G-FAQ – Why is Bit Depth Important?

直接抄：

https://apollomapping.com/2012/August/article15.html

For this month’s Geospatial Frequently Asked Question (G-FAQ), I pivot to a topic that deserves more attention than it gets, and that is bit depth. Some of you may have heard this term when ordering imagery from Apollo Mapping or perhaps when downloading free Landsat data without understanding its implications. As such, let’s delve into this topic, addressing the following set of questions:

What exactly is bit depth and why is it important when ordering satellite imagery? Is 16-bit imagery harder to work with? When should I order 16-bit imagery versus 8-bit depth?

To start of this discussion, it is important to understand the difference between base-10 and binary number systems. In a base-10 system, each digit place in a number represents 10 possible values from 0 to 9 and then each successive digit place increases by ten-fold in its value. Let’s look at this mathematically:

Base-10 Number = 123

This can be written mathematically as: (1 x 10^2) + (2 x 10^1) + (3 x 10^0)
Which can be simplified to: (1 x 100) + (2 x 10) + (3 x 1) 
And final to: 100 + 20 + 3 = 123

Given our familiarity with math since the early days of school, creating base-10 numbers is something we do seamlessly as opposed to writing out a mathematical formula as most beginners working with binary number systems do. In a binary system, each digit place only has two possible values, 0 or 1. This mimics a computer chip which can either be off (0) or on (1) – and this is the reason that computers are based around binary numbers at the system code level. So then, let’s see how a binary number is created mathematically:

Binary Number* = 111 1011

This can be written mathematically using base-10 numbers as: (1 x 2^6) + (1 x 2^5) + (1 x 2^4) + (1 x 2^3) + (0 x 2^2) + (1 x 2^1) + (1 x 2^0)
Which can be simplified to: (1 x 64) + (1 x 32) + (1 x 16) + (1 x 8) + (0 x 4) + (1 x 2) + (1 x 1)
And final to: 64 + 32 + 16 + 8 + 0 + 2 + 1 = 123

* In binary numbers, you add a space every fourth digit to the left of the decimal place.

By following steps similar to what I present above, you can convert any binary number to base-10 – though going from base-10 to binary involves more effort, check this website out for more details on those steps.

Now that we have explored binary number systems, we can pivot our attention back to bit depth. Satellites are based on binary numbers and as such bit depth is measured accordingly. The bit depth of a satellite tells you the maximum number of values it can measure per spectral band. The higher the bit depth, the more information it can measure and thus the more sensitive the sensor is to different illumination values (typically called digital numbers) of the surface of the planet.

As we explained in more detail last month, a passive optical satellite measures the intensity of photons that are reflected from the surface of our planet. Take for example two hypothetical satellites, one with 8-bit depth and the other with 9-bit depth. The 8-bit depth satellite can measure up to 2^8 digital number values (or 256 values) for the intensity of photon reflection; while a 9-bit depth satellite can measure up to 2^9 values (or 512).

To look at that same hypothetical situation another way, for each value an 8-bit depth satellite can measure, the 9-bit depth satellite can measure two values. That means the 9-bit satellite is twice as sensitive in each of its spectral bands as is the 8-bit satellite. Now remember that most satellites offer at least 4 multispectral bands. As such, a 9-bit depth satellite can produce 16 times the number of RGB + NIR values for each pixel in 4 band multispectral imagery than can an 8-bit depth satellite – this has important implications I will discuss later in this G-FAQ series.

The table below shows the imaging characteristics of the most common satellite we work with and then the maximum number of spectral combinations possible per pixel. The final column shows the increase in spectral information as related to a traditional 4-band, 11-bit depth satellite such as IKONOS. You can see from this table that increasing the number of spectral bands has a much larger impact on the maximum number of spectral combinations than does bit depth.

It is important to note that rarely, if ever, will satellite imagery utilize the entire range of pixel values possible with its bit depth. Imaging companies such as DigitalGlobe make the conscience decision to ‘dampen’ satellite systems, assuring that digital numbers close to the maximum possible (i.e. 2047 for 11-bit and 4095 for 12-bit depth) for each pixel are rarely reached. When the maximum value is reached and/or exceeded, flares can occur which destroy the spectral information in this pixel and surrounding ones.

Now that I have explained the basics, let’s take a look at the bit-depth ordering options for satellite imagery. If you have ever placed an order with Apollo Mapping for imagery products, you may have noticed that the two options for bit depth are 8 and 16, not 11 or 12 as in the table above. The reason for defining bit depth in increments of 8 ties back to computer technology. A single bit of information – either a 0 or 1 – is memory’s building block. Since the 1960s, it has been common practice for 8 bits to make up a single byte; so that 2 bytes is comprised of 16 bits of information.

As such, the binary digital numbers that are embedded in satellite imagery files (a TIFF usually) will either be 8-bits or 16-bits in length. In order to make 11 or 12-bit depth imagery 16 digits (or bits) long, 0’s are added to the front of the binary number so that the value itself remains unchanged. When 11 or 12-bit depth data is delivered as 8-bit depth imagery, the values are scaled so that each 8-bit depth pixel value represents multiple 11/12-bit depth values. Accordingly, 8-bit depth imagery will show less spectral variability. This 11/12-bit to 8-bit depth scaling process can also introduce additional signal noise. One term you might hear associated with 8-bit Red, Green, Blue (or Natural Color) imagery is 24-bits. They use this term as it is 3 spectral bands with 8-bit depth each, or 3 x 8 = 24-bit imagery.

In next month’s edition of G-FAQ, I will continue this discussion on bit depth by looking at the advantages and disadvantages of 8 and 11/12-bit depth imagery as well as provide recommendations on when to order each.

Until our next edition of G-FAQ, happy GIS-ing!

Do you have an idea for a future G-FAQ? If so, let me know by email at brock@apollomapping.com. 

Find Out More About This Topic Here:

• Indiana University - Knowledge Base
• Penn State University - Chapter 8: Remotely Sensed Image Data
• Regional and Mesoscale Meteorology Branch - Effect of Bit Depth on GOES Images

Brock Adam McCarty
Map Wizard
(720) 470-7988
brock@apollomapping.com