Maintaining the calibration accuracy of differential reflectivity (ZDR) within 0.2 dB for both X-band and S-band weather radars is a prevailing operational requirement. With the advancement of weather radar detection technologies, X-band phased-array weather radars (XPAR) offer superior spatiotemporal resolution and are particularly advantageous for low-altitude detection. However, they suffer from ZDR bias caused by the deterioration of antenna isolation, which results from the variation of the normal direction along with changes in the beam pointing angle. To systematically investigate the primary factors affecting XPAR data quality, this study selects synchronized observations from XPAR and an S‑band polarimetric radar (SPOL) over multiple periods, and proposes a refined ZDR correction algorithm (RC-ZDR) that addresses four aspects: systematic bias correction, attenuation correction, cross‑polarization isolation degradation compensation specific to phased‑array systems, and correction for anomalous radial interference. For systematic bias, light rain below the zero-degree layer is employed as natural calibration targets, and a weighted correction method (Bias-WZDR) based on the mean ZDR values at three elevation angles whose beam pointing angles are close to the array normal direction (0°) is proposed to mitigate the interference of cross‑polarization isolation on system calibration. Attenuation correction formulas are applied to moderate and heavy precipitation regions. In addition, a 9.4 GHz dual‑polarized microstrip antenna array is modeled using High Frequency Structure Simulator (HFSS) electromagnetic simulations to reveal the variations of antenna gain and 3 dB beamwidth with beam pointing angle. Based on this, a quadratic function correction method for cross‑polarization isolation (QCCPI) is proposed to quantitatively compensate for the degradation bias between different elevation angles. A composite threshold method is adopted to filter out radial interferences in low‑level XPAR observations, further improving the quality of basic data. During a moderate rain event at the ZG100 site, after QCCPI correction, the maximum difference in mean ZDR among different beam pointing angles decreased from 0.45 dB to 0.27 dB. The proposed RC-ZDR method is then applied to a squall line event observed at the ZG100 site on June 6, 2023. Spatiotemporal matching and comparison with SPOL data demonstrate that the corrected ZDR data can more accurately capture the polarimetric signatures of the strong convective core of the squall line. The overall mean ZDR of XPAR across all elevation angles is corrected from -0.591 dB to 0.226 dB, and the bias relative to SPOL observations is reduced from 0.619 dB to 0.198 dB. This work provides a scientific and efficient technical support for operational quality control and performance evaluation of next‑generation phased‑array weather radars.