Thanks @rabernat for the advice. I realized the amount of work after another look at the xarray’s github page. The example @weiji14 showed is very helpful. One thing that I noticed is it only reads back all the variables immediately below the specified group path. The variables at a lower level will not be retrieved. The version 2 of the ICESat-2 ATL10 sea ice freeboard data is an example with such complicated data structure. For example, there are variables under 'gt2r/freeboard_beam_segment/height_segments/' as well as 'gt2r/freeboard_beam_segment/'. I am not sure if the version 3 data will be simpler. ATL07 and ATL09 are slightly simpler, but they still have variables under different subgroups of the same beam (the top level). Can xarray handle such group structure in some way? Otherwise, it seems the accessors might be a promising way for icepyx to take advantage of the existing xarray functionality. @weiji14, do you think it would be easier to write customized intake wrapper for different dataset? The issue with this approach is that the resulting data will be a couple of xarray datasets and we still have to organize them in some way, such as packaging them into the IS2 data class.