Surfdisp96-Roccastrada-10k-2025 / dataset_info.yaml
schattin's picture
Update dataset_info.yaml
075e573 verified
# =====================================================================
# MIGRATE Synthetic Dataset Metadata
# Dataset: Surfdisp96-Roccastrada-10m
# Author: Nils Schaetti (DMML - HEG Genève)
# License: CC BY-SA 4.0
# =====================================================================
dataset_name: "Surfdisp96-Roccastrada-10m" # Unique dataset identifier
dataset_type: "surfdisp"
description: >
This dataset contains synthetic seismic models and their corresponding
Rayleigh-wave dispersion curves generated using forward modeling with
the Roccastrada priors. It is designed for benchmarking inversion
algorithms and training machine learning models in geophysics.
# ---------------------------------------------------------------------
# Priors used to generate the synthetic seismic models
# ---------------------------------------------------------------------
priors:
vs: [0.5, 4.0] # Range of shear-wave velocities (km/s)
z: [0.0, 5.0] # Depth range in kilometers
layers: [2, 20] # Minimum and maximum number of layers
vpvs: 1.3 # Ratio between P-wave and S-wave velocities (Vp/Vs)
mohoest: null # Estimated Moho depth (None for synthetic generation)
mantle: null # Mantle layer properties (not used)
thickmin: 0.1 # Minimum layer thickness in km
lvz: null # Low Velocity Zone (unused)
hvz: null # High Velocity Zone (unused)
# ---------------------------------------------------------------------
# Model-specific parameters for forward simulation
# ---------------------------------------------------------------------
model_parameters:
dispersion_curve_length: 108 # Number of samples per dispersion curve
# ---------------------------------------------------------------------
# Fold configuration for cross-validation splits
# ---------------------------------------------------------------------
folds:
available: ["2-fold", "5-fold", "10-fold"] # Supported fold configurations
fold_file: "folds.json" # File describing shard allocation per fold
# ---------------------------------------------------------------------
# File format and dataset metadata
# ---------------------------------------------------------------------
format: "parquet" # Dataset stored in Apache Parquet format
license: "CC BY-SA 4.0" # Dataset license (Creative Commons Attribution-ShareAlike 4.0)
created_by: "Nils Schaetti" # Creator or generating system
creation_date: "2025-09-16 15:39:27" # Timestamp of dataset generation
# ---------------------------------------------------------------------
# Parameters for synthetic data generation
# ---------------------------------------------------------------------
generation:
seed: 43 # Random seed for reproducibility
random_generator: "numpy.default_rng" # RNG backend used
n_samples: 10000 # Total number of synthetic samples
samples_per_shard: 1000 # Samples per Parquet shard
n_shards: 10 # Total number of shards
source: "sample_model + forward" # Generation method (sampling + forward simulation)
# ---------------------------------------------------------------------
# Feature schema describing structure of each record
# ---------------------------------------------------------------------
features:
vs:
type: "list<float32>" # List of shear-wave velocities (per layer)
variable_length: true # Number of layers varies per sample
z:
type: "list<float32>" # Depth boundaries (km) for each layer
variable_length: true # Same variable-length structure as 'vs'
h:
type: "list<float>" # Layer thickness (km), depends on prior
variable_length: true # Depends on the number of layers
vp:
type: "list<float>" # P-wave velocities per layer (km/s)
variable_length: true # Same variable-length structure as 'vs'
vpvs:
type: "float32" # Single scalar value (Vp/Vs ratio)
nlayers:
type: 'int' # Number of layers
velmap_vs:
type: 'list<float>' # Shear velocitiy profile (vs) interpolated (flat) to 60 points
length: 60
velmap_z:
type: 'list<float>' # Depth grid corresponding to velmap_vs
length: 60
# Low Range Dispersion Curve
L_disp_x:
type: "list<float32>" # Periods or frequencies (x-axis of dispersion curve)
length: 108 # Fixed number of samples per curve
L_disp_y:
type: "list<float32>" # Phase or group velocities (y-axis of dispersion curve)
length: 108 # Same fixed length as disp_x
L_wave_type:
type: "string" # Type of surface wave (e.g., Rayleigh, Love)
L_velocity_type:
type: "string" # Type of velocity model (e.g., Vs, Vp, Vp/Vs)
# Middle Range Dispersion Curve
M_disp_x:
type: "list<float32>" # Periods or frequencies (x-axis of dispersion curve)
length: 108 # Fixed number of samples per curve
M_disp_y:
type: "list<float32>" # Phase or group velocities (y-axis of dispersion curve)
length: 108 # Same fixed length as disp_x
M_wave_type:
type: "string" # Type of surface wave (e.g., Rayleigh, Love)
M_velocity_type:
type: "string" # Type of velocity model (e.g., Vs, Vp, Vp/Vs)
# High Range Dispersion Curve
H_disp_x:
type: "list<float32>" # Periods or frequencies (x-axis of dispersion curve)
length: 108 # Fixed number of samples per curve
H_disp_y:
type: "list<float32>" # Phase or group velocities (y-axis of dispersion curve)
length: 108 # Same fixed length as disp_x
H_wave_type:
type: "string" # Type of surface wave (e.g., Rayleigh, Love)
H_velocity_type:
type: "string" # Type of velocity model (e.g., Vs, Vp, Vp/Vs)