### CyberEye

# Rapid Risk Assessment

The Rapid Risk Assessor (RRA) within CyberEye, founded on an initial standalone executable prototype called Hakou, currently focuses on high fidelity hazard quantification supporting hurricane wave and surge, though the capability to model wind hazards is currently under development. While conventional approaches to hurricane risk assessment are based on analysis of data from historical storms and from simulation of design events or by adopting low-fidelity numerical models to predict hurricane responses, RRA relies on surrogate modeling for efficiently providing predictions of hurricane wave and surge responses from databases of existing high-fidelity numerical simulations, thus providing a single tool for both historical and non-historical event analysis. The foundation of the approach is the simplified description of each hurricane scenario through a small number of model parameters, corresponding to its characteristics at landfall, such as (i) reference location; (ii) track heading; (iii) central pressure; (iv) forward speed; and (v) radius of maximum winds. The scenarios in the database are then parameterized with respect to this model parameter vector and ultimately provide an input-output data set, where the output involves responses of interest such as storm surge or significant wave height over a large coastal region (represented typically by a large number of nodes) and possibly at different times during the hurricane history. A surrogate model is then built to approximate this input-output relationship. Though the initial version of Hakou relied upon a moving least squares response surface methodology for its surrogate model, the implementation into CyberEye now employs a Kriging metamodel for this purpose. Once the metamodel is tuned, it can efficiently provide predictions for any new hurricane scenario. To further reduce the computational burden, pertaining to both speed of execution and more importantly memory requirements, this approach is coupled with principal component analysis (PCA) as a dimensional reduction technique. This PCA implementation contributes to very large computational savings necessary for web-based platforms with minimal impact on the prediction accuracy.

As the overall RRA can be efficiently used to predict the hurricane response for any new scenario whose parameters lie within the range of its backside database, it readily facilitates a highly efficient probabilistic risk assessment. This is established by characterizing the uncertainty in the hurricane parameters through appropriate probability distributions, leading ultimately to quantification of hurricane risk as a probabilistic integral over the uncertain parameter space. This integral is then estimated using Monte Carlo simulation relying on the developed surrogate model for efficient implementation. An advantage of this approach is the fact that it can support our desire to form the initial database using high-fidelity numerical models to estimate the hurricane responses with very high-accuracy.

" -- Tracy Kijewski-Correa |

In the first application of this algorithm (called Hakou), which focused on the Hawaiian islands in a collaboration with the Army Corps of Engineers, a combination of the ADCIRC and SWAN numerical models were used to predict surge and wave responses and form the high-fidelity database. In this process, the underlying high-fidelity ADCIRC+SWAN has been validated against full-scale data from historical hurricanes with exceptional agreement, while the RRA surrogate modeling approach was blind validated against a subset of the high-fidelity database with only 2% misclassified. The resulting Hakou standalone executable facilitated a high-accuracy estimation of risk in only 6 min on a 3.2 GHz single core processor with 4 GB of RAM. This corresponds to a tremendous reduction of computational time compared to the high-fidelity model, which required over 1500 CPU hours to analyze a single hurricane scenario. In the current version of the RRA tool, coupled with PCA implementation, this computational burden has been reduced by a factor of 6. Bringing this tool into CyberEye and employing the backside servers located at the Center for Research Computing at the University of Notre Dame now enables efficient visualization of results regardless of the computational capabilities of the user’s device.

The extensibility of this framework has particular significance for the modularization of risk assessment within CyberEye to encourage users to contribute and expand the available databases in the RRA going forward. The only requirement for users to bring their databases into the RRA is that their database be described by a fixed set of parameters (up to 10). After an offline registration, the database will then be assimilated into the RRA tool as one of the available hurricane basins and ready for use for deterministic or probabilistic analysis.

**Recommended Reading**

Jia, G., and Taflanidis, A. (2013). "Kriging metamodeling for approximation of high-dimensional wave and surge responses in real-time storm/hurricane risk assessment." *Comp. Methods Appl.Mech. and Eng., *261-262: 24-38.

Kennedy, A. B., Westerink, J. J., Smith, J., Taflanidis, A. A., Hope, M., Hartman, M., Tanaka, S., Westerink, H., Cheung, K. F., Smith, T., Hamman, M., Minamide, M., and Ota, A. (2012). "Tropical cyclone inundation potential on the Hawaiian islands of Oahu and Kauai." *Ocean Model., *52-53: 54-68.

Taflanidis, A. A., Kennedy, A. B., Kennedy, A. B., and Smith, J. (2012). "Implementation/Optimization of moving least squares response surfaces for approximation of hurricane/storm surge and wave responses." *Nat.Haz., *66(2): 955-983.

Taflanidis, A. A., Kennedy, A. B., Westerink, J. J., Smith, J., Cheung, K. F., Hope, M., and Tanaka, S. (2013). "Rapid assessment of wave and surge risk during landfalling hurricanes; a probabilistic approach." *ASCE J. of Waterw., Coast. and Port Auth., *139(3): 171-182.