to combine the understanding of spatio-temporal variability of hurricane occurrences and physical aspects of infrastructural stability to better estimate/predict infrastructure reliability and consequently, the losses. This will enable optimal planning and management of resources in the short (e.g., disaster preparedness) and long term (e.g., town planning) by public and private sector agencies.

The proposal is organized as follows: A case for GOALI is first made. Objectives and expected significance of this research are then presented, background on hurricanes and structural reliability are next described, followed by the proposed approach. Finally, we present “proof of concept” from our ongoing SGER to demonstrate the feasibility of the proposed research.

II Why GOALI?

Considerable research is being focused on understanding hurricane variability (by climate scientists) and also on developing improved structural risk assessment techniques (by structural engineers). There is little communication between these two groups which has lead to the following drawbacks:

(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. In particular, this information can be very useful in probabilistic forecast of storm tracks and magnitudes in any given year.

(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. The life cycle cost of structures are also not considered. These can lead to substantial over and under estimation of losses and consequently, sub-optimal decision making.

(iv) The loss data is available with insurance companies and not in the public domain. This makes it very difficult to study the impact of any improved structural reliability model(s).

Clearly, to address these drawbacks, in particular, to make the research in structural reliability estimation due to hurricanes relevant and useful, an industry partner is crucial. From the industry stand point a better loss estimation required a realistic structural reliability model. This mutual convergence of interests and needs naturally led to the proposed GOALI.

This GOALI is in partnership with the company, Risk Engineering (http://www.riskeng.com). Risk Engineering has extensive experience with hurricane track simulation and loss estimation and their clients include several fortune 500 companies. The industry partner will help with testing the impact of structural reliability model in loss estimation. Furthermore, we will work jointly in evaluating and testing the hurricane track simulation model used by Risk Engineering and also the proposed track model proposed in this research.

III Objectives and Expected Significance

As described in the introduction, there is a real need for improved loss estimation. This calls for an integration of knowledge from climate science, structural engineering and loss estimation (private sector). This motivates our proposed research, the objectives of which are to:

1. Develop an improved hurricane track model that can generate synthetic tracks, which will preserve the relative frequencies of observed landfalls, wind speeds, precipitation and flooding. The model should also be able to simulate tracks conditioned upon various climate indicators [e.g., El-Nino Southern Oscillation, North Atlantic Oscillation etc.] to estimate dynamic (time varying) risk in any given year (i.e., in a predictive mode);

2. Develop methods for combining loads from multiple sources (i.e., wind speed, precipitation and flooding) that are essential to structural reliability estimation;

3. Develop failure states that are physically based and associated with the estimated loads. Thereby, providing the capability to estimate structural failure probability;

4. Develop a module for estimating stationary and time varying structural system reliability, based on material degradation, structural failure probability and life cycle;

5. Develop a loss estimation module that will incorporate the structural system reliability estimates from above.

The structural reliability model will be developed for the dominant stock of structures (e.g.,

wood, concrete etc.) in South Eastern US. The proposed framework will also be tested over

this region of US.

The industry partner will help with objective 5 and also provide their hurricane track model

for verification and testing (objective 1)

The following outcomes are expected:

1. Synthetic hurricane track simulation model(s) capable of generating tracks and associated precipitation and flooding potential, conditioned upon the large-scale climate state (e.g., El Nino, state of the North Atlantic Ocean etc.). This will provide probabilistic forecast of climate information relevant to structural reliability.

2. Structural failure models for different type of structures (e.g., wood and concrete) and consequently, estimation of failure probabilities.

4. Model for estimating time-varying structural system reliability. Consequently, life cycle cost and optimal planning.

5. Methodology to classify structures based on their system reliability;

6. Loss estimation model that includes estimates of structural system reliability.

All these outcomes will be extremely helpful for their contributions to the knowledge base and intellectual growth in the area of civil infrastructure assessment as well as the reduction of risks induced by hurricanes. This will enable advances in the management of our nation’s civil infrastructure by developmental agencies, state planners, disaster mitigation agencies, and private sector insurers. The outcomes from this study will supplement the HAZUS initiative by FEMA for enhancing the nation’s risk assessment capabilities under natural hazards. The framework developed here will complement the hurricane loss estimation and wind loss estimation module of HAZUS. We propose to work with FEMA and, consequently, the impact will be wide reaching in the public and private sectors.

IV Background

A review of information relevant to hurricanes and Bayesian structural risk estimation process is presented in this section. First, we briefly describe the connection between large-scale climate features (e.g. ENSO, NAO etc.) and hurricanes in the Atlantic basin. Then we briefly describe the hurricane track generation model that the PI and his collaborators have developed. Description of load combination analysis is next presented, followed by the Bayesian structural risk estimation methodology.

IV.1 Hurricanes and Climate

As noted earlier, tropical and extra-tropical cyclones (called hurricanes in the Atlantic basin) cause sever property damage and loss of life. The hurricanes are named by the National Hurricane Center when the storm reaches a maximum sustained wind speed of at least 18 m/s (Neumann et al., 1993). When the maximum sustained wind speed reaches 50 m/s the hurricanes get into the category 3-5 on the Saffir-Simpson hurricane damage potential scale (Gray, 1990). The major hurricanes can have a storm surge of greater than 2.7m (Wilson, 1999). Clearly, the wind speed is a major factor that causes structural damage and also the flooding due to the accompanying storm surge and precipitation also contribute to structural degradation.

Over the years, investigators have identified large-scale climatic factors, such as ENSO, QBO, Atlantic Sea Surface Temperatures (SSTs), and rainfall over Africa’s Sahel region, that appear to affect the year-to-year variability in tropical cyclone activity (Gray, 1984 and Gray, 1990; Goldenberg and Shapiro 1996; Shapiro, 1982; Shapiro, 1989; Shapiro and Goldenberg, 1998; Bove et al., 1998). Elsner and Kocher (2000) found links between tropical cyclone activity and the preceding winter state of the North Atlantic Oscillation (NAO). Recently, Rajagopalan et al., 2002 find that El-Nino (i.e. warming of western tropical Pacific Ocean) substantially reduces the number of storm days over the Gulf of Mexico and eastern USA relative to other years – this corroborates earlier studies. They also find that a warmer tropical North Atlantic sea surface during spring tends to enhance storms over the Gulf of Mexico and also in the Bermuda region in the following summer, relative to a cooler tropical North Atlantic. One of their interesting findings is that high pressure over the sub-tropical Atlantic Ocean (i.e. high phase of NAO) in the winter tends to favor lesser storms in general over the eastern Atlantic basin (which includes the east coast of US).

Mechanistically, all these climatic phenomena are believed to regulate tropical storm formation via their effects upon upper tropospheric wind shear (Landsea, 1998). Events that increase shear lead to a weaker hurricane season (i.e. less number of hurricanes) while events that lower wind shear make for an active hurricane season. Therefore, for realistic simulation of hurricane tracks and consequently, the risk estimates, these climate features have to be appropriately included (or conditioned upon) in the track simulation model.

III.2 Hurricane Track Simulation

There are several physical models (e.g., Kurihara et al., 1995; Marks, D.G., 1992) that are used for short term forecasting (hours to days) of hurricane tracks once the storm formation is identified. Such models are used by the National Hurricane Center to monitor hurricanes once they have formed and issue warnings. Such models are computationally intensive and run with observed physical data of the storm, Furthermore, they are expensive to run anddo not have the capability to generate “realistic” synthetic hurricane tracks that are key to estimating spatial risks and consequently, structural risk and reliability. Stochastic models are an attractive alternative. Not much work has been done in developing hurricane track simulation models that generate synthetic tracks. There are some models developed in the private sector – but these models are not public domain and often adhoc in nature. Very recently, the PI in collaboration with researchers at Columbia University has developed a Markov Chain model and a Track Segment model (Miller et al., 2002) for hurricane track simulations that reproduce quite well the relative frequency of hurricane occurrences over the entire Atlantic basin. A brief description of these two models is presented below.

Markov Chain Model

In this model the entire Atlantic basin is divided into grid boxes of 5 deg x 5 deg. Based on the historical hurricane tracks (this consists of all hurricane track information since 1900 – containing the location of the storm, wind speed, pressure every 6 hours right from the formation of the storm) at each grid box, transition probabilities are estimated – i.e. the probability of a storm in the box to move into adjacent grid boxes. The estimated transition probabilities at each of the grid boxes are used to generate synthetic tracks. This model performed well in terms of reproducing the relative frequencies of the historical storms over the Atlantic basin (Miller et al., 2002). The main drawback is its inability to generate the wind speeds and other attributes for which a separate stochastic model has to be developed and coupled with this track simulation model.

Track Segment Model

The second model developed in Miller et al., 2002 involves re-sampling segments of historical tracks to generate a synthetic track. Thus, the tracks generated will be a mixture of several historical tracks that are physically consistent. This method is very simple and elegant and has the ability to generate a rich mixture of tracks that are consistent with the historical tracks. We generated 1000 synthetic tracks from this model in the Atlantic basin and compare several statistics to historical values to assess the performance. We present a couple of results from Miller et al., 2002 to demonstrate the capability of the model. Figure 1 shows the cumulative distribution of the track lengths in 6-hour periods – essentially, it is the lifetime of the tracks. The solid line represents the distribution from the historical tracks, while the dashed lines are the lower 5th percentile, lower quartile, upper quartile and upper 95th percentile from bottom to top, respectively, from the simulated tracks.