How many satellites are needed for real-time Earth observation?

Between 204 and 386 LEO satellites depending on resolution: 204 for 0.3m, 254 for 0.5m, or 386 for 1.0m with hourly revisit capability.

What is the estimated cost of a global Earth observation constellation?

The 5-year total cost ranges from about $1.7 billion for moderate resolution to about $18.3 billion for the high-resolution configuration using Falcon 9 rideshare launches.

What AI processing is embedded in each satellite?

Each satellite includes a radiation-hardened Xilinx Virtex-5QV FPGA for image preprocessing and an NVIDIA Jetson Thor AI processor for real-time deep learning inference.

Real-Time Earth Observation Satellite Constellation with Embedded AI

System Architecture

End-to-end orbital data pipeline

From orbital image capture to real-time AI-powered emergency alerting, each satellite acts as a compact sensing and inference node.

IM

Image Acquisition

High-resolution telephoto payload captures sub-meter imagery from a 500 km LEO orbit.

GSD 0.3-1m

FP

FPGA Preprocessing

Xilinx Virtex-5QV handles calibration, denoising, compression, and data reduction before AI inference.

10W rad-hard

AI

AI Inference

NVIDIA Jetson Thor performs on-board deep learning for anomaly detection and object classification.

275 TOPS

KA

Ka-Band Downlink

High-throughput transmitter sends urgent detections and selected imagery to the ground segment.

1.2 Gbps

GS

Ground Stations

A global network receives imagery, confirms events, and routes data into response workflows.

Global network

AL

AI Alert Center

Real-time emergency alerts are dispatched for detected fires, floods, infrastructure damage, and unusual activity.

<60 min latency

Orbital Parameters

500 kmAltitude

6,871 kmOrbital radius

94.5 minPeriod

15.24Orbits/day

7.06 km/sGround speed

Walker ΔConfiguration

Camera Selection

Imaging payload comparison

Five representative Earth-observation camera platforms define the resolution, cost, mass, swath, and power trade space.

WorldView Legion

Maxar

0.30 m

Highest commercial resolution with 15 cm pan-sharpened imagery and high revisit capability.

FOV1.15°

Swath9 km

Mass750 kg

Power~150 W

Price$80M

Recommended

SkySat

Planet Labs

0.50 m

Strong resolution-to-cost ratio, video capture capability, and a compact 117 kg satellite class.

FOV0.92°

Swath8 km

Mass117 kg

Power~50 W

Price$2-5M

NewSat Mk V

Satellogic

1.0 m

Most affordable option, with multispectral and hyperspectral capability for broad monitoring missions.

FOV0.60°

Swath5.3 km

Mass46 kg

Power~15 W

Price$1.5M

Gen-3

BlackSky

0.35 m

Latest-generation imaging platform with AI-integrated delivery and near-real-time operations.

FOV~1.0°

Swath~9 km

Mass~100 kg

Power~60 W

Price$5-10M

NAOMI

Airbus

1.5 m

Heritage SPOT-class instrument with a wide 60 km swath for large-area coverage.

FOV~2.0°

Swath~60 km

Mass60 kg inst.

Power~40 W

BandsPAN + 4 MS

Selection Lens

GSD drives intelligence value, but narrow swath and high satellite cost quickly dominate constellation size and capital needs. The study therefore treats moderate, medium, and high resolution as separate deployment phases.

GSDResolution

FOVField of view

MS/HSSpectral bands

Embedded Processing

On-board AI pipeline

A heterogeneous processor stack combines a radiation-hardened FPGA with a shielded commercial AI accelerator.

Xilinx Virtex-5QV

Radiation-hardened FPGA for real-time image preprocessing, compression, and data reduction before AI inference.

~10 WPower

65 nmSRAM

TMRHardening

NVIDIA Jetson Thor

Supercomputer-class AI processor enabling deep-learning inference for detection, classification, and alert prioritization.

275 TOPSINT8

40-130 WConfig.

ShieldedRadiation

Processing Pipeline

01Image CaptureRaw telescope imagery

02FPGA PipelineDenoise, calibrate, compress

03AI InferenceDetect and classify

04Alert GenerationPriority events flagged

05Ka-Band TXHigh-priority downlink

Constellation Design

Satellites versus resolution

Walker Delta sizing shows the central tradeoff: higher resolution improves detail but raises cost and still requires hundreds of satellites for hourly global revisit.

Hourly Revisit Satellites

High Res 0.3m204

Medium 0.5m254

Moderate 1.0m386

Daily Revisit Satellites

High Res129

Medium Res161

Moderate245

FOV

FOV = 2 x arctan(swath / (2 x h))

Swath

swath = 2 x h x tan(FOV/2)

Planes

N_p = ceil(C_earth / (swath x 0.9))

Per Plane

N_s = ceil(C_orbit / (swath x 0.9))

Power Budget

Energy and solar sizing

NVIDIA Thor dominates the power budget on smaller satellites, making power and thermal design central to feasibility.

Power Consumption by Component

High Res438 W

Medium Res306 W

Moderate258 W

100 WThor baseline

65 WKa-band comms

20%Margin

Solar and Battery Sizing

High Res area2.02 m2

Medium area1.41 m2

Moderate area1.19 m2

30%GaAs cells

35%Eclipse factor

15%5-year degradation

Cost Analysis

Financial assessment

Five-year constellation cost ranges from a moderate-resolution proof-of-concept to a premium high-resolution system.

$1.69B

Most affordable moderate-resolution Falcon 9 scenario.

$2.21B

Best-value medium-resolution SkySat-class scenario.

$18.3B

Premium high-resolution WorldView-class scenario.

~4.5x

Launch cost advantage of Falcon 9 rideshare versus Rocket Lab.

Cost Element	High Res	Medium Res	Moderate
Satellites	204	254	386
Mass per Satellite	760 kg	117 kg	46 kg
Unit Cost	$80M	$4M	$1.5M
Manufacturing	$16,320M	$1,016M	$579M
Launch, Falcon 9	$853M	$164M	$98M
Ground Segment	$100M	$75M	$50M
Operations, 5 years	$1,020M	$952M	$965M
Total, Falcon 9	$18,293M	$2,207M	$1,692M
Total, Rocket Lab	$21,318M	$2,788M	$2,040M

Scenario Comparison

Side-by-side analysis

The recommended route starts with moderate resolution, then upgrades toward medium and high resolution as operational demand and capital availability are proven.

Metric	High Res	Medium Res	Moderate
GSD	0.30 m	0.50 m	1.0 m
Swath Width	10.0 km	8.0 km	5.3 km
Satellites, hourly	204	254	386
Satellite Mass	760 kg	117 kg	46 kg
Power per Satellite	438 W	306 W	258 W
Solar Panel Area	2.02 m2	1.41 m2	1.19 m2
Unit Cost	$80M	$4M	$1.5M
5-year Total, F9	$18.3B	$2.2B	$1.7B
Camera Type	WorldView	SkySat	Satellogic
Spectral Bands	8 MS + PAN	4 MS + PAN	4 MS + 29 HS

Recommendation & Roadmap

Phased deployment strategy

A three-phase path limits early risk while preserving a route to high-resolution real-time global monitoring.

Years 1-3Phase 1

Moderate Resolution Deployment

386 Satellogic-class satellites validate hourly global revisit, hyperspectral value, and operational demand at about $1.7B.

1.0 m GSD with hyperspectral capability
On-board AI with NVIDIA Jetson Thor
Falcon 9 rideshare launch model
Prove alert workflows and demand

Years 3-5Phase 2

Medium Resolution Upgrade

Replace aging satellites with SkySat-class units for 0.5 m GSD and improved data quality while keeping total cost near $2.2B.

254 satellites for hourly coverage
Enhanced models from Phase 1 data
Expanded ground station network
Better resolution-to-cost balance

Years 5+Phase 3

High Resolution Constellation

WorldView-class 0.3 m imaging offers premium intelligence value, but requires much larger capital and stronger launch/operations support.

204 satellites for hourly coverage
Optical laser downlinks as upgrade path
Highest commercial resolution tier
About $18.3B over five years

Resolution Coverage Tradeoff

Higher resolution narrows swath and quickly raises constellation count for the same revisit rate.

Thor Dominates Power

The AI processor is the primary consumer on smaller satellites and shapes thermal design.

Phased Deployment

Moderate resolution validates the concept before committing to high-resolution capital needs.

Launch Choice Matters

Falcon 9 rideshare saves billions compared with smaller dedicated launch options at scale.