Neural network modelling of wave overtopping at coastal structures

Calibration of overtopping roughness factors for concrete armor units in non-breaking conditions using the CLASH database

“…However, no information is obtained regarding the uncertainty of this γf. In order to overcome the uncertainty of the estimation of the γf, a bootstrap resampling technique is used in this study as described by Van Gent et al (2007). A bootstrap resample is a random selection of N datum taken from the N original dataset, with the probability 1/N that a particular datum is selected each time.…”

“…The relative Mean Squared Error (rMSE) of log Q, similar to that used in Van Gent et al (2007), is used here to measure the quality of the overtopping estimator "e" when applied to a group of target or observed data "o" (i=1,2,….,N). The rMSE corresponding to overtopping estimator "e" and observations "o" is given by: where N is the total number of data; i is the data index; WF is the weight factor given in Table 2 from CLASH (see Van Gent et al, 2007); Qe and Qo are the estimated and target dimensionless mean overtopping discharges using estimator "e" and target data "o".…”

“…The rMSE corresponding to overtopping estimator "e" and observations "o" is given by: where N is the total number of data; i is the data index; WF is the weight factor given in Table 2 from CLASH (see Van Gent et al, 2007); Qe and Qo are the estimated and target dimensionless mean overtopping discharges using estimator "e" and target data "o". 0%<rMSEe(o)<100% indicates the proportion of variance of the data not explained by the estimator "e".…”

“…To select the data corresponding to conventional mound breakwaters, the following data filters were applied to the CLASH database: β=0, cotαd=cotαu=cotα, 1.19 ≤ cotα ≤ 4, B=0, tanαb=0, hb=0, RF≤2, CF=1, non-breaking conditions (1.8 Hm0toe<0.8h and/or Irp= Tptoe/cotα[2πHm0toe/g] 0.5 >2 to ensure no wave breaking) and Q≥10 -6 . RF is the Reliability Factor (1≤RF≤4) to account for the quality of the test, and CF is the Complexity Factor (1≤CF≤4) to account for the complexity of the cross section, both defined in Van Gent et al (2007). Furthermore, tests with remarks and/or without references were removed.…”

“…The CLASH database was created as the foundation for a generic overtopping prediction method based on artificial neural networks (see Van Gent et al, 2007). The CLASH neural network (CLNN) uses the γf given by Coeveld et al (2005), which are slightly different from those derived by Pearson et al (2004) which were also obtained within the CLASH Project.…”