近年来，受计算机视觉与自然语言处理领域成功启发，数据驱动的深度学习方法展现出克服这些挑战的巨大潜力。然而，缺乏大规模多样化的基准数据集仍是此类方法开发与评估的主要障碍。为填补这一空白，我们推出OpenSWI综合基准数据集，该数据集通过表面波反演数据集预处理（SWIDP）管道生成。OpenSWI包含两套针对不同研究尺度与应用场景的合成数据集——OpenSWI-shallow与OpenSWI-deep，以及用于泛化评估的人工智能就绪型真实数据集OpenSWI-real。OpenSWI-real整合自开源项目，包含两组观测色散曲线及其对应的一维参考模型，作为评估深度学习模型泛化能力的基准数据集。", "ds_source": "
OpenSWI-shallow源自二维地质模型数据集OpenFWI，包含逾2200万组一维速度剖面及其基本模式相位与群速度色散曲线，覆盖广泛的浅层地质结构（如平坦层、断层、褶皱及真实地层）。OpenSWI-deep由14个全球及区域性三维地质模型构建而成，包含约126万组高精度一维速度-色散数据对，专用于深部地球研究。", "ds_process_way": "
通过表面波反演数据集预处理（SWIDP）管道生成。", "ds_quality": "
为验证OpenSWI实用性，我们基于OpenSWI-shallow和OpenSWI-deep训练深度学习模型，并通过OpenSWI-real进行评估，结果显示预测模型与基准速度模型高度吻合，证实了OpenSWI数据集的多样性与代表性。", "ds_acq_start_time": "2013-01-01 00:00:00", "ds_acq_end_time": "2016-12-31 00:00:00", "ds_acq_place": "全球", "ds_acq_lon_east": -180.0, "ds_acq_lat_south": 90.0, "ds_acq_lon_west": 180.0, "ds_acq_lat_north": 90.0, "ds_acq_alt_low": null, "ds_acq_alt_high": null, "ds_share_type": "login-access", "ds_total_size": 41370974604, "ds_files_count": 2, "ds_format": ".npz", "ds_space_res": "", "ds_time_res": "年", "ds_coordinate": "无", "ds_projection": "", "ds_thumbnail": "70160163-73d4-4508-9416-b462b1b2ca10.png", "ds_thumb_from": 2, "ds_ref_way": "", "paper_ref_way": "", "ds_ref_instruction": "", "ds_from_station": null, "organization_id": "0a4269e1-65f4-45f1-aeba-88ea3068eebf", "ds_serv_man": "敏玉芳", "ds_serv_phone": "0931-4967596", "ds_serv_mail": "ncdc@lzb.ac.cn", "doi_value": "", "subject_codes": [], "quality_level": 3, "publish_time": "2025-11-27 15:50:52", "last_updated": "2026-01-14 11:03:08", "protected": false, "protected_to": null, "lang": "zh", "cstr": "11738.11.NCDC.ZENODO.DB7022.2025", "i18n": { "en": { "title": "OpenSWI: A Massive-Scale Benchmark Dataset for Surface Wave Dispersion Curve Inversion(（2013,2016）", "ds_format": ".npz", "ds_source": "
OpenSWI shalow is derived from the 2D geological model dataset OpenFWI, which contains over 22 million sets of one-dimensional velocity profiles and their basic mode phase and group velocity dispersion curves, covering a wide range of shallow geological structures such as flat layers, faults, folds, and real strata. OpenSWI deep is constructed from 14 global and regional 3D geological models, containing approximately 1.26 million high-precision one-dimensional velocity dispersion data pairs, specifically designed for deep Earth research.", "ds_quality": "
To verify the practicality of OpenSWI, we trained deep learning models based on OpenSWI sharow and OpenSWI deep, and evaluated them through OpenSWI real. The results showed that the predicted model highly matched the baseline velocity model, confirming the diversity and representativeness of the OpenSWI dataset.", "ds_ref_way": "", "ds_abstract": "
In recent years, inspired by the success of computer vision and natural language processing, data-driven deep learning methods have shown great potential to overcome these challenges. However, the lack of large-scale and diverse benchmark datasets remains a major obstacle to the development and evaluation of such methods. To fill this gap, we have launched the OpenSWI comprehensive benchmark dataset, which is generated through the Surface Wave Inversion Dataset Preprocessing (SWIDP) pipeline. OpenSWI includes two synthetic datasets for different research scales and application scenarios - OpenSWI sharow and OpenSWI deep, as well as the AI ready real dataset OpenSWI real for generalization evaluation. OpenSWI real integrates from an open source project, including two sets of observed dispersion curves and their corresponding one-dimensional reference models, as a benchmark dataset for evaluating the generalization ability of deep learning models.
", "ds_time_res": "年", "ds_acq_place": "global", "ds_space_res": "", "ds_projection": "", "ds_process_way": "Generate pipeline through surface wave inversion dataset preprocessing (SWIDP).", "ds_ref_instruction": "" } }, "submit_center_id": "ncdc", "data_level": 0, "license_type": "CC BY 4.0", "ds_topic_tags": [ "地球物理反演", "基准数据集", "OpenSWI" ], "ds_subject_tags": [], "ds_class_tags": [], "ds_locus_tags": [ "全球" ], "ds_time_tags": [ 2013, 2016 ], "ds_contributors": [ { "true_name": "李亚星", "email": "yxli2024@cdut.edu.cn", "work_for": "成都理工大学", "country": "中国" } ], "ds_meta_authors": [ { "true_name": "李亚星", "email": "yxli2024@cdut.edu.cn", "work_for": "成都理工大学", "country": "中国" } ], "ds_managers": [ { "true_name": "李亚星", "email": "yxli2024@cdut.edu.cn", "work_for": "成都理工大学", "country": "中国" } ], "category": "其他" }