*4.7. Overall Analysis*

Table 5 shows the results of the evaluation of the KNN classification results by combining multiple cases of different processes for six Doppler radars in Jiangsu in July, 2012. The POD of KNN for stratiform classification reaches 0.950, the FAR is 0.085, and the CSI is 0.874. From a comprehensive perspective, it is possible to accurately classify more than 85% of the observed stratiform precipitation areas. The POD of the convective classification reaches 0.781, the FAR is 0.137, and the CSI is 0.695. Anagnostou [18] also uses the 2A23 product to classify precipitation types using neural networks, obtaining values of POD = 0.97, FAR = 0.07, and CSI = 0.90 for stratiform precipitation classification and POD = 0.52, FAR = 0.29 and CSI = 0.43 for the classification of convective precipitation. In that paper, the results of SHY95 were also evaluated by 2A23; the stratiform POD, FAR and CSI values were 0.85, 0.05 and 0.81, respectively, and the convective POD, FAR and CSI values were 0.72, 0.59 and 0.36. The cases used are not the same, but with the KNN classification, although the e ffectiveness for stratiform precipitation decreased, the classification accuracy of convective precipitation improved significantly.


**Table 5.** Comprehensive evaluation of the KNN classification results.
