OpenSWI: A Massive-Scale Benchmark Dataset for
Surface Wave Dispersion Curve Inversion

🌐 Main Page | 📃 Paper | 🤗 Dataset

---

OpenSWI: A Massive-Scale Benchmark Dataset for Surface Wave Dispersion Curve Inversion

Feng Liu, Sijie Zhao, Xinyu Gu, Fenghua Ling, Peiqin Zhuang, Yaxing Li*, Rui Su*, Lihua Fang, Lianqing Zhou, Jianping Huang, Lei Bai

Abstract: Surface wave dispersion curve inversion plays a critical role in both shallow resource exploration and deep geological studies, yet it remains hindered by sensitivity to initial models, susceptibility to local minima, and low computational efficiency. Recently, data-driven deep learning methods, inspired by their success in computer vision and natural language processing, have shown promising 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 address this gap, we introduce OpenSWI, a comprehensive benchmark dataset generated through the surface wave inversion dataset pipeline (SWIDP). OpenSWI comprises two synthetic datasets tailored to different research scales and application scenarios, namely OpenSWI-shallow and OpenSWI-deep, as well as an AI-ready real-world dataset for generalization evaluation, OpenSWI-real. OpenSWI-shallow is derived from the 2-D geological model dataset OpenFWI, containing over 22 million 1-D velocity profiles paired with their fundamental-mode phase and group velocity dispersion curves, spanning a broad spectrum of shallow geological structures (e.g., flat layers, faults, folds, and realistic stratigraphy). OpenSWI-deep is built from 14 global and regional 3-D geological models, comprising approximately 1.26 million high-fidelity 1-D velocity-dispersion data pairs for deep earth studies. OpenSWI-real, compiled from open-source projects, contains two sets of observed dispersion curves and their corresponding 1-D reference models, serving as a benchmark for evaluating the generalization of deep learning models. To demonstrate the utility of OpenSWI, we trained deep learning models on OpenSWI-shallow and OpenSWI-deep, and evaluated them on OpenSWI-real. The results show strong agreement between the predicted and reference velocity models, confirming the diversity and representativeness of the OpenSWI dataset. To facilitate the advancement of intelligent surface wave dispersion curve inversion techniques, we release the SWIDP toolbox, the OpenSWI datasets, trained deep learning models, and other examples, aiming to provide comprehensive support and open resources for the research community.

---

📊 Datasets

More Details of the OpenSWI Datasets can be found at huggingface (OpenSWI Huggingface)

• OpenSWI-Shallow: 1D velocity profiles derived from 2D velocity models (OpenFWI dataset), paired with corresponding surface wave dispersion curves.
• OpenSWI-Deep: 1D velocity profiles generated from high-resolution 3D geological models, sourced globally and regionally, tailored for deep geological studies.
• OpenSWI-Real: AI-ready observational data from Long Beach, USA, and the China Seismological Reference Model Project, with 1D velocity profiles and corresponding surface wave dispersion curves.

Group	Datasets	Period Range (s)	Depth Range and Interval (km)	Extracted 1D Velocity	Augmented 1D Velocity
OpenSWI-shallow	OpenFWI-FlatVelA	0.1-10 s	0-2.8 km / 0.04 km	30,000 x 4 x 70	1,490,415 x 4 x 70
	OpenFWI-Flat-FaultA	0.1-10 s	0-2.8 km / 0.04 km	292,941 x 4 x 70	2,925,151 x 4 x 70
	OpenFWI-CurveVel	0.1-10 s	0-2.8 km / 0.04 km	295,773 x 4 x 70	2,952,975 x 4 x 70
	OpenFWI-Fold-Fault	0.1-10 s	0-2.8 km / 0.04 km	537,774 x 4 x 70	5,369,692 x 4 x 70
	OpenFWI-StyleA	0.1-10 s	0-2.8 km / 0.04 km	2,344,958 x 4 x 70	9,345,103 x 4 x 70
OpenSWI-deep	LITHO1.0	1-100 s	0-300 km / 1.0 km	40,959 x 4 x 300	245,771 x 4 x 301
	USTClitho1.0	1-100 s	0-300 km / 1.0 km	9,125 x 4 x 300	54,750 x 4 x 301
	Central-and-Western US	1-100 s	0-300 km / 1.0 km	6,803 x 4 x 300	40,818 x 4 x 301
	Continental China	1-100 s	0-300 km / 1.0 km	4,516 x 4 x 300	27,096 x 4 x 301
	US Upper-Mantle	1-100 s	0-300 km / 1.0 km	3,678 x 4 x 300	22,061 x 4 x 301
	EUcrust	1-100 s	0-300 km / 1.0 km	43,520 x 4 x 300	261,155 x 4 x 301
	Alaska	1-100 s	0-300 km / 1.0 km	19,408 x 4 x 300	116,448 x 4 x 301
	CSEM-Europe	1-100 s	0-300 km / 1.0 km	21,931 x 4 x 300	131,586 x 4 x 301
	CSEM-Eastmed	1-100 s	0-300 km / 1.0 km	12,782 x 4 x 300	76,692 x 4 x 301
	CSEM-Iberian	1-100 s	0-300 km / 1.0 km	9,102 x 4 x 300	54,612 x 4 x 301
	CSEM-South Atlantic	1-100 s	0-300 km / 1.0 km	7,371 x 4 x 300	44,226 x 4 x 301
	CSEM-North Atlantic	1-100 s	0-300 km / 1.0 km	14,541 x 4 x 300	87,246 x 4 x 301
	CSEM-Japan	1-100 s	0-300 km / 1.0 km	14,641 x 4 x 300	87,846 x 4 x 301
	CSEM-Astralasia	1-100 s	0-300 km / 1.0 km	4,131 x 4 x 300	24,786 x 4 x 301
OpenSWI-real	LongBeach	0.263 - 1.666 s	0-1.4 km / 0.04 km	5,297 x 4 x 35	-
	CSRM	8 - 70 s	0-120 km / 1.0 km	12,901 x 4 x 120	-

---

🔄 SWIDP: Integrated Workflow for Dataset Construction

SWIDP provides a fully modular pipeline for constructing large-scale surface-wave dispersion curve datasets:

1. Collect and standardize geological models
   Aggregate 2D/3D geological–geomorphological models from public databases and literature, and unify data format through cleaning, parameter conversion, and normalization.

2. Build 1D velocity models
   Extract representative 1D shear-wave velocity profiles, de-duplicate, optimize thin layers, interpolate to uniform thickness, and complete missing parameters (e.g., (v_p), density).

3. Augment for geological diversity
   Apply perturbation-based and generative-model-based enhancements to improve geological diversity and boundary complexity coverage.

4. Forward modeling of dispersion curves
   Efficiently compute large-scale surface-wave dispersion curves (using an optimized Disba-based solver with parallelization) across diverse period ranges.

flowchart LR
    A[Collect & Standardize Geological Models] --> B[Build 1D Velocity Models]
    B --> C[Augment for Geological Diversity]
    C --> D[Forward Modeling of Dispersion Curves]

Loading

---

📚 Example 1: Building the OpenSWI-shallow Dataset

import numpy as np
import sys
sys.path.append("OpenSWI/Datasets/OpenSWI/")
from SWIDP.process_1d_shallow import augment_workflow
from SWIDP.dispersion import (
    generate_mixed_samples, calculate_dispersion,
    transform_vp_to_vs, transform_vs_to_vel_model
)
from p_tqdm import p_map

# Step 1: Load 1D velocity model n x (depth, vp)
depth_vp = np.loadtxt(
    "./OpenSWI/Datasets/OpenSWI/Datasets-Construction/OpenSWI-shallow/0.2-10s-Aug/vp_demo.txt"
)
depth = depth_vp[:, 0]
vp = depth_vp[:, 1]

# Step 2: Convert vp to vs
vs = transform_vp_to_vs(vp)

# Step 3: Augment vs models
augment_nums = 100
vs_augmented = augment_workflow(
    vs, depth,
    perturb_num=augment_nums,
    vs_perturbation=0.05,           # relative ratio
    thickness_perturbation=0.1,     # relative ratio
    vel_threshold=0.05,             # km/s
    thickness_threshold=0.01,       # km
    min_layers_num=3,
    smooth_vel=False,
    smooth_nodes=10
)

# Step 4: Build full velocity models (depth, vp, vs, rho)
vel_model_augmented = p_map(
    transform_vs_to_vel_model,
    list(vs_augmented),
    [depth] * len(vs_augmented)
)

# Step 5: Generate dispersion curves [t, phase velocity, group velocity]
t = generate_mixed_samples(
    num_samples=100, start=0.2, end=10,
    uniform_num=50, log_num=20, random_num=30
)
disp = p_map(
    calculate_dispersion,
    vel_model_augmented,
    [t] * len(vel_model_augmented)
)

Example of OpenSWI-shallow data:

Details of this example can be found at

Dataset Type	Description	Link
OpenSWI-shallow Example	Example notebook for the OpenSWI dataset	OpenSWI-shallow-example.ipynb
Flat	Flat velocity model construction	OpenFWI-FlatVel-A.ipynb
Flat-Fault	Flat velocity model with fault	OpenFWI-FlatFault-A.ipynb
Fold	Fold velocity model construction	OpenFWI-CurveVel-A.ipynb
Fold-Fault	Fold velocity model with fault	OpenFWI-CurveFault-A.ipynb
Field	Field data modeling	OpenFWI-Style-A.ipynb

---

📚 Example 2: Building the OpenSWI-deep Dataset

import numpy as np
import sys
sys.path.append("../../../")
from SWIDP.process_1d_deep import *
from SWIDP.dispersion import generate_mixed_samples,calculate_dispersion,transform_vs_to_vel_model
from p_tqdm import p_map

# step1: get 1d velocity model (vp model or vs)
depth_vs = np.loadtxt("./Datasets/Datasets-Construction/OpenSWI-deep/1s-100s-Aug/vs_demo.txt")
depth = depth_vs[:,0]
vs = depth_vs[:,1]

# step2-1: remove the thin sandwidth layer
vs = combine_thin_sandwich(vs,
                            depth,
                            thickness_threshold=12, # km
                            uniform_thickness=1,    # km (thickness_threshold/uniform_thickness) = max_check_layers
                            gradient_threshold=0.05, # km/s (gradient_threshold)
                            return_idx=False
                            )

# step2-2: smooth the vs model (selectable)
vs = smooth_vs_by_node_interp(vs,
                            depth,
                            n_nodes=20,
                            method="pchip"
                            )

# step3: find moho index
moho_idx = find_moho_depth(vs,
                           depth,
                           [5,90], # range of moho index
                           gradient_search=False,
                           gradient_threshold=0.01)

# step4: augment the vs model
perturb_nums = 100
vs_augmented = p_map(augment_crust_moho_mantle,
                    [vs]*perturb_nums,
                    list(depth.reshape(1,-1))*perturb_nums,
                    [moho_idx]*perturb_nums,
                    [[-0.1,0.1]]*perturb_nums, # relative ratio
                    [[3,8]]*perturb_nums,     # nodes for crust
                    [[8,15]]*perturb_nums,   # nodes for mantle
                    [3]*perturb_nums,          # km 
                    [2]*perturb_nums,     # km
                    [False]*perturb_nums,
                    np.random.randint(0,1000000,perturb_nums)
                    )

# step5: transform the vs model to vp model
vel_models = p_map(transform_vs_to_vel_model,list(vs_augmented),[depth]*len(vs_augmented))

# step6: calculate the dispersion curve
t = generate_mixed_samples(num_samples=300,start=1,end=100,uniform_num=100,log_num=100,random_num=100)
t = np.ones((len(vel_models),len(t)))*t
disp_data = p_map(calculate_dispersion, vel_models, list(t))
disp_data = np.array(disp_data)
vel_models = np.array(vel_models)

Example of OpenSWI-deep data:

Details of this example can be found at:

Dataset Name	Reference	Link
OpenSWI-deep-example	-	OpenSWI-deep-example.ipynb
LITHO1.0	Pasyanos et al., 2014	LITHO1.0
USTClitho1.0	Xin et al., 2018	USTClitho1.0
Central-and-Western US	Shen et al., 2013	Central-and-Western US
Continental China	Shen et al., 2016	Continental China
US Upper-Mantle	Xie et al., 2018	US Upper-Mantle
EUCrust	Lu et al., 2018	EUCrust
Alaska	Berg et al., 2020	Alaska
CSEM-Europe	Blom et al., 2020; Fichtner et al., 2018; Çubuk-Sabuncu et al., 2017	CSEM-Europe
CSEM-Eastmed	Blom et al., 2020; Fichtner et al., 2018	CSEM-Eastmed
CSEM-Iberian	Fichtner et al., 2018; Fichtner et al., 2015	CSEM-Iberian
CSEM-South Atlantic	Fichtner et al., 2018; Colli et al., 2013	CSEM-South Atlantic
CSEM-North Atlantic	Fichtner et al., 2018; Krischer et al., 2018	CSEM-North Atlantic
CSEM-Japan	Fichtner et al., 2018; Simutė et al., 2016	CSEM-Japan
CSEM-Australasia	Fichtner et al., 2018; Fichtner et al., 2010	CSEM-Australasia

---

🧠 Deep Learning-based Surface Wave Dispersion Curve Inversion

This method leverages deep learning to directly invert surface wave dispersion curves (including period, phase velocity, and group velocity) into a 1D shear wave velocity profile of the subsurface. We use a Transformer-based framework for end-to-end inversion, where the input dispersion curves are processed through CNN modules to extract features, and Transformer blocks capture long-range dependencies to enhance inversion accuracy and stability. A depth-aware strategy dynamically adjusts the inversion depth range based on the input data, improving model performance. The loss function is based on Mean Squared Error (MSE), calculated only for the effective depth range. Additionally, noise and missing data are simulated during training to enhance the model's robustness.

flowchart LR
    subgraph row1 [ ]
        style row1 fill:#f0f8f0,stroke:#4d4d4d,stroke-width:2px
        A[Input: Dispersion Curves] --> B[Embedding]
        B --> C[Transformer Blocks]
        C --> D[Output: 1D vs Profile]
        style A fill:#e3f9e5,stroke:#4d4d4d,stroke-width:2px
        style B fill:#d9f7d0,stroke:#4d4d4d,stroke-width:2px
        style C fill:#c6f5c1,stroke:#4d4d4d,stroke-width:2px
        style D fill:#aee7b5,stroke:#4d4d4d,stroke-width:2px
    end

    subgraph row2 [ ]
        style row2 fill:#f0f8f0,stroke:#4d4d4d,stroke-width:2px
        D --> E[Depth-Aware Strategy Adjusts Inversion Depth Range]
        E --> F[Loss Function: MSE for Effective Depth Range]
        F --> G[Training with Noise and Missing Data Simulation]
        style E fill:#d0f8d9,stroke:#4d4d4d,stroke-width:2px
        style F fill:#b7f2b0,stroke:#4d4d4d,stroke-width:2px
        style G fill:#a8e6a3,stroke:#4d4d4d,stroke-width:2px
    end

Loading

• Input: Surface wave dispersion curves (period, phase velocity, and group velocity).
• Neural Network:
  • CNN Embedding: Extracts features from the input data.
  • Transformer Modeling: Captures long-range dependencies to improve inversion accuracy.
  • Output: Produces a 1D shear wave velocity profile.
• Depth-Aware Strategy:
  Dynamically adjusts the inversion depth based on the data.
• Loss Function:
  Optimizes using Mean Squared Error (MSE) for the effective depth range.
• Training:
  Simulates noise and missing data to enhance model robustness.

The package for training a deep learning model for Surface Wave Inversion can be found at SWInversion. We encourage users to utilize the pre-existing transformer-based SWI inversion network, Dispformer, for enhanced performance.

---

📚 Training Example

import torch
from torch.utils.data import ConcatDataset, Dataset, DataLoader, random_split
import numpy as np
from scipy import interpolate
import sys
sys.path.append("../../../")
from SWInversion.data_augmentation import *
from SWInversion.dataloader_pretrain import *
from SWInversion.misfits import *
from SWInversion.training import *
from SWInversion.args import *
from tqdm import tqdm
import json, os, shutil, pprint
pp = pprint.PrettyPrinter(indent=4)

# loading args
args = parse_args(json.load(open("training_example_settings.json")),jupyter=True)
pp.pprint(vars(args))

#----------------------------------------------------
# Load datasets
#----------------------------------------------------
base_path  = "Basic Path to OpenSWI datasets"
model_path = os.path.join(base_path,"model.npz")
disp_path = os.path.join(base_path,"disp.npz")

datasets = DispersionDatasets(
            input_data_path=disp_path, input_label_path=model_path,train=args.train,
            interp_layer=args.interp_layer, interp_layer_thickness=args.interp_layer_thickness, interp_layer_number=args.interp_layer_number,
            interp_kind=args.interp_kind, layer_used_range=args.layer_used_range,
            num_workers=args.num_workers, normalize_input_data=args.normalize_input_data,
            aug_train_data=args.aug_train_data, aug_noise_level=args.aug_noise_level,
            aug_random_mask_ratio=args.aug_random_mask_ratio, aug_max_masking_length=args.aug_max_masking_length, aug_remove_group_ratio=args.aug_remove_group_ratio,
            aug_remove_phase_ratio=args.aug_remove_phase_ratio, aug_varylength=args.aug_varylength, aug_varylength_start_range=args.aug_varylength_start_range,
            aug_varylength_min_length=args.aug_varylength_min_length,
            aug_varylength_c1=args.aug_varylength_c1, aug_varylength_c2=args.aug_varylength_c2, aug_varylength_c3=args.aug_varylength_c3,
        )

# split the datasets into train/valid/test = 8/1/1 for each sub-dataset
train_dataset, valid_dataset, test_dataset = random_split(datasets, [0.9, 0.05, 0.05], generator=torch.Generator().manual_seed(args.random_seed))

train_loader = DataLoader(train_dataset, batch_size=args.batch_size,shuffle=True,num_workers=args.num_workers,pin_memory=True,collate_fn=auto_collate_fn)
val_loader = DataLoader(valid_dataset, batch_size=args.batch_size,shuffle=False,num_workers=args.num_workers,pin_memory=True,collate_fn=auto_collate_fn)
test_loader = DataLoader(test_dataset,batch_size=args.batch_size,shuffle=False,num_workers=args.num_workers,pin_memory=True,collate_fn=auto_collate_fn)

# deep learning model
model = get_dl_model(args)

# loss function
loss_fn = LossFunction(args.loss_type)
reg_fn = VelocityRegularization(reg_type=args.reg_type, reg_weight=args.reg_weight) if args.reg_type is not None else None

# learning rate update scheduler 
total_steps = args.num_epochs * len(train_loader)
scheduler = get_scheduler(optimizer, args, total_steps)

# early stopping
if args.early_stopping_flag:
    early_stopping = EarlyStopping(patience=args.early_stopping_patience, verbose=True)
else:
    early_stopping = None

# save path
save_path,save_path_name = get_save_path_name(args)

#----------------------------------------------------
#  Dispersion Transformer training
#----------------------------------------------------
pbar = tqdm(range(args.num_epochs), desc="Training")
train_loss_epoch,valid_loss_epoch = [],[]
train_step_loss_epoch,valid_step_loss_epoch = [],[]
best_valid_loss = float('inf')
for epoch in pbar:
    # training
    train_loss,train_step_loss = train_model(model,train_loader,optimizer,scheduler,loss_fn,reg_fn,device=args.device,model_name=args.model_name,args=args)
    train_loss_epoch.append(train_loss)
    train_step_loss_epoch.append(train_step_loss)
    
    # validation
    valid_loss,valid_step_loss = valid_model(model,val_loader,loss_fn,device=args.device,model_name=args.model_name,args=args)
    valid_loss_epoch.append(valid_loss)
    valid_step_loss_epoch.append(valid_step_loss)
    
    # pbar set the loss
    pbar.set_description(f"Epoch: {epoch+1}, Train {args.loss_type}: {train_loss_epoch[-1]:.6f}, Valid {args.loss_type}: {valid_loss_epoch[-1]:.6f}")
    
    # early stopping
    if early_stopping is not None:
        early_stopping(valid_loss)
        if early_stopping.early_stop:
            print("Early stopping")
            break
    
    # save the model
    if (epoch+1) % args.save_epochs == 0:
        # copy the best model to the current model
        shutil.copy(f"{save_path}/best_model.pth", f"{save_path}/model_{epoch+1}.pth")
        
    # save the best validation loss model
    if valid_loss_epoch[-1] < best_valid_loss:
        best_valid_loss = valid_loss_epoch[-1]
        torch.save(model.state_dict(), f"{save_path}/best_model.pth")
    
    # update the loss to the .npy file
    np.save(f"{save_path}/train_loss_epoch.npy", np.array(train_loss_epoch))
    np.save(f"{save_path}/valid_loss_epoch.npy", np.array(valid_loss_epoch))
    np.save(f"{save_path}/train_step_loss_epoch.npy", np.array(train_step_loss_epoch))
    np.save(f"{save_path}/valid_step_loss_epoch.npy", np.array(valid_step_loss_epoch))

print(f"Best valid loss: {best_valid_loss:.4f}")

Details of Training Examples

Dataset	Link
OpenSWI-shallow	OpenSWI-shallow Example
OpenSWI-deep	OpenSWI-deep Example

---

🔖 Citation

If you use this dataset in your research, please cite:

@article{liu2025openswi,
  title     = {OpenSWI: A Massive-Scale Benchmark Dataset for Surface Wave Dispersion Curve Inversion},
  author    = {Liu, Feng and Zhao, Sijie and Gu, Xinyu and Ling, Fenghua and Li, Yaxing and Su, Rui and Zhuang, Peiqin and Lihua Fang and Zhou, Lianqing and Huang, Jianping and Bai, Lei},
  year      = {2025},
  journal   = {arXiv preprint arXiv:2508.10749},
  url       = {https://arxiv.org/abs/2508.10749},
  doi       = {10.48550/arXiv.2508.10749},
  note      = {Dataset and code available at https://github.com/liufeng2317/OpenSWI}
}

---

📧 Contact

Principal Developer

Feng Liu

🏛️ Affiliations:

• Shanghai Artificial Intelligence Laboratory
• Shanghai Jiao Tong University

📧 Contact Information:

• 
• 

---

⚖️ License

This project is licensed under the Creative Commons Attribution 4.0 International License - see the LICENSE file for details.

---

🔍 Reference

Click to expand related works

1. X. Xiao et al., "CSRM‐1.0: A China Seismological Reference Model," JGR Solid Earth, vol. 129, no. 9, p. e2024JB029520, Sept. 2024, doi: 10.1029/2024JB029520.

2. L. Fu et al., "Improved high‐resolution 3D vs model of Long Beach, CA: Inversion of multimodal dispersion curves from ambient noise of a dense array," Geophys. Res. Lett., vol. 49, no. 4, p. e2021GL097619, Feb. 2022, doi: 10.1029/2021GL097619.

3. C. Deng et al., “OpenFWI: Large-scale multi-structural benchmark datasets for full waveform inversion,” Neural Information Processing Systems, Nov. 2021, Accessed: Feb. 29, 2024. [Online]. Available: https://www.semanticscholar.org/paper/2d13799dcbd0aefb08c379f58cd6004b1376ca33

4. N. Blom et al., "Seismic waveform tomography of the central and eastern Mediterranean upper mantle," Solid Earth, vol. 11, no. 2, pp. 669–690, Apr. 2020, doi: 10.5194/se-11-669-2020.

5. E. M. Berg et al., "Shear velocity model of Alaska via joint inversion of Rayleigh wave ellipticity, phase velocities, and receiver functions across the Alaska transportable array," J. Geophys. Res.: Solid Earth, vol. 125, no. 2, Feb. 2020, doi: 10.1029/2019jb018582.

6. H. Xin et al., "High‐resolution lithospheric velocity structure of continental China by double‐difference seismic travel‐time tomography," Seismol. Res. Lett., vol. 90, no. 1, pp. 229–241, Jan. 2019, doi: 10.1785/0220180209.

7. J. Xie et al., "3-D upper-mantle shear velocity model beneath the contiguous United States based on broadband surface wave from ambient seismic noise," Pure Appl. Geophys., vol. 175, no. 10, pp. 3403–3418, Oct. 2018, doi: 10.1007/s00024-018-1881-2.

8. Y. Lu et al., "High-resolution surface wave tomography of the European crust and uppermost mantle from ambient seismic noise," Geophys. J. Int., vol. 214, no. 2, pp. 1136–1150, Aug. 2018, doi: 10.1093/gji/ggy188.

9. L. Krischer et al., "Automated large‐scale full seismic waveform inversion for North America and the North Atlantic," J. Geophys. Res.: Solid Earth, vol. 123, no. 7, pp. 5902–5928, July 2018, doi: 10.1029/2017JB015289.

10. A. Fichtner et al., "The collaborative seismic earth model: generation 1," Geophys. Res. Lett., vol. 45, no. 9, pp. 4007–4016, May 2018, doi: 10.1029/2018gl077338.

11. Y. Çubuk-Sabuncu et al., "3-D crustal velocity structure of western Turkey: constraints from full-waveform tomography," Phys. Earth Planet. Inter., vol. 270, pp. 90–112, Sept. 2017, doi: 10.1016/j.pepi.2017.06.014.

12. S. Simutė et al., "Full‐waveform inversion of the Japanese islands region," J. Geophys. Res.: Solid Earth, vol. 121, no. 5, pp. 3722–3741, May 2016, doi: 10.1002/2016jb012802.

13. W. Shen et al., "A seismic reference model for the crust and uppermost mantle beneath China from surface wave dispersion," Geophys. J. Int., vol. 206, no. 2, pp. 954–979, Aug. 2016, doi: 10.1093/gji/ggw175.

14. A. Fichtner and A. Villaseñor, "Crust and upper mantle of the western Mediterranean – constraints from full-waveform inversion," Earth Planet. Sci. Lett., vol. 428, pp. 52–62, Oct. 2015, doi: 10.1016/j.epsl.2015.07.038.

15. M. E. Pasyanos et al., "LITHO1.0: An updated crust and lithospheric model of the Earth," JGR Solid Earth, vol. 119, no. 3, pp. 2153–2173, Mar. 2014, doi: 10.1002/2013JB010626.

16. W. Shen et al., "A 3‐D model of the crust and uppermost mantle beneath the Central and Western US by joint inversion of receiver functions and surface wave dispersion," JGR Solid Earth, vol. 118, no. 1, pp. 262–276, Jan. 2013, doi: 10.1029/2012JB009602.

17. F. Rickers et al., "The Iceland–Jan Mayen plume system and its impact on mantle dynamics in the North Atlantic region: evidence from full-waveform inversion," Earth Planet. Sci. Lett., vol. 367, pp. 39–51, Apr. 2013, doi: 10.1016/j.epsl.2013.02.022.

18. L. Colli et al., "Full waveform tomography of the upper mantle in the South Atlantic region: imaging a westward fluxing shallow asthenosphere?," Tectonophysics, vol. 604, pp. 26–40, Sept. 2013, doi: 10.1016/j.tecto.2013.06.015.

19. C. Trabant et al., "Data products at the IRIS DMC: stepping stones for research and other applications," Seismol. Res. Lett., vol. 83, no. 5, pp. 846–854, Sept. 2012, doi: 10.1785/0220120032.

20. A. Fichtner et al., "Full waveform tomography for radially anisotropic structure: new insights into present and past states of the Australasian upper mantle," Earth Planet. Sci. Lett., vol. 290, no. 3–4, pp. 270–280, Feb. 2010, doi: 10.1016/j.epsl.2009.12.003.

21. A. Fichtner et al., "Full seismic waveform tomography for upper-mantle structure in the Australasian region using adjoint methods," Geophys. J. Int., vol. 179, no. 3, pp. 1703–1725, Dec. 2009, doi: 10.1111/j.1365-246x.2009.04368.x.