In this guide
Imagery Processing Decision Tables
Last Updated: March 9, 2022
Information for users to decide what imagery processing stretch and bit depth options are most appropriate for varying purposes.
Imagery Stretch & Bit Depth Outline
| Byte (8-bit integer) | UInt16 (16-bit integer) | Float32 (32-bit Decimal) | |
| No stretch (ns) |
Not Applicable |
Terrain Corrected with digital number (DN) values |
Not Applicable |
| Reflectance (rf) |
Optimized image contrast over snow and ice surfaces |
Comparative analysis between multiple images or cross-sensor imagery |
Spectral response analysis |
| Modified reflectance (mr) |
Optimized image contrast in temperate or tropical regions |
Not Applicable |
Not Applicable |
| Radiance (rd) |
Not Applicable |
Not Applicable |
Use in atmosphere correction model |
Imagery Stretch & Bit Depth Use Cases
|
Purpose |
UInt16 ns |
Byte rf |
Byte mr |
UInt16 rf |
Float32 rf |
Float32 rd |
| Comparison & analysis between multi-temporal images | ▬ | ⚫ | ⚫ | ⚫ | ⚫ | ⚫ |
| Visual interpretation of snow/ice regions (Antarctica/Greenland) | ▬ | ⚫¹ | ▬ | ⚪ | ⚪ | ⚪ |
| Visual interpretation of diverse land cover (nonpolar/vegetation) | ▬ | ▬ | ⚫¹-² | ⚪ | ⚪ | ⚪ |
| Full 11-bit radiometric depth | ▬ | ▬ | ▬ | ⚫ | ⚫ | ⚫ |
| Reflectance values (TOA) | ▬ | ⚪³ | ▬ | ⚪³ | ⚫ | |
| Radiance values (absolute) | ▬ | ▬ | ▬ | ▬ | ▬ | ⚫ |
| Original digital number values (no radiometric calibration) | ⚫ | ▬ | ▬ | ▬ | ▬ | ▬ |
| Has an absolute unit | Percent reflectance scaled to 200 | N/A | Percent reflectance scaled to 2000 | Percent reflectance | w/m²/str |
▬ not recommended ⚪ appropriate ⚫ optimal
¹ – 8-bit rasters are rendered faster than 16 or 32 bit data. A GIS does not have to apply a histogram stretch to take 11 to 32 bits of data and render them in 8-bit. However, radiometric detail in shadows or very bright areas may be lost in reducing the data to 8 bit.
² – Modified reflectance is based on reflectance, but shifted to enhance darker pixels. It’s useful to make non-ice-covered areas bright enough interpret, but pixel values are not a valid percent reflectance value.
³ – Reflectance values scaled to fit into an integer package.