The original mixed phase partition plot released with v2.9 is based on mass mixing ratio (CLDICE and CLDLIQ). Suggested by @zyuying, using cloud fraction based simulator output (CLD_CAL_ICE and CLD_CAL_LIQ) will have fair comparison with obs. The additional variable needed is Z3 which is needed to convert T to COSP space.

The observation data is from:
Cesana, G. V., Ackerman, A. S., Fridlind, A. M., Silber, I., & Kelley, M. (2021). Snow reconciles observed and simulated phase partitioning and increases cloud feedback. Geophysical Research Letters, 48, e2021GL094876. https://doi.org/10.1029/2021GL094876