|
| | Abstract | | Natural earthquake tomography is one of the typical geophysical method and has been used widely as a technique for imaging subsurface structure with good accuracy. However, there have been some problems in conventional travel time tomography. One is the resolution of the analysis result. The subsurface structures must be discretized into many grids or small elements, and the grids spacing or element size must be manually optimized on the basis of ray path density. Moreover, even when a high-density ray path area is in a target zone, the grid spacing must be set according to the lowest ray path density area. The other is the initial model. The analysis result depends on the initial model. To address these problems, we develop a new natural earthquake tomography method by use of neural network. The new method is realized by making use of the excellent capability of multilayer neural network to approximate an arbitrary function for which the network training is carried out by minimize the squared residuals of an integral equation. We applied a new method to three-dimensional numerical experiments. As a result we revealed that the new method by neural network is efficient. |
|
|
|
| |
|