A differential reflectivity (Zdr) observation operator incorporating ice-phase hydrometeor classification was developed to explicitly account for ice-phase hydrometeor contributions in the data assimilation. This proposed observation operator is then implemented within a three-dimensional variational (3DVAR) data assimilation framework for direct assimilating radar Zdr observations. Its performance was evaluated using both single-observation and cycling assimilation experiments for a typhoon case. Results from the single-observation test indicate that assimilating Zdr introduces additional microphysical constraints, leading to more effective adjustments of hydrometeor compared with reflectivity-only assimilation. In the cycling assimilation experiments, the combined reflectivity and Zdr (RFZDR) configuration produces a more physically consistent representation of reflectivity and Zdr, along with a more realistic vertical and horizontal distribution of hydrometeors. These improvements propagate into short-term precipitation forecasts, with RFZDR exhibiting lower root-mean-square errors, enhanced spatial consistency, and gains in heavy precipitation metrics.