Quality influencing factors of dispersion curves from short period dense arrays based on a convolutional neural network across the north section of the Xiaojiang fault area

The number of dispersion curves increases significantly when the scale of a short-period dense array increases. Owing to a substantial increase in data volume, it is important to quickly evaluate dispersion curve quality as well as select the available dispersion curve. Accordingly, this study quantitatively evaluated dispersion curve quality by training a convolutional neural network model for ambient noise tomography using a short-period dense array. The model can select high-quality dispersion curves that exhibit a ≤ 10% difference between the results of manual screening and the proposed model. In addition, this study established a dispersion curve loss function by analyzing the quality of the dispersion curve and the corresponding influencing factors, thereby estimating the number of available dispersion curves for the existing observation systems. Furthermore, a Monte Carlo simulation experiment is used to illustrates the station-pair interval distance probability density function, which is independent of station number in the observational system with randomly deployed stations. The results suggested that the straight-line length should exceed 15 km to ensure that loss rate of dispersion curves remains < 0.5, while maintaining the threshold ambient noise tomography accuracy within the study area.