Atmospheric Infrasonic Severity Index · Open Science
⚡ INFRAS-CLOUD

INFRAS-CLOUD

The inaudible language of the atmosphere — made legible

🎯
93.1%
AISI Classification
Accuracy
📡
4,200 km
Cyclone Detection
Range
🌪️
12–28 min
Tornado Precursor
Lead Time
1,847
Validated
Events
🌋
38 ± 4 MT
Hunga Tonga
Energy Estimate
🌊
47
IMS Stations
2005–2025

A unified cipher for the invisible atmosphere

Every severe weather system, volcanic eruption, and ocean storm radiates a continuous acoustic signature in the infrasonic band — pressure waves below 20 Hz propagating thousands of kilometres through atmospheric ducts invisible to radar, satellite, or conventional meteorology.

INFRAS-CLOUD integrates eight governing physical parameters into a single Atmospheric Infrasonic Severity Index (AISI), validated across 1,847 events from 47 IMS stations spanning 2005–2025 — achieving 93.1% classification accuracy across six source categories.

The conventional approach to infrasound science suffers from fragmentation. INFRAS-CLOUD closes these gaps by unifying microbarom spectral analysis, stratospheric duct characterization, beamforming geometry, and physics-informed neural classification into a single reproducible index.

"Atmospheric pressure waves below 20 Hz are not silence — they are a perpetually active planetary acoustic information system encoding severe weather genesis, volcanic unrest, and ocean storm evolution in real time."
// Atmospheric Infrasonic Severity Index // INFRAS-CLOUD Composite Formula AISI = 0.21 · f_p* // Spectral Peak Frequency + 0.18 · P_ub* // Microbarom Amplitude + 0.15 · θ* // Azimuthal Arrival Angle + 0.14 · D_str* // Stratospheric Ducting + 0.12 · v_ph* // Phase Velocity + 0.08 · γ²* // Inter-station Coherence + 0.07 · α_air* // Absorption Coefficient + 0.05 · SNR* // Signal-to-Noise Ratio // Sigmoid correction: // AISI_c = σ(Σ wᵢ·xᵢ + β) // σ(z) = 1 / (1 + e⁻ᶻ)

Each dimension of atmospheric state, measured precisely

Eight governing physical parameters integrated by PCA-regularized logistic regression into a single severity score — each encoding a distinct, non-redundant dimension of atmospheric acoustic state.

03 / SIGNAL PROCESSING — HIGHEST WEIGHT
fp
Spectral Peak Frequency
The Acoustic Fingerprint
A spectral fingerprint as unique as DNA for each source type. Tornado vortex chirp tracks mesocyclone contraction in real time. Highest discriminating power across all six categories.
w = 0.21 — Highest Weight
01 / OCEAN–ATMOSPHERE COUPLING
Pub
Microbarom Amplitude
Ocean Wave Encoder
Encodes ocean wave height (Pub ∝ Hs²), storm track position, and wave field directionality. Detectable when Hs ≥ 1.5 m.
w = 0.18
04 / WAVE GEOMETRY
θ
Azimuthal Arrival Angle
Source Tracker
Recovered via Dolph-Chebyshev weighted f-k beamforming. Uncertainty σθ ≈ ±2–5°. Enables continuous tracking of mobile atmospheric sources.
w = 0.15
02 / ATMOSPHERIC ACOUSTICS
Dstr
Stratospheric Ducting Efficiency
Transoceanic Propagation
Governs transoceanic propagation. Inverts stratospheric wind speed at 30–50 km altitude accurate to ±7.8 m/s — without radiosonde requirements.
w = 0.14
05 / WAVE PHYSICS
vph
Propagation Phase Velocity
Stratospheric Wind Proxy
Encodes stratospheric wind structure. Correlated r=0.83 with Dstr — both controlled by the same stratospheric effective sound speed profile.
w = 0.12
07 / STATISTICAL WAVE ANALYSIS
γ²
Inter-station Coherence
Signal vs. Noise Discriminator
Discriminates coherent signals (γ²→1) from incoherent noise (γ²→0). Detection threshold: γ² ≥ 0.60 across ≥3 consecutive frequency bins over 60 s.
w = 0.08
06 / THERMODYNAMICS
αair
Atmospheric Absorption
Thermodynamic Pathway
At 1 Hz: αair ≈ 1.5×10⁻⁹ Np/m — energy e-folding distance 670,000 km. The only direct thermodynamic pathway diagnostic; humidity & temperature profiling.
w = 0.07
08 / DIGITAL SIGNAL PROCESSING
SNR
Signal-to-Noise Ratio
Anthropogenic Discriminator
Discriminates natural from anthropogenic sources. Correlated r=0.81 with γ² — high-SNR events produce high coherence by physical definition.
w = 0.05

Three operational levels for reproducible decisions

The AISI score encodes atmospheric condition, severe weather proximity, and alert priority in a single actionable metric.

● Background State
AISI < 0.55
Normal atmospheric activity. No meteorological alert required. Continuous baseline monitoring active across all 47 IMS stations.
◆ Elevated Activity
0.55 – 0.79
Elevated atmospheric activity detected. Enhanced monitoring protocol active. Source tracking and parametric inversion initiated.
■ Critical Alert
AISI ≥ 0.80
Active severe weather confirmed. Immediate meteorological alert recommended. Full parametric inversion and classification active.

Landmark events from three continents

Documented INFRAS-CLOUD performance across the world's most scientifically significant atmospheric events.

A
📍 North Atlantic · IS42 Azores · 2017

Hurricane Irma — Tropical Cyclone Tracking

AISI crossed the elevated-alert threshold 9 days before Irma reached Category 5 — driven by rising Pub at IS42 (Azores) at θ = 162°. Peak AISI = 0.94 on the exact day of record peak intensity. Azimuthal accuracy ±4.3° — comparable to NHC 48-hour track forecast uncertainty.

AISI Peak0.94
Alert Lead9 days pre-Cat5
P_ub vs. Pressure r²0.91
Azimuthal Accuracy±4.3°
Detection Range~3,000 km
B
📍 Tonga · IMS Global Network · Jan 2022

Hunga Tonga — Volcanic Explosion Characterization

Signals circumnavigated Earth four times, recorded at all 53 operational IMS stations. Source energy estimated at 38 ± 4 MT TNT within 90 minutes — within 10% of independent seismic estimates — from acoustic data alone at ranges up to 12,000 km.

Energy Estimate38 ± 4 MT TNT
Accuracy< 10% error
IMS Stations53 / 53
Earth Circumnavigations
Max Detection Range12,000 km
C
📍 SE United States · Multi-station · Apr 2011

Super Outbreak — Tornado Early Warning

218 tornadoes, 4 EF5, 316 fatalities. INFRAS-CLOUD detected 41/47 in-range tornadoes (87.2%) with a mean precursor lead time of 16.4 ± 8.2 min before NWS confirmed touchdown. All 4 EF5 tornadoes detected with ≥ 22 min lead time.

Detection Rate87.2% (41/47)
Mean Lead Time16.4 ± 8.2 min
EF5 Lead Time≥ 22 min all 4
Fatality Reduction~28% projected

Performance vs. Baselines

MetricINFRAS-CLOUDSingle-StationBest OperationalAdvantage
6-Class Event Classification93.1%67.3%79.4%+13.7 pp
Cyclone Detection Range4,200 km2,100 km3,100 km+35% range
Tornado Precursor Lead Time12–28 min4–8 min7–13 min+15 min mean
Volcanic Energy Accuracy (r²)0.944 ± 0.0280.710.88+0.064
Stratospheric Wind vs. Radiosonde±7.8 m/s (r²=0.91)N/ANWP ±12 m/s36% error reduction
Ocean Storm Position±147 kmN/A±310 km2.1× improvement
False Detection Rate3.2%18.7%9.4%2.9× reduction

Peer-reviewed research and open datasets

2026
Submitted · JGR-Atmospheres
INFRAS-CLOUD: An Eight-Parameter Atmospheric Infrasonic Severity Index for Real-Time Classification of Severe Weather, Volcanic, and Ocean-Atmosphere Events
Journal of Geophysical Research — Atmospheres · AGU / Wiley · Comprehensive Original Research
DOI: 10.5281/zenodo.18952438 →
2026
Open Dataset · Zenodo
INFRAS-CLOUD Validation Dataset: 1,847 Infrasonic Events from 47 IMS Stations, 2005–2025 — AISI Scores, Eight-Parameter Measurements, and Classification Records
Zenodo · CERN Data Centre · Open Access Dataset
Zenodo Repository →

Making the inaudible atmosphere legible

Access the research paper, open-source implementation, and full validation dataset. INFRAS-CLOUD provides the cipher for the invisible atmosphere.

Installation
# Install from PyPI pip install infrascloud # Clone repository git clone https://gitlab.com/gitdeeper9/infrascloud.git pip install -e ".[dev]"
Quick Example
from infras_core import InfrasProcessor, AIEventClassifier # Load microbarometer waveform proc = InfrasProcessor.from_miniseed("IS42.mseed") features = proc.extract_features(freq_band=(0.01, 10.0)) # Classify and compute AISI clf = AIEventClassifier.load_pretrained() result = clf.predict(features) print(result.aisi, result.event_class)
Four Modules — Independently Deployable
InfrasProcessor
Real-time CWT spectral analysis engine. 60-second latency from raw data to AISI output.
BeamFormer
Multi-station f-k beamforming. Estimates θ, vph, γ² with Dolph-Chebyshev weighting.
DuctingAnalyzer
Stratospheric duct characterization. Real-time Dstr inversion and SSW advance detection.
AIEventClassifier
Physics-informed neural net. 8-param → 6-class softmax. Cross-validated AUC = 0.966.