GCU Wall of Text has arrived in this system.
Normal maps used in FS2 are tangent space normal maps that only use two channels of information. Tangent space means that the normal map affects the surface's local normal vector (which is the normal of the surface's tangent plane).
There are other types of normal maps that can use two or three channels (three channels is enough to specify a vector in three-dimensional space, which means it can be used as object space normal mapping. These normal maps work a bit differently - namely, they specify the normal vector's direction in the object's reference frame rather than relative to the surface tangent plane at each pixel's mapped UV location.
But FS2 normal mapping uses tangent space normals, so two channels is enough. One channel contains the tilt of the surface on horizontal direction, and the other on vertical direction.
In the most typical tangent space normal maps, the normals information is in red and green channels, while blue is used to save some height information (which may or may not be used by the application). This results in the typical bluish-purple normal maps that most normal map plugins produce as tangent space normal maps. In this type of normal map, "flat" or unaffected surface normal vector results from either colour #7f7fff or #8888ff (RGB(127,127,255) or RGB(128,128,255)). On a range from 0 to 255, there's 256 values, which means there's no exact middle point that can be expressed as an integer, which means the middle point is somewhere between values 127 and 128.
As far as normal maps are concerned, the colour that is interpreted as "neutral" would depend on the shader system; I do not know how FS2_Open's normal map shaders do the math. I do know that GIMP's normal map plugin uses #7f7fff as the "flat" colour... but I digress.
For the shader mathematics, it doesn't matter how the information is delivered. In most cases, though, compressed textures deliver sufficient quality while reducing memory load by a whole lot, so using DXT compression is beneficial.
Now, DXT compression causes a certain amount of interaction between red, green and blue channels due to how the interpolation processes work. This means that what is in the red channel will affect the contents of green channel and vice versa. This isn't so bad in diffuse, shine and glow maps; however it is practically fatal for normal maps.
However, DXT formats with a proper alpha channel (DXT3 and DXT5 - DXT5 is typically better for normals) save the alpha channel distinct from the RGB compression.
Because of this, it is beneficial to save one channel of normal map information on the RBG side of the file, and the other channel is saved on the alpha channel. This means there's no crosstalk between the channels, and higher quality is achieved. Usually, red channel is transferred to alpha channel. Green channel information is either copied to red and blue channels (which results in "grey" normal maps) or alternatively, red and blue channels are filled with solid black to fill the channel information with a bunch of zeroes (which results in green normal maps).
In other words, to achieve best quality for the green channel, red and blue channels must be either identical to green, or they must contain no relevant information which is achieved by having them fully black.
This DDS format where green and alpha channel contain normals information is called DXT5_nm (nm like normal map). Functionally, it is exactly identical to the DXT5 compression, the only difference is the purpose. Some DDS converters include a DXT5_nm option; what this does is essentially throw the red channel to alpha, then either copy green to red/blue or fill red/blue with black, then save as DXT5. This can be done manually as well.
Some converters results in greenish, some in grayish normal maps as they are opened in image editor. Functionally, there's no difference, as only green and alpha channels are read.
