Hi Harry,
I have downloaded your image from our anonymous ftp server. As far as I can tell, there's nothing wrong with it.
The file is a standards-compliant three-channel, 32-bit floating point FITS image (presumably a RGB color image). It has pixel values ranging from -29420 to +50299, approximately. This is an arbitrary range that must have a perfectly sound meaning in the context of the application that has created the file.
Unfortunately, the FITS standard says absolutely nothing about the range of numerical values that can be stored in a floating point FITS image. Much worse than that is the fact that there's no (standard) way to know anything about the actual range of numerical values a particular floating point image has been referred to. In simpler words, there's no way to know which values correspond to the black and white points to which pixel sample values refer in a floating point FITS image.
PixInsight uses the normalized [0,1] range to store and manipulate floating point images, where 0=black (meaning "no signal") and 1=white (meaning "full signal"). Besides a number of technical advantages that we won't discuss here, the [0,1] range allows PixInsight to store and manipulate five different data formats in a completely transparent way (for example, you define histogram parameters from 0 to 1, irrespective of the data format of a particular target image).
Since you are reading a floating point FITS image whose pixels don't fit into the [0,1] range, PixInsight asks you how that FITS image should be interpreted, in order to know how to rescale it to the [0,1] range. This behavior can be configured with the FITS Format Preferences dialog (available from the Format Explorer window). The default option is to ask the user, but you can configure it to perform a number of actions automatically.
To avoid problems associated with arbitrary floating point ranges, the best option is to export FITS files in 32-bit integer format, if possible, instead of 32-bit floating point. 32-bit integers provide an effective range that is much larger than the range of 32-bit floating point, and unambiguously define the actual dynamic range of the data (from 0 to 2^32-1). To maximize compatibility with other applications, it is advisable to use unsigned integers instead of signed integers.
