NoData is a concept used to describe the areas of a raster image where there are no real data values. For all raster images, this includes the area outside the extent of the raster dataset. Raster images are always organized as a rectangular array of pixels. A pixel is either a member of the original datasets or it is NoData.
In general, any image operation involving a NoData value will result in a NoData value. A notable exception is when two images are mosaicked; in this case NoData results only if ALL operands resolve to NoData for a given pixel.
There are many common image operations that can give rise to NoData areas in an image.
Mosaicking two or more raster datasets into a single dataset can give rise to NoData in non-image areas. These operations are illustrated below where the original raster datasets are shaded and "ND" indicates areas of NoData in the new raster image.
NoData can be introduced into an image if a raster dataset containing all real data is geocorrected (or rotated) and rewritten to a new image creating a new dataset. It is common to produce an image with NoData when georeferencing tools are employed. Calibrated images produce NoData in their displayed outputs even if there are not any NoData pixels saved in the dataset. Simple resampling; that is, changing only the image cell size, will not produce NoData.
If a non-rectangular area of interest (AOI) is cut from a dataset and saved to a new image, all areas of the new dataset outside the original area of interest contain NoData.
Areas of NoData are not automatically identified as such by the operations that create them. You must explicitly create a NoData node and assign a value to represent NoData.