Satellite Remote Sensing Failures: Technical Background

This document explains the real-world satellite sensor and software failures simulated by the satellite malfunction operations. Understanding these failure modes provides context for using the operations effectively and appreciating their basis in actual remote sensing challenges.

Overview

Satellite remote sensing systems face unique environmental and operational challenges that lead to characteristic image artifacts. These failures fall into three categories:

Sensor Hardware Failures - Physical detector issues (cosmic rays, calibration drift)
Software/Transmission Failures - Data corruption during processing and downlink
Mechanical Hardware Failures - Physical component failures (scan mechanisms)

Sensor Hardware Failures

Salt & Pepper Noise: Cosmic Ray Hits

Physical Mechanism

In the space environment, high-energy particles (cosmic rays, solar protons) constantly bombard satellite electronics. When these particles strike a sensor’s photodetector array:

Direct ionization - Particle passes through silicon, creating electron-hole pairs
Charge collection - These charges are misinterpreted as signal
Single Event Effect (SEE) - One particle hit → one corrupted pixel

Characteristics

Random distribution - No spatial correlation (cosmic rays are isotropic)
Bimodal response - Pixel reads either maximum (255) or minimum (0) value
Transient - Only affects single frame, next frame is clean
Frequency - Depends on orbital altitude and solar activity
- Low Earth Orbit (LEO): ~0.01-0.1% pixels per frame during quiet periods
- Solar storms: Can increase by 10-100× for hours or days
- South Atlantic Anomaly: 5-10× higher rates

Real-World Examples

Hubble Space Telescope: Visible cosmic ray hits in long exposures
Mars rovers: Random bright pixels in Mastcam images
Landsat: Occasional white/black pixel speckles in thermal bands

Why Both Salt (White) and Pepper (Black)?

The polarity depends on:

Particle energy and angle
Depth of penetration into detector
Whether charge accumulates or depletes Result: approximately equal probability of saturation high or low

Corduroy Striping: Detector Calibration Drift

Physical Mechanism

Multispectral satellites use linear detector arrays (push-broom) or 2D arrays (whisk-broom) to build up images by scanning:

Manufacturing variations - Each detector element has slightly different sensitivity
Initial calibration - Ground calibration compensates for these differences
Orbital degradation - Over time, detectors age differently:
- Radiation damage accumulates non-uniformly
- Thermal cycling causes mechanical stress
- Contamination affects some detectors more than others
Calibration drift - Original compensation no longer accurate

Characteristics

Stripe pattern - Repeating lines perpendicular to scan direction
- Push-broom: Vertical stripes (satellite moves forward, scans across-track)
- Whisk-broom: Horizontal or diagonal stripes
Persistent - Same pattern appears in every image
Brightness variation - Typically ±5-20% brightness
Worsens over time - Years 1-2: subtle, Years 5+: significant

Mathematical Model

If detector i has gain drift Δgᵢ:

I_observed(i) = I_true × (1 + Δgᵢ)

Where Δgᵢ varies slowly over detector array (correlated with manufacturing lot, position in array, thermal gradients).

Real-World Examples

Landsat 4/5 Thematic Mapper - Visible vertical striping in bands 5 & 7 after 10+ years
MODIS Terra - “Bowtie” striping visible in some bands, especially at scan edges
Aqua MODIS - Detector 6 degradation created visible stripe in Band 6

Correction Methods

Ground processing applies:

Periodic re-calibration - Update gain/offset coefficients
Destriping algorithms - Histogram matching or Fourier filtering
Replacement with backup detectors - If available

Software/Transmission Failures

Band Swap: Packet Header Corruption

Physical Mechanism

Satellite downlink transmits image data as packets with metadata headers:

[Packet Header: Band ID, Position] [Pixel Data: Raw Values]

Bit flips in packet headers (from cosmic rays, transmission noise) can corrupt metadata while leaving data intact:

Header corruption - Band identifier bits flip (e.g., Band 1 → Band 3)
Data received correctly - Pixel values transmitted without error
Misinterpretation - Ground station assigns data to wrong spectral band
Result - Red channel data interpreted as Blue, etc.

Characteristics

Rectangular regions - Size corresponds to packet length
Sharp boundaries - Instant transition between correct and swapped regions
Spatial structure preserved - Shapes and textures intact, only colors wrong
Deterministic per tile - Same swap pattern within each corrupted packet

Why Rectangular Tiles?

Satellite data is packetized in fixed-size tiles for efficient transmission:

Typical size: 256×256 to 512×512 pixels
Each tile has independent header
Header corruption affects entire tile uniformly

Real-World Examples

NOAA AVHRR - Occasional band swaps in real-time transmission
MSG SEVIRI - Rare channel mis-identification in rapid scan mode
Suomi-NPP VIIRS - Documented cases of incorrect band labeling in early mission

Probability

Given cosmic ray SEU rate ~10⁻¹⁰ bits/second:

Packet header: ~100 bits
Data downlink rate: 100 Mbps
Expected header corruption: ~1 packet per 10⁵ transmitted

Buffer Corruption: Single Event Upsets in Memory

Physical Mechanism

On-board image buffers (DRAM/SRAM) hold raw image data before compression and transmission. Cosmic ray hits can flip bits in memory:

Particle strike - High-energy proton/heavy ion hits memory cell
Charge deposition - Creates electron-hole pairs
Bit flip - If charge exceeds threshold, bit flips (0→1 or 1→0)
Data corruption - Pixel values corrupted before encoding

Types of Corruption

XOR Corruption (Most common):

Original pixel: 10110011 (179)
XOR mask:       00001000 (bit 3 flipped)
Corrupted:      10111011 (187)

Inversion (Less common):

Original pixel: 10110011 (179)
Inverted:       01001100 (76)

Pointer Corruption (Multi-planar buffers):

RGB stored as: R[1024×1024], G[1024×1024], B[1024×1024]
If pointer corrupted: R data read as B, G data read as R, etc.

Characteristics

Localized regions - Typically affect memory pages (4-64KB blocks)
Pixel-level corruption - Individual values altered
Bitwise patterns - Systematic bit flips create specific visual signatures
Varies by memory type:
- DRAM: Large regions (row/column corruption)
- SRAM: Small regions (individual cells)

Real-World Examples

Mars rover cameras - Occasional “glitch blocks” in Navcam/Hazcam images
Cassini ISS - Random corruption in raw image buffers
Hubble WFC3 - Memory bit flips in electronics

SEU Rates by Orbit

Low Earth Orbit (400-800 km): 10⁻⁶ - 10⁻⁵ upsets/bit/day
Geostationary (36,000 km): 10⁻⁷ - 10⁻⁶ upsets/bit/day
During solar storms: 100× - 1000× higher

Compression Artifacts: On-board Encoder Failures

Physical Mechanism

Satellites use on-board JPEG/JPEG2000 compression to reduce data volume. Encoder failures create localized artifacts:

Rate Control Failures:

Target bitrate - Encoder must fit scene into fixed bandwidth
Complexity variation - Some image regions harder to compress
Buffer overflow - When complex region exceeds budget
Emergency quality reduction - Encoder drops quality mid-scene

Hardware Malfunctions:

DCT engine errors - Discrete Cosine Transform unit malfunction
Quantization table corruption - Encoding parameters corrupted
Thermal throttling - Overheating reduces clock speed → reduced quality

Characteristics

Tile-based - JPEG compresses in 8×8 or 16×16 blocks
Variable quality - Different regions at different compression levels
8×8 blocking - Visible DCT block boundaries
Color bleeding - Chroma subsampling artifacts

Real-World Examples

Landsat 7 ETM+ - Early mission compression artifacts from encoder bugs
Sentinel-2 MSI - Rare JPEG2000 artifacts in L1C products
Commercial satellites - Quality variations in rapid-revisit systems

Compression Ratios

Target: 10:1 to 20:1 (high quality)
Failure mode: 50:1 to 100:1 (severe artifacts)

Mechanical Hardware Failures

SLC-Off: Landsat 7 Scan Line Corrector Failure

Physical Mechanism

The Scan Line Corrector (SLC) was a mechanical mirror system on Landsat 7’s Enhanced Thematic Mapper Plus (ETM+):

Normal Operation (Before May 31, 2003):

Whisk-broom scanner - Oscillating mirror scans perpendicular to flight
Forward motion - Satellite moves ~7 km/s
SLC compensation - Second mirror compensates for motion
Result - Straight, continuous scan lines

After Failure (May 31, 2003 onwards):

SLC mirror stuck - Compensation mechanism failed
Uncompensated forward motion - Satellite movement creates diagonal scan displacement
Zig-zag gap pattern - Diagonal gaps alternate direction (left/right) on successive scans
Data gaps - Missed regions between scan lines form diagonal wedges
Wedge pattern - Gaps widen with distance from nadir (subsatellite point)

Geometry

The gap width as a function of position:

gap_width(y) = gap_max × |y - y_center| / (y_max / 2)

Where:

y = row position in image
y_center = nadir (center) row
gap_max ≈ 22% of scene width (historical value)

Characteristics

Diagonal gaps - Uncompensated forward motion creates shallow diagonal wedges across image
Alternating directions - Each scan line’s gap tilts opposite direction (left/right zig-zag)
Symmetric wedges - Gaps widen from center to edges
No gaps at center - Nadir point has complete coverage
Maximum at edges - ~22% data loss at top/bottom
Repeating pattern - Gaps every ~14 scan lines (ETM+ geometry)
Persistent - Same pattern in every scene since May 31, 2003

Impact

14% data loss - Average across entire scene
22% maximum loss - At scene edges
Continued operation - Satellite still providing valuable data
Gap-filling algorithms - Post-processing interpolates missing data

Real-World Timeline

Launch: April 15, 1999
SLC failure: May 31, 2003 (4 years into mission)
Continued operation: Still operational as of 2024 (21+ years total)
Legacy: 20+ years of SLC-Off data in archives

Scientific Impact

Despite gaps, Landsat 7 ETM+ SLC-Off data remains valuable:

Long-term continuity with Landsat 4/5
Higher resolution than Landsat 8/9 for some applications
Gap-filling algorithms make data usable for many purposes

Failure Mode Interactions

Real satellite systems experience multiple simultaneous failures:

Radiation Environment Correlations

Solar particle events create correlated failures:

Solar Storm Event
├── Increased cosmic ray flux → salt_pepper
├── Memory upsets → buffer_corruption
├── Transmission errors → band_swap
└── Detector damage → corduroy (long-term)

Long-duration missions accumulate failures:

Mission Years 1-3: Minimal degradation
├── salt_pepper: 0.0001 (baseline cosmic rays)
└── corduroy: strength 0.1 (minor drift)

Mission Years 5-10: Moderate degradation
├── salt_pepper: 0.0003 (detector damage)
├── corduroy: strength 0.4 (calibration drift)
└── compression_artifact: encoder wear

Mission Years 15+: Severe degradation
├── salt_pepper: 0.001 (accumulated damage)
├── corduroy: strength 0.8 (severe drift)
├── compression_artifact: frequent failures
└── Potential mechanical failures (SLC-Off scenario)

Cascading Failures

One failure can trigger others:

Solar Particle Event
→ Memory corruption (buffer_corruption)
→ Corrupts compression tables
→ Encoder failure (compression_artifact)
→ Reduced data quality

Orbital Considerations

Failure rates vary by orbit type:

Low Earth Orbit (LEO)

Altitude: 400-800 km Examples: Landsat, MODIS Terra/Aqua, Sentinel-2

Radiation Environment:

Moderate cosmic ray flux
South Atlantic Anomaly (SAA) passages: 5-10× higher rates
Solar particle events: Significant impact

Typical Failure Rates:

salt_pepper: 0.0001-0.001% per frame
buffer_corruption: ~1 event per 10⁶ frames
corduroy: Develops over 5-15 years

Sun-Synchronous Orbit (SSO)

Altitude: 600-800 km Examples: Most Earth observation satellites

Characteristics:

Fixed local time (e.g., always 10:30 AM)
Consistent solar illumination
SAA passages every ~90 minutes

Failure Patterns:

Regular SAA-induced spikes in salt_pepper
Thermal cycling → accelerated detector drift

Geostationary Orbit (GEO)

Altitude: 36,000 km Examples: GOES, MSG, Himawari

Radiation Environment:

Lower overall flux (outside Van Allen belts)
Greater solar wind exposure
Higher energy particles

Typical Failure Rates:

Lower salt_pepper rate than LEO
Higher energy SEU events (buffer_corruption)

Mitigation Strategies

Understanding failures informs both simulation realism and appreciation of operational challenges:

Radiation Hardening

Detector Design:

Shielding (tungsten, tantalum)
Redundant detectors
Error correction codes (ECC)

Memory Protection:

EDAC (Error Detection And Correction)
Triple Modular Redundancy (TMR)
Scrubbing (periodic memory refresh)

Calibration Management

On-orbit Calibration:

Solar calibrator targets
Lunar observations (stable reference)
Cross-calibration with other satellites

Periodic Updates:

Quarterly coefficient updates
Detector health monitoring
Drift prediction models

Data Quality Flags

Operational systems flag suspect data:

Pixel quality masks
SAA passage indicators
Calibration confidence levels

Using This Information

For Realistic Simulation

Match parameters to mission phase:

Early Mission (Years 0-3):

- operation: "salt_pepper"
  params:
    amount: 0.0001  # Baseline
- operation: "corduroy"
  params:
    strength: 0.1   # Minimal drift

Mid-Mission (Years 5-10):

- operation: "salt_pepper"
  params:
    amount: 0.0005  # Some detector damage
- operation: "corduroy"
  params:
    strength: 0.4   # Noticeable drift

Late Mission (Years 15+):

- operation: "salt_pepper"
  params:
    amount: 0.002   # Significant damage
- operation: "corduroy"
  params:
    strength: 0.7   # Severe drift

For Aesthetic Goals

Use understanding to create intentional effects:

Documentary realism: Match historical parameters (SLC-Off: 0.22)
Glitch art: Exaggerate specific failures (band_swap tile_count: 50)
Scientific visualization: Isolated failures to understand individual effects

References and Further Reading

Landsat 7 SLC-Off:

USGS: “Landsat 7 SLC-off Products: Background and Information” (2003)
Storey et al., “Landsat 7 Scan Line Corrector-Off Gap-Filled Product Development” (2005)

Radiation Effects:

Barth et al., “Space, Atmospheric, and Terrestrial Radiation Environments” (2003)
Holmes-Siedle & Adams, “Handbook of Radiation Effects” (2002)

Detector Calibration:

Xiong et al., “MODIS On-Orbit Calibration and Characterization” (2006)
Chander et al., “Overview of Landsat Calibration History” (2009)

Single Event Effects:

Normand, “Single Event Effects in Avionics” (1996)
Petersen, “Single Event Effects in Aerospace” (2011)

Conclusion

The six satellite malfunction operations simulate well-documented failure modes from decades of remote sensing operations. Understanding the physical mechanisms, orbital environments, and historical examples provides:

Realistic parameter selection - Match real-world failure rates
Intentional aesthetic choices - Know what you’re simulating
Scientific appreciation - Recognize challenges of space-based imaging

These operations bridge artistic expression and technical accuracy, allowing creation of both realistic simulations and intentional glitch aesthetics grounded in actual satellite physics.

Next Steps: Apply this understanding in the Satellite Malfunctions Tutorial and explore parameter options in the Satellite Operations Reference.

Satellite Remote Sensing Failures: Technical Background

Overview

Sensor Hardware Failures

Salt & Pepper Noise: Cosmic Ray Hits

Corduroy Striping: Detector Calibration Drift

Software/Transmission Failures

Band Swap: Packet Header Corruption

Buffer Corruption: Single Event Upsets in Memory

Compression Artifacts: On-board Encoder Failures

Mechanical Hardware Failures

SLC-Off: Landsat 7 Scan Line Corrector Failure

Failure Mode Interactions

Radiation Environment Correlations

Age-Related Degradation

Cascading Failures

Orbital Considerations

Low Earth Orbit (LEO)

Sun-Synchronous Orbit (SSO)

Geostationary Orbit (GEO)

Mitigation Strategies

Radiation Hardening

Calibration Management

Data Quality Flags

Using This Information

For Realistic Simulation

For Aesthetic Goals

References and Further Reading

Conclusion