Provide a way in sunpy to make composite image maps that combine short and long exposure time images

[kreevescfa]

Describe the feature

It would be nice to have a function that would make a composite map from short and long exposure time images that are close in time. Saturated pixels from the (exposure-time normalized) long exposure time image would be substituted with pixels from the (exposure-time normalized) short exposure time image. Care would have to be taken with the meta data, since two images are combined into one.

Proposed solution

Here is my (basic!) code to do this with AIA images if helpful (some things hardcoded for AIA).

import copy
def make_composite_image(file1,file2):
map1 = sunpy.map.Map(file1)
map2 = sunpy.map.Map(file2)

#check wavelengths are the same
if map1.wavelength.value == map2.wavelength.value:
    #check that there is a short exposure time
    if map1.exposure_time.value < 1.0 or map2.exposure_time.value < 1.0:
        if map1.exposure_time > map2.exposure_time:
            long_map = map1
            short_map = map2
        else: 
            long_map = map2
            short_map = map1

        saturated_indicies = (long_map.data > 4000).nonzero()
 
        composite_map = copy.deepcopy(long_map)
        composite_map.data[saturated_indicies] = short_map.data[saturated_indicies]

        return composite_map
    else: print("no short exposure time")
else: print("files need to be the same wavelength")

[wtbarnes]

Thanks for raising this @kreevescfa!

A few questions:

• In general (i.e. independent of a particular instrument), how would you decide which set of metadata to pass to the resulting map?
• Are there any metadata keys that would need to be modified to indicate that this is a composite map? Naively, I would expect this is probably a pretty instrument-dependent choice, but maybe not.
• Is it assumed that these maps are close enough together in time? Should there be some option to reproject one into the pixel grid of the other so you can be sure the pixels correspond to approximately (within the resolution of the instrument) the same region of the Sun?

My gut reaction is that something like this would be better suited to live in an instrument-specific package, particularly when it comes to metadata manipulation. However, if it is used by multiple instruments, it would be silly to duplicate the code in each instrument package.

[kreevescfa]

The XRT routines use the long exposure for the FITS header. Most of the pixels are probably from the long image, so maybe that makes sense? The XRT software updates the history tag in the FITS header to indicate that it is a composite image. Maps should be close together in time.

Both AIA and XRT would certainly use something like this, since they both run sequences of long/short exposure time images to increase dynamic range.

[kreevescfa]

From the documentation for mk_xrt_composite.pro:
"The index structure for the long image will be used for the combined index structure with the history updated with the short image information."

XRT has software for triple composites too (combining three images with different exposure times), but that's probably unique to us.