Tsunami run-up height forecasting by using artificial neural networks

Abstract

Tsunami run-up heights (R) were predicted by using two different artificial neural network (ANN) methods such as feed forward back propagation (FFBP) and generalised regression neural networks (GRNN). The R records resulting from the ground motions, which occurred between 1900 and 2007, were used during the applications. These records were gathered from three coastal states of the USA, namely California (CA), Oregon (OR) and Washington (WA). First, the earthquake moment magnitude (M w ), the distance from the earthquake source to the run-up location (D), the latitude of the run-up location (L x ) and the longitude of the run-up location (L y ) were used as inputs of each ANN method. In order to evaluate the effects of the L x and L y on the R prediction, a second input combination consisting of the M w and D was used. Each ANN method structured for each input combination was applied to estimate the R of both the separate state and the Western Seaboard. In general, the forecasting performance of the FFBP model that used the M w , D, L x and L y in the input layer was found superior to the other models under the conditions of the used data and model structures.