This was a feature of CMIP5 too and the CMIP6 data request paper mentions it:
... while near-surface relative humidity values of 140 % can, in principle, be realistic at a point in space and time, many of the high values in the CMIP5 archive, which represented time and grid cell averages, are likely to be caused by processing errors. Hence the upper-limit is set at 100.001 % and categorized as suggested, in contrast to the limit for sea ice extent that has a robust limit of 100.001 %
Juckes et al (2020): The CMIP6 Data Request (DREQ, version 01.00.31), Geosci. Model Dev., 13, 201–224, https://doi.org/10.5194/gmd-13-201-2020
So modeling centers were permitted to clip their data to 100.001 %, but presumably many of them chose not to. That papers cites Ruosteenoja et al (2017) for the CMIP5 analysis, who suggest several reasons:
- in some models supersaturations originate from the core of the model algorithm
- in others they have been formed only in the stage of creating the surface air moisture fields
- in several models both factors play a role: some degree of supersaturation has developed within the dynamic calculations but the excessive relative humidities have further been amplified when building the surface air humidity fields
- isobaric level fields [do not] represent the actual intrinsic model data but have been interpolated from the values given at model hybrid levels
They also contacted the modeling groups, who gave a variety of answers that fall into three categories:
- the definition of RH with respect to ice rather than liquid water
- contradictions in the determination of specific humidity and air temperature for the near-surface level
- nonlinearity of saturated specific humidity as a function of temperature
That paper is open access (I think) and worth a read for more details on all those points. It's also worth pointing out that there are plenty of screwy features of CMIP data, which is inevitable given the volumes of data, complexity of post-processing chains and expense of running the models. It gets better each round, but there will always be something odd in there. The important thing is to do what you're doing and query anything that doesn't seem right, rather than taking it at face value.
