cftime and open_mfdataset conflict

[jrobrien91]

Describe the bug
During the 2026 ARM Summer School, when attempting to open MFRSRAOD data via open_mfdataset, users received errors associated with the additional coordinate fields denoting the calibration times, particularly non-monotonic global indexes along these additional dimensions.

From the user:

The mfrsr7nchaod1michM1.c1 datastream contains the normal ARM usage of base_time, time_offset, and time, each with units specified as “second since …”.  But in addition, it also contains two extra fixed-length coordinates fields “Io_interquartile_time” and “Io_gauss_time” that specify the times of calibration values.  Here is how they are defined:
 
float Io_interquartile_time(Io_interquartile_time=72);
  :long_name = "Time for interquartile Io sample times";
  :units = "seconds since 1970-01-01 00:00:00";
 
float Io_gauss_time(Io_gauss_time=61);
  :long_name = "Time for Ios after gaussian filter";
  :units = "seconds since 1970-01-01 00:00:00";

Issue appears to be within 200+ ARM datastreams that follow this cftime convention

To Reproduce
ds = xr.open_mfdataset(mfrsr7nchaod1michM1.c1_files[-3:])

Expected behavior
ds = xr.open_mfdataset(files[-3:])
ValueError: Resulting object does not have monotonic global indexes along dimension Io_gauss_time

Screenshots

Desktop (please complete the following information):

Smartphone (please complete the following information):

Additional context
Will forward the email chain to maintainers

[cgodine]

Hi @jrobrien91,
I took a look at combining mfrsr7nchaod1mich files, and I was able to produce a working Dataset with mulitple days' worth of data. It appears that the Io_gauss_time variable is a rolling window of 30 datetimes on either side of the current day, meaning that when xr.open_mfdataset attempts to concatenate it sees that all values except the first value of the earliest file and the last value of the latest file are overlapping. In other words, for a file on 2026-06-20, the dates in Io_gauss_time are dates from 2026-05-21 to 2026-07-21; combining files from 2026-06-19 through 2026-06-21 would then result in overlapping dates from 2026-05-19 through 2026-07-20.

Using xarray.merge solves the issue:

import xarray as xr

def combine_aod(mfrsr7nchaod1michM1.c1_files):
    objs = []
    for f in mfrsr7nchaod1michM1.c1_files:
        ds = xr.open_dataset(f)
        objs.append(ds)
    return xr.merge(objs, compat="minimal")

It might be worth adding some sort of exception handling or an additional function to merge multiple of these VAP files within io/arm.py. If you'd like I can assign this to myself and investigate a bit further.

[zssherman]

@cgodine Hi Corey! Yeah if your willing to investigate more that would be much appreciated!