New perspectives on tsunami source inversion

Abstract

Many tsunami source inversion techniques have already been developed to infer tsunami source models with the assumption that tsunami generation is due to slip on a single large fault. Therefore, these inversion techniques cannot determine to what extent subsidiary phenomena - such as submarine landslides, block movement, or slip on splay faults - have contributed to the tsunami generation. In addition to taking into account subsidiary phenomena, it is necessary to consider all the model complexities associated with tsunami generation and propagation including model physics, source kinematics and source discretization to infer a source model that can produce tsunami waveforms having a good agreement with observed waveforms. In my thesis, I have developed new methods using tsunami waveforms to estimate the properties of the source that depend as little as possible on assumptions about how it was generated. I first consider the importance of model parametrization, including dispersion, source kinematics and source discretization, in a conventional least-squares approach to tsunami source inversion. I implement single and multiple time window methods for dispersive and non-dispersive wave propagation to estimate source models for the tsunami generated by the 2011 Tohoku-oki earthquake. The results show that tsunami source models can strongly depend on such model choices, in particular for high quality data available today from ocean bottom pressure and global positioning system gauges. My results show that it is important to consider them together, rather than separately as has been done previously, in order to obtain more meaningful inversion results. I have also proposed a new method that can be used to derive source models without requiring the assumption of slip on a fault of pre-determined geometry or even knowing the earthquake source area. The proposed method is based on "Time Reverse Imaging (TRI)" technique, which has been used in underwater acoustic and medical imaging. We have applied TRI to recover the initial sea surface displacement associated with the tsunami source. To show the application of this method we have chosen the same tsunami event triggered by the March 11, 2011 Tohoku earthquake, for which an unprecedented number of high-quality observations are available. This method has been extended by combining TRI with conventional source inversion method. In conventional source inversion, the linear problem is solved by weighted least-squares method with regularization constraints to obtain stable and physically meaningful solutions. This approach solves the same linear problem but without using additional constraints or imposing weights of the stations. Instead, the parameters are determined by using time reverse imagining with the same Green's function from source to receiver and reversed observed waveforms. This method can overcome the uncertainties associated with observation weights and the choice of hyperparameters. The method has potential for use tsunami warning systems, as the method is computationally efficient and can be used to estimate initial source model by using precomputed Greens function in order to provide more accurate and realistic tsunami prediction.