Randomly distributed unit sources to enhance optimization in tsunami waveform inversion

Abstract

Abstract. In tsunami waveform inversion using the conventional Green's function technique, an optimal solution is sometimes difficult to obtain because of various factors. This study proposes a new method to both optimize the determination of the unknown parameters and introduce a global optimization method for tsunami waveform inversion. We utilize a genetic algorithm that further enhanced by a pattern search method to find an optimal distribution of unit source locations prior to the inversion. Unlike the conventional method that characterized by equidistant unit sources, our method generates a random spatial distribution of unit sources inside the inverse region. This leads to a better approximation of the initial profile of a tsunami. The method has been tested using an artificial tsunami source with real bathymetry data. Comparison results demonstrate that the proposed method has considerably outperformed the conventional one in terms of model accuracy.

Estimate of tsunami source using optimized unit sources and including dispersion effects during tsunami propagation: The 2012 Haida Gwaii earthquake

“…Without using earthquake fault parameters, initial sea surface elevation in the source region can be estimated by inversion of tsunami waveforms [Satake et al, 2005;Saito et al, 2010;Hossen et al, 2015;Mulia and Asano, 2015]. A combination of genetic algorithm (GA) methods for tsunami source inversion Asano, 2015, 2016] is used in this study to determine the initial sea surface elevation in the source region of the 2012 Haida Gwaii earthquake.…”

“…Unlike most of other tsunami inversion techniques that fix the distribution of unit sources (Figure 2a), our GA uses the least squares method iteratively to find the optimal number and distribution of unit sources. In the second stage, the GA adjusts the locations of the selected unit sources from the first stage in order to further improve the waveform fit [Mulia and Asano, 2015]. This leads to a reduction of the unit sources because the GA removes any unit source that has similar information in terms of surface height from the adjacent source points (black dots in Figure 2b) [Mulia and Asano, 2016].…”

“…We quantify the waveform fit based on a combination of root-mean-square error (RMSE) and Pearson correlation coefficient (r) [Mulia and Asano, 2015] (see Supporting Information S1). We quantify the waveform fit based on a combination of root-mean-square error (RMSE) and Pearson correlation coefficient (r) [Mulia and Asano, 2015] (see Supporting Information S1).…”

“…For every new location of unit source, synthetic tsunami waveforms at the stations are computed by applying nearest neighbor-weighted interpolation of waveforms from four nearest initial unit sources [Mulia and Asano, 2015]. For every new location of unit source, synthetic tsunami waveforms at the stations are computed by applying nearest neighbor-weighted interpolation of waveforms from four nearest initial unit sources [Mulia and Asano, 2015].…”

Fast evaluation of tsunami scenarios: uncertainty assessment for a Mediterranean Sea database

“…Liu and Wang, 2008;Saito and Furumura, 2009;Saito et al, 2011;Tsushima et al, 2014;Mulia and Asano, 2015), pyramidal (Zaibo et al, 2003), conical or circular with positive elevation in the centre and negative at the edge (Choi et al, 2005;Zaitsev and Pelinovsky , 2011), rectangular prisms (Miranda et al, 2014) or cosine-tapered (Hossen et al, 2015) ES. The approach has been proposed in retrospective for rapid near-field forecasting of the Tohoku 2011 tsunami with tFISH/RAPiD (Tsushima et al, 2014), for PTHA (Selva et al, 2016) and for source inversion (Liu and Wang, 2008;Saito et al, 2011;Mulia and Asano, 2015;Hossen et al, 2015). However, such an approach has never been fully validated by a systematic assessment of the uncertainties that it introduces in the tsunami modelling or forecasting.…”