Aiming at the problem of uncertainty regarding the threshold parameters in the quality control of radiosonde temperature using the bi-weight standard deviation method, an optimization algorithm for threshold Z parameters based on the skewness and normal waveform indicators is being proposed. By utilizing the temperature data from three radiosonde stations in Fujian in 2017 and the ERA5 model reanalysis temperature data, a comparative analysis of the different datasets of the sounding temperature increment data before and after quality control is conducted. The distribution curve of the kurtosis-skewness value C_KS with threshold Z indicates that a higher threshold Z will result in incomplete quality control, while a lower threshold Z will result in excessive quality control. The optimal threshold Z of the bi-weight standard deviation method, as determined by the kurtosis-skewness value C_KS, is more consistent with the normal distribution requirements of the model assimilation system than the fixed threshold of the bi-weight standard deviation method. This provides a better method for quality control of radiosonde temperature data in model assimilation. Regarding the distribution characteristics of temperature increment before and after quality control, the temperature observation increment consistently removes outliers from the entire pressure atmosphere using a fixed increment value. Around 57.15% of outliers are primarily distributed in the range of 0-100 hPa, while the remaining 42.85% of outliers are evenly distributed in the range of 100-1000 hPa. The abnormal points, where the absolute value of temperature increment within the range of 0-100 hPa is greater than 10 ℃, are caused by factors such as solar radiation affecting the temperature sensor, which exceed the normal detection error range. These abnormal points can be removed in advance before quality control, further improving the quality of radiosonde temperature data.