In case any others come across this issue:
if you start with dicom files and use dcm2niix, and two magnitude nifti files will be generated with one correspoding to each echotime, and you will also see the following warning:
"Slices not stacked: echo varies (TE 4.92, 7.38; echo 1, 2). Use 'merge 2D slices' option to force stacking
Warning: Interslice distance varies in this volume (incompatible with NIfTI format)."
  
More digging gives me the following answer
"Specifically, when dcm2niix generates the warning Interslice distance varies for a Siemens fieldmap, it typically suggests that one of your tools that handles DICOM images has deleted some of the images. You can fix this by getting the images directly off of the console, or checking the tools that archive your DICOM images. It is not a problem with dcm2niix, but reflects data loss upstream. The core issue is that Siemens will generate identical Session, Series and Instance IDs for each echo in a field map. Many tools assume this means that these "identical" images are redundant, and randomly deletes one of the echoes. Which echo is deleted varies randomly for the 2D slices in a 3D stack."
 
Since fieldmap correction does not care which magnitude image to use (https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/F... also see below), you can force stack data from two echos to obtain one magnitude image using -m option.
"If you have data from a SIEMENS scanner then we strongly recommend that the tool fsl_prepare_fieldmap is used to generate the required input data for FEAT or fugue. Fieldmap data from a SIEMENS scanner takes the form of one phase difference image and two magnitude images (one for each echo time). In the following, where a magnitude image is required, pick the "best looking" one. This image is used for registration and masking but the process is not particularly sensitive to the quality and typically either image will work fine."

Oliver