Abstract:

Time-averaged shear-wave velocity to 30m(V_S30)is one of the most widely used parameter for characterizing local site conditions, classifying engineering sites, evaluating site amplification, and developing ground motion prediction models. Direct determination of V_S30 generally requires borehole investigation and in situ shear-wave velocity testing to a depth of at least 30m. Such measurements are reliable but costly and spatially sparse, making them difficult to obtain over large regions. Consequently, regional V_S30 maps are commonly developed from proxy variables such as geology, geomorphology, and topographic slope. Existing slope-based models are simple and broadly applicable, but a single slope parameter cannot fully represent regional differences in terrain and near-surface materials. Terrain-classification-based models incorporate multiple terrain descriptors, yet they usually assign one representative V_S30 value to all locations within the same terrain class, thereby neglecting within-class variability and producing spatially discontinuous estimates at class boundaries.

To address these limitations, this study proposes an improved V_S30 estimation method that combines terrain features with slope information. Xinjiang, China, was selected as the study area. Three terrain attributes—slope, surface texture, and local convexity—were derived from a 30-arc-second digital elevation model. Following the Iwahashi-Pike terrain-classification scheme, the study area was divided into 16 terrain classes through a nested decision-tree procedure. A total of 4 115 engineering boreholes with measured shear-wave velocity profiles were compiled to develop the model.

Model development involved three steps. First, for each terrain class, 65% of the boreholes were randomly selected as the training subset and the remaining 35% were used for validation. This procedure was repeated 1 000 times, and the class-specific representative V_S30 value associated with the minimum mean squared prediction error was retained. Second, because several terrain classes contained relatively few boreholes, F-tests at a significance level of 0.05 were used to evaluate whether sparsely sampled classes with similar terrain characteristics could be combined. The results supported merging classes 3 and 4, classes 13 and 14, and classes 15 and 16, whereas classes 1 and 2 were retained separately. Third, within each final terrain group, a logarithmic regression between V_S30 and slope was examined. When the slope coefficient was statistically significant, a class-specific regression equation was used to produce continuously varying V_S30 estimates; otherwise, the cross-validated representative value was assigned to that terrain group. Statistically significant V_S30-slope relationships were identified for seven terrain groups.

The resulting terrain-slope hybrid model was used to generate a regional V_S30 map for Xinjiang and was evaluated against two widely used proxy approaches: the WA07 slope-only model and the Yong12 terrain-only model. Compared with the two reference models, the proposed method better represented within-class spatial variability and reduced abrupt changes caused by assigning a single constant value to an entire terrain class. For logarithmic V_S30 residuals in the validation dataset, the hybrid model produced a mean residual of 0.018 8, a standard deviation of 0.264 8, a mean absolute error of 0.174 9, and a root-mean-square error of 0.265 2. The corresponding standard deviation, mean absolute error, and root-mean-square error were 0.278 9, 0.183 0, and 0.280 7 for Yong12 and 0.328 4, 0.260 9, and 0.335 7 for WA07, respectively. An additional assessment based on locally smoothed means and standard deviations yielded a typical standard deviation of 77.42m/s for the hybrid model, lower than 79.06m/s for Yong12 and 83.40m/s for WA07. Site-classification performance was further examined using the NEHRP B, C, D, and E categories. The terrain-slope hybrid model achieved an overall classification accuracy of 80.85%, compared with 75.56%for Yong12 and 69.79%for WA07.hese results demonstrate that incorporating slope-dependent variation within terrain classes can improve both the accuracy and spatial continuity of regional V_S30 estimation. The proposed method provides a practical approach for regional seismic hazard and risk studies, rapid site-conditioning mapping, and engineering applications in areas where borehole measurements are limited. Nevertheless, its applicability should be further examined in extensive plains, deep basins, and densely urbanized areas, where the relationship between topography and shallow subsurface conditions may be weak or where buildings and vegetation may bias DEM-derived slope estimates.

Key words: terrain features, terrain classification, regression analysis, V_S30 estimation, site effects

摘要：

地下30m范围内的等效剪切波速(V_S30)作为评估场地效应和地震动预测模型的关键指标, 其准确性直接影响地震动放大效应的评估及地震动预测的可靠性。文中提出了一种融合地形特征和坡度信息的V_S30 宏观估计改进方法, 通过引入坡度信息, 改进了基于单一地形特征信息的V_S30 估计方法无法考虑地形类别内V_S30 变异性的问题。以新疆地区为例, 基于该地区4 115个钻孔数据, 应用该方法进行V_S30 宏观估计, 并给出了新疆地区V_S30 的空间分布地图。通过V_S30 预测值的残差分析和场地类别预测结果的混淆矩阵, 将文中方法与基于单一坡度信息和单一地形特征信息的V_S30 预测方法进行对比。 结果表明, 文中提出的地形-坡度混合模型在预测值的残差指标和场地类别预测结果准确度方面均优于传统预测模型。

关键词: 地形特征, 地形分类, 回归分析, V_S30估计, 场地效应