There used to be a time where a graphics designer would have to spend hours organizing and manipulating multiple files that represent the same image. First, the image is saved as a .raw file with low contrast and low saturation in order to preserve the data (allow for maximum dynamic range and widest gamut) for future adjustments. Depending on what the image is used for, the image is manipulated using a series of layers and changing the color space. After the image is optimized for its appropriate output requirements, the image is saved, and any value that exceeds the visible range as a result of manipulation is discarded. This process would have to be repeated each time there is a different use for the image—for example, the image used for print would look different than the image used for monitor display. All this work was necessary when pixel formats were represented as unsigned integers.

Representing pixels using floating point numbers changed this. This type of image is referred to as a HDR image, which uses 32-bits to describe a more extended dynamic range. This allows information to be more flexible; the choice of the appropriate pixel format depending on the usage is easier. Because the range is preserved throughout the manipulation process, there is no need to worry about color profiles and spaces. Saving an image in this format would be not be efficient for compression purposes because the human eye can only detect 10 million colors, which is less than a 24-bit representation of the image. However, using a 32-bit HDR image makes it easier for manipulations, without the hassle of multiple files.

http://en.wikipedia.org/wiki/High_dynamic_range_imaging
http://blogs.msdn.com/billcrow/archive/2006/06/22/642213.aspx
http://www.cambridgeincolour.com/tutorials/high-dynamic-range.htm