Abstract
This paper establishes various advancements for the application of surrogate modeling techniques for storm surge prediction utilizing an existing database of high-fidelity, synthetic storms (tropical cyclones). Kriging, also known as Gaussian process regression, is specifically chosen as the surrogate model in this study. Emphasis is first placed on the storm selection for developing the database of synthetic storms. An adaptive, sequential selection is examined here that iteratively identifies the storm (or multiple storms) that is expected to provide the greatest enhancement of the prediction accuracy when that storm is added into the already available database. Appropriate error statistics are discussed for assessing convergence of this iterative selection, and its performance is compared to the joint probability method with optimal sampling, utilizing the required number of synthetic storms to achieve the same level of accuracy as comparison metric. The impact on risk estimation is also examined. The discussion then moves to adjustments of the surrogate modeling framework to support two implementation issues that might become more relevant due to climate change considerations: future storm intensification and sea level rise (SLR). For storm intensification, the use of the surrogate model for prediction extrapolation is examined. Tuning of the surrogate model characteristics using cross-validation techniques and modification of the tuning to prioritize storms with specific characteristics are proposed, whereas an augmentation of the database with new/additional storms is also considered. With respect to SLR, the recently developed database for the US Army Corps of Engineers’ North Atlantic Comprehensive Coastal Study is exploited to demonstrate how surrogate modeling can support predictions that include SLR considerations.
Similar content being viewed by others
References
Bachoc F (2013) Cross validation and maximum likelihood estimations of hyper-parameters of Gaussian processes with model misspecification. Comput Stat Data Anal 66:55–69
Bass B, Bedient P (2018) Surrogate modeling of joint flood risk across coastal watersheds. J Hydrol 558:159–173
Bengio Y, Grandvalet Y (2004) No unbiased estimator of the variance of k-fold cross-validation. J Mach Learn Res 5:1089–1105
Das HS, Jung H, Ebersole B, Wamsley T, Whalin RW (2010) An efficient storm surge forecasting tool for coastal Mississippi. Paper presented at the 32nd international coastal engineering conference, Shanghai, China
Fischbach JR, Johnson DR, Kuhn K (2016) Bias and efficiency tradeoffs in the selection of storm suites used to estimate flood risk. J Mar Sci Eng 4(1):10
Ginsbourger D, Dupuy D, Badea A, Carraro L, Roustant O (2009) A note on the choice and the estimation of Kriging models for the analysis of deterministic computer experiments. Appl Stoch Models Bus Ind 25(2):115–131
Hartigan JA, Wong MA (1979) Algorithm AS 136: a K-means clustering algorithm. J Roy Stat Soc Ser C (Appl Stat) 28(1):100–108
Irish J, Resio D, Cialone M (2009) A surge response function approach to coastal hazard assessment. Part 2: quantification of spatial attributes of response functions. Nat Hazards 51(1):183–205
Jia G, Taflanidis AA (2013) Kriging metamodeling for approximation of high-dimensional wave and surge responses in real-time storm/hurricane risk assessment. Comput Methods Appl Mech Eng 261–262:24–38
Jia G, Taflanidis AA, Nadal-Caraballo NC, Melby J, Kennedy A, Smith J (2015) Surrogate modeling for peak and time dependent storm surge prediction over an extended coastal region using an existing database of synthetic storms. Nat Hazards 81(2):909–938
Kennedy AB, Westerink JJ, Smith J, Taflanidis AA, Hope M, Hartman M, Tanaka S, Westerink H, Cheung KF, Smith T, Hamman M, Minamide M, Ota A (2012) Tropical cyclone inundation potential on the Hawaiian islands of Oahu and Kauai. Ocean Model 52–53:54–68
Kijewski-Correa T, Smith N, Taflanidis A, Kennedy A, Liu C, Krusche M, Vardeman C (2014) CyberEye: development of integrated cyber-infrastructure to support rapid hurricane risk assessment. J Wind Eng Ind Aerodyn 133:211–224
Kim S-W, Melby JA, Nadal-Caraballo NC, Ratcliff J (2015) A time-dependent surrogate model for storm surge prediction based on an artificial neural network using high-fidelity synthetic hurricane modeling. Nat Hazards 76(1):565–585
Kleijnen JP, Beers WV (2004) Application-driven sequential designs for simulation experiments: Kriging metamodelling. J Oper Res Soc 55(8):876–883
Kohavi R (1995) A study of cross-validation and bootstrap for accuracy estimation and model selection. In: International joint conference on artificial intelligence, pp 1137–1145
Lin N, Emanuel K, Oppenheimer M, Vanmarcke E (2012) Physically based assessment of hurricane surge threat under climate change. Nat Clim Change 2(6):462–467
Liu H, Ong Y-S, Cai J (2017) A survey of adaptive sampling for global metamodeling in support of simulation-based complex engineering design. Struct Multidiscipl Optim 57(1):393–416
Lophaven SN, Nielsen HB, Sondergaard J (2002) DACE-A MATLAB Kriging toolbox. Technical University of Denmark
Luettich RA, Jr., Westerink JJ, Scheffner NW (1992) ADCIRC: an advanced three-dimensional circulation model for shelves, coasts, and estuaries. Report 1. Theory and methodology of ADCIRC-2DDI and ADCIRC-3DL. Dredging Research Program Technical Report DRP-92-6, U.S Army Engineers Waterways Experiment Station, Vicksburg, MS
Meckesheimer M, Booker AJ, Barton RR, Simpson TW (2002) Computationally inexpensive metamodel assessment strategies. AIAA J 40(10):2053–2060
Nadal-Caraballo NC, Melby JA, Gonzalez VM, Cox AT (2015) North Atlantic coast comprehensive study—coastal storm hazards from Virginia to Maine, ERDC/CHL TR-15-5. U.S. Army Engineer Research and Development Center, Vicksburg
Niedoroda AW, Resio DT, Toro GR, Divoky D, Reed C (2010) Analysis of the coastal Mississippi storm surge hazard. Ocean Eng 37(1):82–90
Pronzato L, Müller WG (2012) Design of computer experiments: space filling and beyond. Stat Comput 22(3):681–701
Rao RB, Fung G, Rosales R (2008) On the dangers of cross-validation. An experimental evaluation. In: Proceedings of the 2008 SIAM international conference on data mining. SIAM, pp 588–596
Resio DT, Boc SJ, Borgman L, Cardone V, Cox A, Dally WR, Dean RG, Divoky D, Hirsh E, Irish JL, Levinson D, Niedoroda A, Powell MD, Ratcliff JJ, Stutts V, Suhada J, Toro GR, Vickery PJ (2007) White paper on estimating hurricane inundation probabilities. Consulting Report prepared by USACE for FEMA
Resio D, Irish J, Cialone M (2009) A surge response function approach to coastal hazard assessment—part 1: basic concepts. Nat Hazards 51(1):163–182
Resio DT, Irish JL, Westering JJ, Powell NJ (2012) The effect of uncertainty on estimates of hurricane surge hazards. Nat Hazards 66(3):1443–1459
Resio DT, Asher TG, Irish JL (2017) The effects of natural structure on estimated tropical cyclone surge extremes. Nat Hazards 88(3):1609–1637
Rohmer J, Lecacheux S, Pedreros R, Quetelard H, Bonnardot F, Idier D (2016) Dynamic parameter sensitivity in numerical modelling of cyclone-induced waves: a multi-look approach using advanced meta-modelling techniques. Nat Hazards 84(3):1765–1792
Sacks J, Welch WJ, Mitchell TJ, Wynn HP (1989) Design and analysis of computer experiments. Stat Sci 4(4):409–435
Santner TJ, Williams BJ, Notz WI (2013) The design and analysis of computer experiments. Springer, Berlin
Smith JM, Sherlock AR, Resio DT (2001) STWAVE: Steady-state spectral wave model user’s manual for STWAVE, Version 3.0. DTIC Document
Smith JM, Westerink JJ, Kennedy AB, Taflanidis AA, Smith TD (2011) SWIMS Hawaii hurricane wave, surge, and runup inundation fast forecasting tool. In: 2011 Solutions to coastal disasters conference, Anchorage, Alaska, 26–29 June
Sundararajan S, Keerthi SS (2001) Predictive approaches for choosing hyperparameters in Gaussian processes. Neural Comput 13(5):1103–1118
Taflanidis AA, Kennedy AB, Westerink JJ, Smith J, Cheung KF, Hope M, Tanaka S (2013) Rapid assessment of wave and surge risk during landfalling hurricanes; probabilistic approach. ASCE J Waterw Port Coast Ocean Eng 139(3):171–182
Tanaka S, Bunya S, Westerink J, Dawson C, Luettich R (2011) Scalability of an unstructured grid continuous Galerkin based hurricane storm surge model. J Sci Comput 46:329–358. https://doi.org/10.1007/s10915-010-9402-1
Toro GR, Resio DT, Divoky D, Niedoroda A, Reed C (2010) Efficient joint-probability methods for hurricane surge frequency analysis. Ocean Eng 37:125–134
USACE (2015) North Atlantic coast comprehensive study: resilient adaption to increasing risk. US Army Corps of Engineers, Washington
Wynn H (2004) Maximum entropy sampling and general equivalence theory. In: Di Bucchianico A, Läuter H, Wynn HP (eds) mODa 7—advances in model-oriented design and analysis. Contributions to statistics. Physica, Heidelberg, pp 211–218
Zijlema M (2010) Computation of wind-wave spectra in coastal waters with SWAN on unstructured grids. Coast Eng 57(3):267–277
Acknowledgements
This work has been done under contract with the US Army Corps of Engineers (USACE), Engineer Research and Development Center, Coastal and Hydraulics Laboratory (ERDC-CHL). The support of the USACE’s Flood and Coastal R&D Program is also gratefully acknowledged.
Funding
The funding was provided by Engineer Research and Development (Grant No. W912HZ-16-P-0083-P00001).
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: Surrogate model validation
Validation of the surrogate model is established by comparing its predictions to the actual response (high-fidelity simulations of storm surge) for storm scenarios that do not belong to database X. These scenarios represent the validation set. Two common approaches are used for forming this set (Kohavi 1995): test-sample or cross-validation. The first one requires new observations, or, equivalently, splitting of the initial database to a single training set and a single validation set. The second approach repeats comparisons of the accuracy over different divisions of the entire database to training set and validation set and has the benefit that it requires no new observations. Perhaps the most common implementation of the latter is the leave-one-out cross-validation (LOOCV): Each of the observations from the database is sequentially removed, then the remaining training points are used to predict the output for it, and the error between the predicted and real responses is evaluated. A more general CV approach is k-fold CV, established by partitioning the entire observation set into k equal-size subsets and calculating errors similarly as in the LOOCV, but removing each time the entire subset, rather than a single observation.
As error statistics popular choices are the coefficient of determination, the correlation coefficient and the square root of the mean squared error. For output yj and the LOOCV approach these are given by
where \(\hat{y}_{j} ({\mathbf{x}}^{h} |{\mathbf{X}}_{ - h} )\) represents the surrogate model predictions for storm xh established by using the entire database excluding that storm. Using the closed form of Eq. (7) the statistics in Eqs. (18)–(20) can be calculated without explicitly evaluating \(\hat{y}_{j} ({\mathbf{x}}^{h} |{\mathbf{X}}_{ - h} )\), rather simply substituting LOO predictions for entire output vector by
where [e]h is the vector corresponding to the hth row of matrix with elements ehi, i.e., to the LOO errors associated with the latent outputs for storm xh. For the test-sample approach the set \(\{ {\mathbf{x}}^{h} ;h = 1, \ldots ,n\}\) needs to be replaced with the training set of new observations \(\{ {\mathbf{x}}_{n}^{h} ;h = 1, \ldots ,n_{n} \}\) in these equations, whereas the entire database X (rather than X–h) is used for the surrogate model predictions. It should be pointed out that the aforementioned statistics evaluate the average accuracy over the examined domain X. An alternative approach would have been to look at the maximum error over X which examines the worst-case prediction performance.
The accuracy across all outputs is evaluated by calculating the average error statistics over all outputs.
Values of \(\bar{R}^{2}\) and \( \overline{{cc}}\) close to 1 and values of \( \overline{{rmse}}\) close to zero indicate better accuracy. It should be also pointed out that \( \overline{{rmse}}\) does not correspond to normalized statistics; in other words, its value depends on the magnitude of the response.
Appendix 2: Calculation of gradients for LOOCV objective function
This appendix discusses the estimation of the gradient of the objective function of Eq. (6) utilized in the LOOCV tuning of the hyper-parameters. The kth component of the gradient vector, the partial derivative with respect to hyper-parameter sk, is
with partial derivative for LOO error given by
where
in which \(\frac{{\partial {\mathbf{R}}({\mathbf{X}})}}{{\partial s_{k} }}\) and \(\frac{{\partial {\varvec{\upbeta}}^{*} }}{{\partial s_{k} }}\) are matrices of element-wise derivatives with the latter given by
with \({\mathbf{Q}}({\mathbf{X}})^{\text{T}} = {\mathbf{F}}({\mathbf{X}})^{\text{T}} {\mathbf{R}}({\mathbf{X}})^{ - 1}\).
Appendix 3: Optimization for storm selection
This appendix discusses a numerical optimization scheme for the storm selection. First, select the number of candidate samples nc for the stochastic search, the percentage reductions ar, ac for determining the retained experiments, and the number of samples for the Monte Carlo integration (MCI) Ns. Perform then the following steps, also shown in Fig. 11 for the same example considered in Fig. 1.
-
(1)
[MCI samples]. Generate Ns {xq; q = 1,…, Ns} samples representing distribution φ(x), and determine the corresponding weights wq(xq). For the common implementation that samples are directly from distribution φ(x) the weights are wq(xq) = 1.
-
(2)
[Candidate experiment generation]. Generate nc samples for x with uniform distribution in domain D, forming set \(\{ {\mathbf{x}}_{new}^{c} ;c = 1, \ldots ,n_{c} \}\). Skip this step if candidate experiments are already provided (Part a of Fig. 11).
-
(3)
[Ranking of experiments]. Evaluate \(\{ \sigma_{n}^{2} ({\mathbf{x}}_{new}^{c} |{\mathbf{X}});c = 1, \ldots ,n_{c} \}\) and order candidate samples using a descending order for \(\sigma_{n}^{2} ({\mathbf{x}}_{new}^{c} |{\mathbf{X}})\). Retain only the arnc candidate experiments that correspond to the highest values of \(\{ \sigma_{n}^{2} ({\mathbf{x}}_{new}^{c} |{\mathbf{X}});c = 1, \ldots ,n_{c} \}\) (Part b of Fig. 11). Skip this step if ar = 1.
-
(4)
[Clustering of solutions]. Cluster retained experiments from Step 3 into acarnc clusters, using for example K-means cluster (Hartigan and Wong 1979), and keep only one experiment per cluster, the one corresponding to the largest value of \(\sigma_{n}^{2} ({\mathbf{x}}_{new}^{c} |{\mathbf{X}})\) (Part c of Fig. 11). Skip this step if ac = 1.
-
(5)
[Calculation of IMSE]. For each retained candidate experiment after Step 4, \(\{ {\mathbf{x}}_{new}^{c} ;c = 1, \ldots ,a_{r} a_{c} n_{c} \}\), calculate \(\{ \sigma_{n}^{2} ({\mathbf{x}}^{q} |{\mathbf{X}},{\mathbf{x}}_{new}^{c} );q = 1, \ldots ,N_{q} \}\) using Eqs. (12) and (13). Approximate IMSE through MCI as
$$IMSE({\mathbf{x}}_{new}^{c} ) = \frac{1}{{N_{s} }}\sum\limits_{q = 1}^{{N_{s} }} {w_{q} ({\mathbf{x}}^{q} )\sigma_{n}^{2} ({\mathbf{x}}^{q} |{\mathbf{X}},{\mathbf{x}}_{new}^{c} )} .$$(27) -
(6)
[Final selection]. Select as new experiment the one that provides the minimum value for \(\{ IMSE({\mathbf{x}}_{new}^{c} );c = 1, \ldots ,a_{r} a_{c} n_{c} \}\) (Part d of Fig. 11).
The most computationally demanding step of this process is Step 5, which requires MCI for each of the candidate experiments examined. Steps 3 and 4 have been introduced to reduce this burden. Step 3 removes candidate experiments belonging to domains where the surrogate model accuracy is already high. For such experiments, it is anticipated that their IMSE will not correspond to a minimum over D, as addition of experiments in domains of already adequate accuracy [corresponding to lower values of \(\sigma_{n}^{2} ({\mathbf{x}}|{\mathbf{X}})\)] is anticipated to provide small benefits. Step 4 keeps only one candidate experiment among experiments that are close to one another. All such experiments are expected to provide similar improvement of IMSE, and therefore, examining all of them as candidate solutions is redundant. Only the experiment with lowest accuracy is retained in each cluster. Removal of experiments in Steps 3 and 4 provides an acar-fold reduction in computational burden without compromising the quality of the identified solution of the optimization, provided that values of ar and ac are not too small. These values should be chosen in range [0.7–0.3].
Rights and permissions
About this article
Cite this article
Zhang, J., Taflanidis, A.A., Nadal-Caraballo, N.C. et al. Advances in surrogate modeling for storm surge prediction: storm selection and addressing characteristics related to climate change. Nat Hazards 94, 1225–1253 (2018). https://doi.org/10.1007/s11069-018-3470-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11069-018-3470-1