Datasets:

DBbun
/

Extreme_Environment_Generator_v1.0

pretty_name: Extreme Environment Generator
dataset_name: Extreme_Environment_Generator
annotations_creators:
  - no-annotation
language_creators:
  - no-annotation
language:
  - en
license: other
multilinguality:
  - monolingual
size_categories:
  - 1M<n<10M
source_datasets:
  - generated
task_categories:
  - time-series-forecasting
task_ids:
  - multivariate-time-series-forecasting

DBbun / Extreme_Environment_Generator

Dataset Summary
Environments
Dataset Structure
Sensors
State Fields
Use Cases
Limitations
License
Citation

Dataset Summary

Extreme_Environment_Generator is a large synthetic dataset produced by a configurable Python simulator. It generates multi-sensor time-series data from extreme physical environments inspired by:

deep-Earth subsurface
subsea environments beneath oceans
deep ice-world and icy-planet structures

Each generated scenario includes:

the true physical state of the environment over time
corrupted/noisy sensor measurements
layer and material profiles
environmental configuration
rare failure events
scenario-level summary statistics

The dataset is designed for machine learning research, including:

multivariate forecasting
sensor fusion
anomaly and rare-event detection
robustness testing under noise, drift, and missing data
reinforcement learning in extreme synthetic environments
foundation model pretraining for time series

Environments

1. Deep Subsurface

A high-pressure, high-temperature environment inspired by Earth’s crust, mantle, outer core, and inner core.
Includes 5 fixed layers from 0 km to 6371 km depth.

2. Subsea

Includes water, sediment, and crust layers, followed by deep subsurface structure.
Lower temperatures and pressures in early depth stages, transitioning to mantle-like profiles.

3. Ice World

Inspired by icy moons and exoplanets.
Includes multi-kilometer ice shells, ice–rock transition, and deep interior structure.

All three environments use the same schema, allowing direct cross-environment comparison.

Dataset Structure

state_true

Ground-truth physical state at every time step.

Field	Description	Unit
scenario_id	scenario index	—
time_sec	time since start	seconds
depth_km	simulated depth	km
true_temperature_C	real internal temperature	°C
true_pressure_GPa	real pressure	GPa
true_density_kg_m3	real density	kg/m³
true_phase	solid/liquid/plastic	—
true_seismic_noise	baseline seismic activity	arbitrary
true_vibration_level	structural vibration	arbitrary
stability_score	stability (0–1)	—
rare_event_flag	True if an extreme event occurred	bool

sensors

Corrupted sensor readings derived from state_true.

Sensor	Field Prefix	Description
Temperature	temperature_	noisy, drifting temperature
Pressure	pressure_	noisy pressure with dropout
Accelerometer	accel_	acceleration in g
Seismic	seismic_	seismic noise + dropout
Quantum Resonance	quantum_resonance_	fictional high-sensitivity sensor
Neutrino Flux	neutrino_flux_	fictional deep-environment reading
Spin Coherence	spin_coherence_	fictional material spin measurement
Gravity Wave	gravity_wave_	fictional micro-gravity perturbations

All sensors include controlled noise, drift, and dropout behavior.

failures

Rare catastrophic or near-catastrophic events.

Field	Description
scenario_id	scenario index
time_sec	event time
depth_km	event depth
event_type	“thermal_spike”, “pressure_surge”, etc.
severity	0–1 normalized severity

layer_profile

Static description of each environment’s internal layers.

Field	Description
scenario_id	scenario index
layer_id	layer index
name	layer name
depth_start_km	start depth
depth_end_km	end depth
material_name	material
density_kg_m3	perturbed density
phase	solid/plastic/liquid

environment_config

The full configuration of each scenario (planet, sensors, perturbations, etc.).

summary

High-level statistics per scenario, including max temperature, average pressure, number of failures, and stability metrics.

Detailed Sensor Descriptions

Temperature Sensor

Gaussian noise in °C
Linear drift over time
Occasional dropout

Pressure Sensor

Multiplicative noise (%)
Dropout probability
Tracks large pressure gradients

Accelerometer

Measures vibration and acceleration
Useful for instability detection

Seismic Sensor

Low-frequency seismic noise
Sensitive to layer boundaries and rare events

Quantum Resonance Sensor (fictional)

Extremely small fluctuations
Useful for ML models requiring additional modality

Neutrino Flux Sensor (fictional)

Sensitive to density changes
Inspired by high-energy particle flux

Spin Coherence Sensor (fictional)

Measures microscopic oscillatory behavior
Correlates weakly with phase transitions

Gravity-Wave Sensor (fictional)

Very small signal
Produces ultra-low-level noise
Adds complexity to multimodal fusion tasks

Detailed State Fields

true_temperature_C

Real physical temperature at depth, following the environment’s temperature profile.

true_pressure_GPa

Real pressure increasing with depth.

true_density_kg_m3

Density of the current layer (with ±5% random perturbation).

true_phase

Categorical:

solid
plastic
liquid

true_seismic_noise

Base seismic level.

true_vibration_level

Simulated mechanical vibration.

stability_score

A scalar between 0 and 1 representing the local stability of the environment.

rare_event_flag

Boolean indicating whether a rare catastrophic event occurred at that time step.

Use Cases

Multimodal sensor fusion
Long-horizon forecasting
Rare event detection
Model robustness evaluation
Foundation-model pretraining for time series
Reinforcement learning in synthetic physical environments
Synthetic data research
Instrumentation testbeds

Limitations

Physics is approximate and synthetic
Sensor noise is parameterized, not empirically derived
Rare events are artificially triggered
Layer boundaries are simplified

Citation

Kartoun, U. (2025). Extreme Environment Generator — Synthetic extreme-environment dataset for multimodal time-series research. DBbun LLC.