Currently the regrid_time preprocessor sets the center points of the time bounds, e.g. monthly data at the 15th day of each month, 00 UTC.

I replicated this behaviour for multimodel stats in #685, but @sloosvel noticed in #673 that this creates slightly offset time bounds, e.g. the bounds of the time coordinate are at 12:00 UTC instead of midnight at the start of the new month.

Note that #744 proposes to move all time handling from multimodel to regrid_time, so I think this issue will have to be addressed in the regrid_time preprocessor, not in multimodel.

@bouweandela propopsed the following approach:

set the lower bound to the start of the month, the upper bound to the start of the next month and then set the points by computing 0.5 * (lower_bound + upper_bound).