M33 / Learn

Field Manual

A technical reference for satellite data, sensor physics, and the provenance architecture behind defensible earth observation.

Entries

12 articles

OG-001

Observational Grammar

A constitution for evidence derived from sensor physics, independent of external incentives

Observational Grammar (OG) is the idea that sensors — satellites, radar, spectrometers — impose a physical grammar on what can be observed. That grammar is not optional. Resolution, revisit rate, spectral sensitivity, geometric distortion — these are constraints that shape the evidence before any analyst, model, or decision-maker touches it. Understanding OG means understanding what your data is actually capable of saying.

→ COM-001

Latency: From Orbit to Application

The satellites are fast enough — the ground segment is where the time goes

Latency in satellite systems is the total time between a sensor observing something and an analyst or system acting on that observation. The satellite is rarely the bottleneck. Most latency accumulates in the ground segment: contact windows, downlink queues, processing pipelines, format conversions, and delivery infrastructure. Understanding where time goes is prerequisite to designing systems that can meet operational latency requirements.

→ SEC-001

Data Provenance

Hard to argue that knowing where your data came from is simply good hygiene!

Data provenance is the complete, verifiable record of where a piece of data came from, what transformations it has undergone, and who or what touched it at each stage. In earth observation, provenance is not a metadata nicety — it is the mechanism by which an observation can be defended as evidence rather than asserted as fact.

→ DAT-004

Analysis-Ready Data

The concept that satellite imagery should arrive ready for science, not ready for more preprocessing

Analysis-Ready Data (ARD) is satellite imagery that has been processed to a standard allowing direct use in analysis without additional preprocessing. ARD specifications define what corrections must be applied — geometric, radiometric, atmospheric — and what metadata must be present. The goal is interoperability: data from different sensors, dates, and providers that can be stacked and compared without bespoke preprocessing for each combination.

→ SEC-003

Trusted Execution Environments

When provenance needs to be a proof, not a claim, the answer is in silicon

A Trusted Execution Environment (TEE) is a hardware-enforced isolated region within a processor where code executes in a way that is cryptographically verifiable. What runs inside a TEE cannot be observed or tampered with by the operating system, hypervisor, or other software — even if those systems are compromised. TEEs allow geospatial processing pipelines to generate signed receipts attesting that a specific input produced a specific output via a specific, unmodified algorithm.

→ SEC-002

Chain of Custody in Multi-Sensor Fusion

When provenance stops being a chain and becomes a graph

When multiple sensor datasets are combined — SAR with optical, optical with terrain models, radar with thermal — the provenance of the resulting product is no longer a chain but a graph. Each input has its own lineage, its own uncertainty, its own processing history. Fusing them produces a new artifact whose trustworthiness depends on how well the lineage of every contributing input is preserved and propagated.

→ SEC-004

Space Cybersecurity: The Attack Surface Above Us

Satellites underpin global infrastructure yet remain among the most exposed digital targets

Space systems are among the most critical and least defended digital infrastructure on Earth. Satellites that underpin GPS, communications, weather forecasting, and earth observation present an attack surface that spans ground stations, uplink/downlink communications, onboard software, and supply chains — often with minimal security controls and long patch cycles. Understanding this attack surface is prerequisite to designing resilient systems.

→ SEN-005

SAR Fundamentals

How synthetic aperture radar sees through clouds, in darkness, and reveals what optical sensors cannot

Synthetic Aperture Radar (SAR) is a radar imaging system carried on aircraft or satellites that creates high-resolution images of the Earth's surface by emitting microwave pulses and measuring what bounces back. Unlike optical sensors that rely on sunlight, SAR generates its own illumination — so it works at night. Unlike optical sensors that are blocked by clouds, SAR's microwave frequencies pass through clouds, rain, and smoke.

→ SAT-002

Sentinel-2

ESA's multispectral workhorse: 13 bands, 10-meter resolution, and the backbone of global land monitoring

Sentinel-2 is a pair of optical satellites operated by the European Space Agency (ESA) as part of the Copernicus Earth observation program. Together, Sentinel-2A and Sentinel-2B provide 13 spectral bands, a 10-meter spatial resolution in key bands, and a 5-day revisit time at the equator. Free, open-access, and systematically archived since 2015, Sentinel-2 has become the default optical dataset for land cover mapping, agriculture monitoring, disaster response, and change detection.

→ PHI-003

Epistemic Architecture

How the structure of information systems determines what can be known — and what remains invisible

Epistemic architecture is the idea that how you structure a system for handling information determines not just how efficiently that information flows, but what kinds of truths the system can produce. The architecture shapes the epistemology. A system without uncertainty quantification cannot produce confident knowledge — it can only produce assertions. A system without provenance cannot defend its outputs as evidence — it can only claim them.

→ PHI-004

Information Networks & Truth

Why the architecture of information systems determines whether they produce understanding or delusion

The structure of an information network — not just the data flowing through it — determines whether that network produces truth or delusion. A network with self-correction mechanisms, error detection, and distributed verification tends toward truth. A network optimized for speed, engagement, or institutional convenience tends toward whatever narrative serves its operators.

→ EO-001

Why Geospatial Intelligence Resists General-Purpose AI

The most effective geospatial AI systems are not the largest or most general — they encode domain knowledge into their structure

General-purpose machine learning treats location as just another feature in a tabular dataset. But geospatial intelligence operates under physical constraints that are not encoded in training data: sensor physics, orbital mechanics, atmospheric conditions, geometric distortion, spectral limitations. Models that ignore these constraints produce outputs that are statistically plausible but physically incoherent.

→

Learning Paths

Curated sequences

By time

15 minutes

15 min

A minimal route: evidence, latency, provenance.

60 minutes

A practical model of observation → trust → action.

Half day

Design, critique, and defend an EO system under real constraints.

By role

EO / GEOINT Analyst

Interpretations you can defend: measurement vs inference with uncertainty intact.

Data Engineer (EO pipelines)

1h 15m

Build pipelines that respect latency budgets and preserve lineage.

Decision-maker / Strategy

Avoid buying a story that cannot survive physics, operations, or adversaries.

Security / Provenance Engineer

1h 5m

Make transformations verifiable and products defensible.

By mission

Disaster response / NRT mapping

55 min

Time-to-answer without speed theater.

Audit-grade intelligence products

1h 10m

Defensibility, reproducibility, verifiability.

Training data for spatial AI

Training signals that match reality — not just benchmarks.