Skip to main content
U.S. flag

An official website of the United States government

Impulse Noise

Landsat data are systematic, geometric, radiometric, and terrain corrected to provide the highest quality data to the user communities. Occasionally, anomalies occur and artifacts are discovered that require research and monitoring.

Return to Landsat Known Issues Overview

 

Impulse Noise (IN) is a general term for single-pixel bright or dark spots that are not authentic imagery. This artifact can have several different causes, each with a slightly different appearance.
 

Bit-Flip Noise

Bit-Flip Noise is a specialized form of IN that causes single-pixel shifts in the data that are set powers of 2—a pixel may be 128, 64, or 32 DN higher or lower than its actual value. (Smaller shifts occur—16, 8, 4, and 2 DN—but are generally too small to see.) This artifact arises from some digital source, often a transmission error or a problem in the ground processing systems. Unlike most sources of IN, Bit-Flip Noise is often correctable.

Example of Bit-Flip Noise in Landsat 5 Thematic Mapper (TM) Band 3 data
Example of Bit-Flip Noise in Landsat 5 Thematic Mapper (TM) Band 3 data. 

Transcription Artifacts

Noise pixels appearing in a regular pattern throughout the scene often indicate a ground processing or transcription problem. Reprocessing the scene usually eliminates the artifact.

Example of a Transcription Artifact that resembles Impulse Noise in Landsat 2 Wideband Video Tape data
Example of a Transcription Artifact that resembles Impulse Noise in Landsat 2 Wideband Video Tape data. 

Fire Noise

If the IN pixels appear only in Bands 5, 6, or 7, it might not be IN—it might be Fire Noise. Fire Noise is a common ground phenomenon, not an artifact. It mimics IN and occurs when small fires on the ground appear as very bright pixels in the near-infrared and thermal bands. These Fire Noise signals are found only in inhabited areas, generally rural areas, during certain seasons, and in weather conditions that are favorable for burning brush and refuse. They appear as bright points in near-infrared and thermal bands (Bands 5, 6, and 7) and are often saturated or oversaturated in Bands 5 and 7. They do not generally appear in visible band data, but large fires may have visible smoke plumes in high-resolution imagery.

Example of Fire Noise artifacts in Landsat 7 Enhanced Thematic Mapper Plus (ETM+) data, Bands 5 and 7.
Example of Fire Noise artifacts in Landsat 7 Enhanced Thematic Mapper Plus (ETM+) data, Bands 5 and 7.

Fire Noise is common in Landsat 7 Enhanced Thematic Mapper Plus (ETM+) imagery because the high spatial resolution of the instrument allows it to see smaller fires. While Fire Noise does occur in Landsat 5 Thematic Mapper (TM) imagery, it is more rare and only very large fires can cause the artifact.

Although Fire Noise is often mistaken for IN or Single Event Upset (SEU), it is not a true artifact. Fire Noise is an accurate observation of the ground. However, if these ground fires are bright enough, they can cause oversaturation artifacts in Bands 5 and 7.

Other Types of Impulse Noise

Different forms of IN have slightly different appearances, and other names for it include Salt and Pepper Noise and Low Saturation Noise. If any form of IN pixels appear only in certain detectors, it could be a warning of an imminent dead or flat detector problem. This artifact is not correctable and might indicate future problems.

If the IN only occurs in dark regions and has a value of 0 DN, it is called Low Saturation Noise. This artifact occurs in instruments with low dark currents and high coherent noise—Landsats 5 and earlier, usually in Bands 5 and 7. Low Saturation Noise is the result of expected amounts of Random Noise plus expected amounts of Coherent Noise reinforcing each other to occasionally low saturate the detector at 0 DN. Low Saturation Noise is uncorrectable and is not a cause for concern.

If the IN appears in groups of several pixels, and the scene was acquired over South America or in polar regions, it may not be IN—it may be caused by SEUs.

If the pixels are evenly distributed across the scene and have no apparent pattern or set value, further analysis is required to determine the cause and nature of the noise.