xpertsystems/oil001-sample

OIL-001 — Synthetic Seismic Survey Dataset (Sample)

XpertSystems.ai Synthetic Data Platform · SKU: OIL001-SAMPLE · Version 1.0.0

This is a free preview of the full OIL-001 — Synthetic Seismic Survey Dataset product. It contains roughly ~25% of the full dataset at identical schema, physics modeling, and seismic interpretation features, so you can evaluate fit before licensing the full product.

File	Rows (sample)	Rows (full)	Description
seismic_traces.csv	~28,800	~100,000	Per-trace amplitude + attributes (25 cols)
horizon_catalog.csv	~5,850	~30,000	Interpreted horizons with rock physics (27 cols)
survey_geometry.csv	~3	~6	Per-survey geometry & acquisition (23 cols)
interpretation_summary.csv	~3	~6	Per-survey interpretation KPIs (18 cols)

Dataset Summary

OIL-001 is a full seismic simulation engine producing realistic 2D/3D seismic waveforms with subsurface structure interpretation labels — the kind of data exploration geophysicists, reservoir engineers, and seismic AI companies (Bluware, Earth Science Analytics, OspreyData, Geoteric) build their models on, but synthetic and free under CC-BY-NC for research.

Physics modeling:

• Convolutional seismic model: reflectivity series × wavelet (Ricker or zero-phase)
• Aki-Richards approximation: angle-dependent reflection coefficients for AVO analysis
• Gassmann fluid substitution: mineral/dry-frame/fluid moduli per layer
• Gardner's relation: density from Vp by lithology
• NMO (Normal Moveout): velocity-based gather flattening for CDP stack
• Hilbert transform: instantaneous amplitude, phase, frequency attributes
• Negative-pressure coherence: fault-zone similarity attribute

Survey acquisition modeling:

• 3D land, 3D marine, 2D 2D-towed-streamer survey types
• Configurable inline/crossline grid spacing
• 9 offset panels (0°-40° in 5° steps) for AVO gathers
• CDP fold, sample interval (2 ms = 500 Hz), 4-second TWT record length

Subsurface structure model (8 structure labels):

• Anticline crest / Anticline flank / Syncline
• Fault plane / Fault shadow
• Salt flank
• Stratigraphic trap
• Flat background

DHI (Direct Hydrocarbon Indicator) injection (6 DHI labels):

• No DHI (most traces)
• Bright spot (amplitude anomaly indicating gas)
• Dim spot (amplitude reduction indicating oil)
• Flat spot (fluid contact reflection)
• Polarity reversal (Class IIp AVO response)
• AVO anomaly (Class III gas response)

8 lithology types:

sandstone (reservoir), shale (seal), limestone, dolomite, salt, anhydrite, volcanic, basement

5 fluid types (with Gassmann fluid substitution):

brine, oil, gas, gas_condensate, tight_dry

Seismic noise injection (7 noise types):

• Surface multiple reflections
• Interbed multiple reflections
• Ambient random noise
• Source interference
• Coupling variation
• Statics anomaly
• Clean (no noise)

Per-trace seismic attributes (industry-standard interpretation features):

• Raw and processed amplitudes
• Instantaneous frequency, phase, amplitude (Hilbert transform)
• Envelope (dB)
• Coherence score (fault-zone similarity)
• Curvature (k1 principal curvature)
• AVO intercept and gradient (Aki-Richards)
• Noise type and flag

Per-horizon rock physics:

• Vp, Vs, Vp/Vs ratio
• Density (g/cc, Gardner-derived)
• Acoustic impedance
• Porosity %, water saturation %
• Net pay, dip angle, dip azimuth

Prospect risking (per-horizon):

• PGOS (Probability of Geological Success) — typical industry 0.20-0.40
• Trap closure area (km²)
• Reservoir quality class
• Fault association

Calibrated Validation Results

Sample validation results across 10 seismic-domain KPIs:

Metric	Observed	Target	Source	Verdict
n_surveys_represented	3	3	Sample survey count	✓ PASS
n_survey_types	2	2	2D + 3D + 4D coverage	✓ PASS
n_basins_represented	3	3	Geographic diversity	✓ PASS
dominant_frequency_hz	37.00	38.00	Standard seismic dominant freq	✓ PASS
snr_db_mean	21.03	22.00	Industry SNR target (post-stack)	✓ PASS
horizon_pick_confidence	0.861	0.880	SEG interpreter confidence	✓ PASS
horizon_continuity_index	0.868	0.830	Lateral continuity benchmark	✓ PASS
fault_detection_confidence	0.813	0.790	Coherence-based fault picking	✓ PASS
sandstone_porosity_pct_mean	16.93	17.50	Industry sandstone reservoir φ	✓ PASS
n_horizon_labels_represented	8	3	Multi-class horizon diversity	✓ PASS

Note: This dataset is designed for seismic interpretation AI training — buyers building horizon-picking auto-trackers, fault-detection neural networks, DHI classification models, or AVO inversion solvers can use these labels for supervised training. The full product includes 6 surveys with larger 3D grids and full structural diversity.

Schema Highlights

seismic_traces.csv (primary file, 25 columns)

Trace identification & geometry:

Column	Type	Description
survey_id, gather_id	int	Survey and CDP gather IDs
cdp_x, cdp_y	float	CDP location (meters)
cdp_inline, cdp_crossline	int	3D grid coordinates
offset_m	float	Source-receiver offset (m)
two_way_time_ms	float	Two-way time (ms)

Amplitudes & attributes:

Column	Type	Description
amplitude_raw	float	Pre-processing amplitude
amplitude_processed	float	After deconv + statics + multiples
frequency_hz	float	Instantaneous frequency
instantaneous_phase_deg	float	Hilbert phase (degrees)
instantaneous_amplitude	float	Hilbert envelope
envelope_db	float	Envelope in dB
coherence_score	float	Coherence (0-1, fault zones low)
curvature_k1	float	Principal curvature
avo_intercept, avo_gradient	float	Aki-Richards AVO A, B

Labels:

Column	Type	Description
horizon_label	string	8 horizon labels (basement, reservoir, etc.)
structure_label	string	8 structure labels (anticline, fault, salt)
dhi_label	string	6 DHI labels
noise_flag, noise_type	string	Noise contamination flag
fault_proximity_m	float	Distance to nearest fault
reservoir_quality_flag	string	Reservoir quality class

horizon_catalog.csv (27 columns)

Per-horizon rock physics + structural interpretation:

Column	Description
horizon_name, horizon_label	Named horizon + 8-class label
two_way_time_ms, depth_m	TWT and depth
dip_angle_deg, dip_azimuth	Structural dip
interval_velocity_ms	Interval Vp (m/s)
acoustic_impedance_mpa	Acoustic impedance
vp_ms, vs_ms	Vp, Vs (m/s)
density_gcc	Density (g/cc, Gardner)
porosity_pct	Porosity (%)
water_saturation_pct	Water saturation (%)
lithology_type	8 lithology classes
fluid_type	5 fluid types
trap_style	Trap classification
dhi_class, dhi_label	DHI class + 6-class label
reservoir_quality_class	Reservoir quality
net_pay_m	Net pay thickness
fault_id	Associated fault
prospect_pgos	Probability of Geological Success

survey_geometry.csv (23 columns)

Per-survey acquisition geometry: survey_type, basin_name, n_inlines, n_crosslines, cdp_spacing, fold, source_type, water_depth, acquisition environment, processing vintage, migration type, SNR, dominant frequency.

interpretation_summary.csv (18 columns)

Per-survey interpretation KPIs: horizon_pick_confidence, fault_confidence, horizon_continuity_index, multiple_contamination_pct, etc.

Suggested Use Cases

• Auto horizon picking — train CNNs to follow horizons across 3D
• Fault detection — train coherence/curvature-based fault networks
• DHI classification — 6-class hydrocarbon indicator prediction
• AVO inversion — predict elastic properties from intercept/gradient
• Salt body segmentation — geometric salt detection
• Structural label segmentation — 8-class structural interpretation
• Lithology prediction — 8-class lithology from seismic attributes
• Fluid type prediction (with Gassmann substitution targets)
• Reservoir quality scoring from acoustic impedance and porosity
• Multiple suppression — deep-learning-based denoise training
• Statics correction — anomaly detection at trace level
• Wavelet estimation & deconvolution — Ricker/zero-phase fitting
• Velocity model building — interval velocity prediction from traces
• Time-to-depth conversion — TWT-depth modeling
• Prospect screening — PGOS prediction from seismic features
• Trap closure area estimation from structural attributes
• Insurtech-style geophysical AI training without proprietary survey data
• University geophysics curriculum — synthetic teaching corpus

Loading the Data

import pandas as pd

traces   = pd.read_csv("seismic_traces.csv")
horizons = pd.read_csv("horizon_catalog.csv")
geometry = pd.read_csv("survey_geometry.csv")
interp   = pd.read_csv("interpretation_summary.csv")

# Multi-class horizon label target (8 classes)
y_horizon = traces["horizon_label"]

# Multi-class structural interpretation target (8 classes)
y_structure = traces["structure_label"]

# Multi-class DHI prediction target (6 classes)
y_dhi = traces["dhi_label"]

# Binary noise contamination
y_noise = traces["noise_flag"]

# Regression: AVO inversion targets
y_avo_a = traces["avo_intercept"]
y_avo_b = traces["avo_gradient"]

# Multi-class lithology from horizon (8 classes)
y_lithology = horizons["lithology_type"]

# Multi-class fluid type from horizon (5 classes)
y_fluid = horizons["fluid_type"]

# Regression: reservoir porosity
y_porosity = horizons["porosity_pct"]

# Regression: PGOS prospect risking
y_pgos = horizons["prospect_pgos"]

# Build 3D seismic cube for ML
cube = traces.pivot_table(
    index=["cdp_inline", "cdp_crossline"],
    columns="two_way_time_ms",
    values="amplitude_processed",
    aggfunc="mean"
)

License

This sample is released under CC-BY-NC-4.0 (free for non-commercial research and evaluation). The full production dataset is licensed commercially — contact XpertSystems.ai for licensing terms.

Full Product

The full OIL-001 dataset includes 6 surveys with 120×80 3D grids each, full structural diversity (anticlines, synclines, salt diapirs, stratigraphic traps), 14-layer velocity models with Gassmann substitution, and full DHI/AVO anomaly catalogs. Calibrated to SEG seismic interpretation standards, Aki-Richards AVO theory, and Gassmann fluid substitution theory.

📧 pradeep@xpertsystems.ai 🌐 https://xpertsystems.ai

Citation

@dataset{xpertsystems_oil001_sample_2026,
  title  = {OIL-001: Synthetic Seismic Survey Dataset (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  url    = {https://huggingface.co/datasets/xpertsystems/oil001-sample}
}

Generation Details

• Generator version : 1.0.0
• Random seed : 42
• Generated : 2026-05-16 22:41:59 UTC
• Surveys : 3 × (40 inlines × 30 crosslines × 9 offsets × 14 layers)
• Wavelet : Ricker, 38 Hz dominant
• Physics : Aki-Richards AVO + Gassmann + Gardner + Hilbert
• Calibration basis : SEG seismic interpretation standards
• Overall validation: 100.0 / 100 (grade A+)