Signal Methodology Document

SatClimate Signal Methodology

Institutional-grade documentation of data sources, signal extraction pipeline, validation methodology, limitations, and intended use. Version 2.0 — May 2026.

What's in this document

14 data sources listed with frequency, latency, and direct vs. modeled classification
Data Latency Summary table — consolidated end-to-end signal delay per source
Third-party dependency risk disclosure — ESA, NASA, NOAA, ECMWF reliance
6-stage signal extraction pipeline — ingestion, QC, spatial aggregation, fusion, normalisation
Fusion weight circular-logic clarification — price data excluded from real-time generation
Activity Index definition — interpretation bands, cross-region constraints
Signal output schema — all fields including data_coverage and confidence_score with explicit definitions
No look-ahead bias statement — all signals built on data available at observed_at only
Survivorship bias disclosure — asset dataset limitations in historical backfill
Signal stability & revision policy — 24-hour stability window, retroactive revision criteria
Validation methodology — event definition, sample period 2018–2024, scope & limitations
Signal uncertainty & error bounds — qualitative and quantitative uncertainty framing
Regulatory classification — alternative data, not a regulated financial instrument
HFT / sub-daily use prohibition — not designed for high-frequency execution
Model governance & change control — versioning, reproducibility, API version pinning
Expanded limitations — satellite, AIS, model, processing, and usage constraints
Legal disclaimer and due diligence contact details

20-parameter scientific quality framework

Every SatClimate-powered output carries a 20-parameter scientific quality profile. Absent metrics are reported explicitly — never fabricated. This framework is shared with all AOI capabilities.

Measurement integrity

accuracy

Validation accuracy against ground truth. NOT_CALCULATED — no independent validation pipeline.

uncertainty

Quantified uncertainty range from the satellite sensor and processing chain.

confidence

Model self-reported confidence. NOT_VALIDATED — not a scientific accuracy claim.

observed_vs_derived

Direct observation (e.g., FIRMS fire detection) vs derived index (e.g., NDVI from Sentinel-2).

completeness

Fraction of the 20 parameters available for this specific output.

Resolution & freshness (per satellite source)

spatial_resolution

Sensor-dependent: 10m (Sentinel-2), 375m (VIIRS), 1km (ERA5), 7km (Sentinel-5P).

temporal_resolution

Revisit cadence: 5 days (Sentinel-2), daily (Sentinel-3, ERA5), 2× daily (VIIRS).

temporal_coverage

Historical archive depth (e.g., Sentinel-2 from 2015, ERA5 from 1940).

data_freshness

Age of the underlying satellite observation at signal generation time.

latency

End-to-end delay from satellite overpass to signal availability.

applicable_geographic_domain_scope

Geographic validity — global, regional, or per-facility.

Source & methodology

source_dataset

Specific dataset: Sentinel-2 L2A, Sentinel-5P L2, NASA FIRMS VIIRS, ERA5, OpenAQ, GFW.

model_methodology

Algorithm version used to derive the signal (e.g., NDVI formula, activity index weighting).

provenance

Full chain: satellite → processing level → ingestion pipeline → signal output.

quality_control_status

QC flags from the source (cloud mask, data quality flag, sensor anomaly).

missing_data_behavior

Explicit NO_DATA when source returns empty — never interpolated or fabricated.

Validation & limitations

validation_method

Retrospective event validation methodology. NOT_VALIDATED for real-time outputs.

validation_period

Historical validation sample period (2018–2024). NOT_AVAILABLE for real-time.

validation_population_area

Geographic scope of validation events. NOT_AVAILABLE for real-time.

limitations

Satellite, model, processing, and usage constraints — stated explicitly per signal.

Seven semantic data modes

DIRECTLY_OBSERVED

Direct measurement from satellite or API (e.g., FIRMS fire detection, VIIRS night lights).

DERIVED_FROM_SATELLITE

Computed from satellite observations (e.g., NDVI from Sentinel-2, flood extent from Sentinel-1).

MODELLED_FROM_SATELLITE

Model inference based on satellite inputs (e.g., risk scores, activity index).

HISTORICAL

Archive data — not current observation (e.g., historical deforestation baseline).

LLM_KNOWLEDGE

LLM reasoning without physical observation (e.g., advisory capabilities).

NO_DATA

Source attempted but returned no observation — never fabricated.

NOT_APPLICABLE

Capability does not require physical data (e.g., confidence assessment).

This document describes a retrospective signal validation methodology. Past signal accuracy does not guarantee future performance. No independent third-party audit has been completed as of v1.0. The platform does not provide investment advice.

SatClimate_Signal_Methodology_v2.0.pdf · Carbonaa SA · Geneva, Switzerland

For Institutional Due Diligence

We understand that hedge funds, commodity trading desks, and facility managers require transparent methodology documentation before incorporating any third-party data signal into their workflow. This document is provided as part of our standard due diligence plan.

For questions or to arrange a technical call with our data team, contact our team

base44
Edit with Base44