An improved online calibration method for weather radar based on ground clutter reflectivity has been proposed, aiming to promptly detect and correct measurement errors in reflectivity caused by radar hardware failures or performance degradation, thereby enhancing the accuracy of quantitative precipitation estimation. Utilizing the Gabella weather radar clutter identification algorithm, the unfiltered reflectivity data is labeled for ground clutter. Based on this, a fixed-dimensional ground clutter data matrix is obtained through resampling, and the frequency of ground clutter occurrence is statistically analyzed. The cumulative probability density function of the ground clutter data is calculated, and the reflectivity value corresponding to its 95% distribution is selected as the monitoring value to identify stable ground clutter data. The baseline value of the ground clutter reflectivity is then obtained using the metal sphere calibration method. Finally, the relative calibration adjustment value of the radar is calculated by the difference between the monitoring value and the baseline value. Sensitivity tests were conducted on the selection range of ground clutter data, range correction, and large-scale precipitation conditions, optimizing the screening threshold and significantly improving the sensitivity and reliability of online radar system bias calibration. Experimental results from the CINRAD/SA-D weather radar at Changsha Lianhuashan showed that on March 28, 2024, under large-scale precipitation conditions, the average hourly relative calibration adjustment value of the radar was 0.22 dB, with a standard deviation of 0.76 dB. The daily relative calibration adjustment value standard deviation in March was 0.75 dB, and a change in radar calibration constant of approximately 2 dB was effectively monitored from March 8 to 9. On September 11, the ground clutter reflectivity baseline value obtained from the metal sphere fixed-point scan was compared with the theoretical value, revealing a radar system bias of -0.13 dB and a standard deviation of 0.26 dB. This method effectively enhances the sensitivity and reliability of online monitoring of weather radar system bias, providing a feasible technical approach for the online calibration of weather radar.