Seismic structure inversion of refraction and reflection waves by progressive model development method

提出了一种二维折射实验反演方法——渐进式模型展开法（PMDM）。1)，允许通过从一维（1- d）模型到伪二维模型，然后是二维模型的顺序逐步改进速度结构（Sato和Kennett， 2000）。本研究对PMDM进行了改进。1 .能够使用后期阶段，如反射。首先，我们挑选出一组相位似乎相同的数据。然后，我们使用仅用首次到达数据计算的速度结构模型（PMDM ver.1的结果）进行最小二乘法估计每组最可能的相位。计算机自动识别所选数据的各个阶段。只有数据挑选和分组需要目视检查。由于对组进行相位识别，因此识别的可靠性高于对每个选取的数据进行相位识别。如果仅用首次到达数据计算的速度结构模型与真实结构的差异较大，则上述方法存在一定的相错识别概率。为了避免这种情况，我们估计了错误识别概率高的群体的最佳2阶段。然后计算每个相位的二维反演。这种修改增加了计算量。当有N个组分两个阶段时，阶段的组合是2的N次方。然而，这种修改使该方法更加健壮。我们使用上述方法（PMDM ver.2）进行了数值模拟。结果表明，该方法能较好地提高下层边界的分辨率和各层下层的速度。即使我们使用一个不好的初始模型，使用正确相位的反演结果也会成为最好的估计模型。我们还对一个真实的数据进行了该方法。

We developed an inversion method for 2-D refraction experiments, Progressive model development method (PMDM ver.1), that allows progressive improvement of velocity structure through a sequence from one-dimensional (1-D) models to pseudo 2-D models, and then 2-D models (Sato and Kennett, 2000). This study modified PMDM ver.1 to be able to use later phases such as reflection. First, we make groups of picked data whose phase seems to be same. Then we conduct the least square method to estimate the most probable phase for each group using the velocity structure model calculated only with the first arrival data (the result of PMDM ver.1). This identifies phases of picked data by computer automatically. Only data picking and grouping need visual inspection. Because of phase identification for groups, the reliability of the identification is higher than phase identification for each picked datum.If the difference between the velocity structure model calculated only with the first arrival data and the true structure is not small, the above method has some probability to make mis-identification of phases. To avoid this, we estimate the best 2 phases for groups that have high probability of mis-identification. Then we calculate 2-D inversion for each phase. This modification increases the amount of computation. When there are N groups that take 2 phases, combination of phases is 2 to the N-th. However, this modification makes the method more robust.We conducted a numerical simulation using the above method (PMDM ver.2). The results show that the resolution of lower layer boundaries and velocities of lower parts of each layer are much improved. Even if we use a bad initial model, the result of the inversion using correct phases becomes the best estimated model. We also conducted the method for a real data.