Impact Modelling of Hurricane Sandy on the Rockaways | 1 Impact Modelling of Hurricane Sandy on the Rockaways | 2 Report type MSc thesis TU Delft Title Impact Modelling of Hurricane Sandy on the Rockaways Subtitle Relating high-resolution storm characteristics to observed impact with use of Bayesian Belief Networks Date 7 September 2014 Author name: Huub C.W. van Verseveld student no: 1365010 Programme/track: Hydraulic Engineering, Master of Science Civil Engineering Specialisation: Coastal Engineering Examination Committee Chairman Prof. Dr. Ir. M.J.F. Stive, Coastal Engineering (TU Delft) First Supervisor Dr. Ir. A.R. van Dongeren, Coastal Morphology (Deltares) Supervisor Dr. N.G. Plant, Center for Coastal & Watershed Studies (USGS) Supervisor MSc. Ir. W.S. Jäger, Applied Mathematics (TU Delft) Supervisor Dr. Ir. C. den Heijer, Coastal Engineering (TU Delft) In collaboration with Deltares U.S. Geological Survey Short summary Hurricane Sandy (2012), which made landfall in New Jersey on October 29th, made devastating impact on the East Coast of the USA and struck major parts of New York City, including the economic centre of Manhattan. The total damage (in the USA and Caribbean) is in excess of 100 billion US$ with estimates ranging between 78 and 97 billion US$ for direct damage and over 10 to 16 billion US$ for indirect damage due to business interruption (M. Kunz et al., 2013). Modelling impact (e.g. damage, fatalities) in the coastal zone due to hazardous storm events is a hardly explored practice. It is difficult to predict damage correctly where damage observations are scarce and the physical processes causing the damage are complex, diverse and can differ from site to site and event to event. Moreover, an increasing interest exists in getting insights in the uncertainty of prediction. This report explores on the possibilities in coupling physics-based hydraulic and morphodynamic modelling to the practice of impact mapping by using Bayesian Belief Networks (BBN’s). Using BBN’s enables to look at the impact in a probabilistic context, which fits well to the highly unpredictable and rare nature of hurricanes. The morphodynamic storm impact model XBeach (Roelvink et al., 2009) is proposed as one of the process-based models, since Hurricane Sandy pointed out that morphodynamic aspects can be of great importance for the amount of damage. Part of the Rockaway Peninsula, NY, which has been severely damaged by Hurricane Sandy’s surges, is used as case study. Only damage to residential buildings is considered in present study, in which delivering a proof of concept for the presented approach is the main goal. In general it can be concluded that the approach succeeded. Impact Modelling of Hurricane Sandy on the Rockaways | 3 Impact Modelling of Hurricane Sandy on the Rockaways | 4 Acknowledgements I am most grateful to my committee, consisting of Marcel Stive, Ap van Dongeren, Nathaniel Plant, Wiebke Jäger and Kees den Heijer, for helping me performing this thesis. Their knowledge, interests, insights, comments, criticism and optimism definitely made the results better and helped me in keeping track. Special thanks to Ap (Deltares) and Nathaniel (USGS), who gave me the opportunity and means to go on an adventurous and educational trip to Florida and New York. I really appreciated the hospitality and felt always valued, both at Deltares and USGS. In that perspective I also like to thank Cheryl Hapke, Soupy Dalyander, Dave Thompson and Joe Long, who helped me in finding my way in the States, facilitated in data, contacts and knowledge on computational modelling. Special thanks also go to Wiebke, Kees, and (in the beginning) Jaap van Thiel de Vries, who supported me during all phases of the process. Thanks, Wiebke, for the company in Saint Petersburg and all the interesting conversations and discussions; these were really helpful in structuring my thoughts. Furthermore, I’d really like to thank Joost den Bieman, who was always available and willing to answer my questions and Mohd Shahrizal Ab Razak from UNESCO-IHE, who helped me tremendously in finishing the last and largest XBeach run. Next to the above mentioned people, I had the privilege to speak to a lot of experts on varying matters; they were always interested and willing to help. To these people from amongst others Deltares, TU delft, VU Amsterdam, USGS, Office of the Mayor of NYC and FEMA I would like to say: thank you for your attribution. Impact Modelling of Hurricane Sandy on the Rockaways | 5 Impact Modelling of Hurricane Sandy on the Rockaways | 6 Abstract Hurricane Sandy (2012), which made landfall in New Jersey on October 29th, made devastating impact on the East Coast of USA and struck major parts of New York City, including the economic centre of Manhattan. The total damage (in the USA and Caribbean) is in excess of 100 billion US$ with estimates ranging between 78 and 97 billion US$ for direct damage and over 10 to 16 billion US$ for indirect damage due to business interruption (M. Kunz et al., 2013). Modelling impact (e.g. damage, fatalities) due to hazardous storm events is a hardly explored practice; especially in the coastal zones where predominantly wind induced hazards from sea (e.g. inundation, wave attack) cause the damage. It is difficult to predict damage correctly where damage observations are scarce and the physical processes causing the damage are complex, diverse and can differ for different sites and events. Moreover, an increasing interest exists in getting insights in the uncertainty of prediction. This report explores on the possibilities in coupling physics-based hydraulic and morphodynamic modelling to the practice of impact mapping by using Bayesian Belief Networks (BBN’s). Using BBN’s enables to look at the impact in a probabilistic context, which fits well to the highly unpredictable and rare nature of hurricanes. Part of the Rockaway Peninsula, NY, which has been severely damaged by Hurricane Sandy’s surges, is used as case study. Only damage to residential buildings is considered in present study, in which delivering a proof of concept is the main goal. The morphodynamic storm impact model XBeach (Roelvink et al., 2009) is proposed as one of the process- based models, since Hurricane Sandy pointed out that morphodynamic aspects can be of great importance to the amount of damage, especially for the barrier islands in front of the US coastlines. The hazards, predominantly coming from sea in the case of Hurricane Sandy, are propagated from large scale (100-1000 km) to the building level (1-10 m) with a nested routine which includes XBeach on the lowest scale level. With the use of extraction methods local hazard indicators are generated for every single residential building, which have been successfully used in combination with building type indicators to predict damage. 5300 observations of a qualitative damage assessment were enough to train a Bayesian Belief Network that is capable of reproducing the spatial pattern of the damage. Multiple analysis tools are available to analyse the quality of prediction and uncertainty quantification and it is possible with ease to visualize that in space with the use of an SQL database coupled to GIS software. In general it can be concluded that the presented approach succeeded. XBeach is capable of providing multiple local hazard indicators on the building level, which proved to having predictive capacity. The indicators “inundation depth”, “wave attack”, “flow velocity” and (to a lesser extend) “scour depth” give together much better predictions than they do alone. The implementation of XBeach therefore fits perfectly in the multi-hazard approach that Bayesian Belief Networks make possible. On the contrary, it must be said that setting up an XBeach model for high resolution simulations and relatively large areas (order of 10 km wide) is a time consuming job. The results of the XBeach model used for present study show a structural overestimation of storm conditions, which can be addressed to a poor calibration. This assumedly limits the predictive skill of the statistical BBN model. There is still a large scope for improvement. Some aspects of the approach have been studied in more detail; roughly, these aspects do either influence the predictive qualities of the individual indicators or have to do with the configuration of the statistical model, the Bayesian Belief Network. Considering the first category, it appeared that model resolution is not as important as expected. Indicator values based on model runs with grid cell sizes of 3x3 m2 in the urban areas give hardly better predictions than runs with grid cell sizes of 9x9 m2 where computational expenses are 25 times higher. Next to this aspect, the extraction method, in which hazard indicators are generated out of the XBeach output data, appeared to be very important for the predictive capacity of these indicators. The usage of (polygons of the) building perimeter outlines to determine buffer zones around these buildings works fine. Moreover, it can Impact Modelling of Hurricane Sandy on the Rockaways | 7 be concluded that the extraction formulations and the size of extraction buffer zones around the buildings can make a substantial difference for the predictive skill of the local hazard indicators. A courser grid asks for a larger buffer zone in order to prevent that the buildings are predicted to be non-flooded where they in reality have been flooded. Considering the statistical model part, using Bayesian Belief Networks gives the opportunity to relate the damage to multiple aspects instead of only one, which has great advantages over the market standard approaches (in which the majority only considers the depth-damage relation). Comparing the spatial distribution of the means of the conditional Probability Mass Functions (PMF’s) to the observed damage, it can be concluded that the established BBN’s are good in capturing the spatial variability in risk (given the event) per building.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages174 Page
-
File Size-