In this study, regional frequency analysis of extreme rainfall were conducted using two different methods based on the most updated rainfall records in Hong Kong. The results of the analysis were also compared. The two methods are index-event L-moment based method of Hosking and Wallis (HW) and hierarchical region (HR) method. This study analyzed the annual maximum rainfall series of duration 5-, 30-, 60-, 120-, 240-, 360-, 480-, 1080-, and 1440-min from a total of 122 automatic raingauges with record lengths varying from 10 to 26 years. HW method divided the entire territory of Hong Kong into 3 to 5 hydrologically homogeneous sub-regions according to geographical attributes (latitude, longitude and elevation) and mean annual rainfall (MAR). HR method classified Hong Kong into...[ Read more ]

In this study, regional frequency analysis of extreme rainfall were conducted using two different methods based on the most updated rainfall records in Hong Kong. The results of the analysis were also compared. The two methods are index-event L-moment based method of Hosking and Wallis (HW) and hierarchical region (HR) method. This study analyzed the annual maximum rainfall series of duration 5-, 30-, 60-, 120-, 240-, 360-, 480-, 1080-, and 1440-min from a total of 122 automatic raingauges with record lengths varying from 10 to 26 years. HW method divided the entire territory of Hong Kong into 3 to 5 hydrologically homogeneous sub-regions according to geographical attributes (latitude, longitude and elevation) and mean annual rainfall (MAR). HR method classified Hong Kong into 5 MAR-based sub-regions and established regional functional relationships between extreme rainfall statistics and MAR. Goodness-of-fit Z-test and LMR diagram were used to select the best-fit regional distributions from five three-parameter probability distributions (GEV, GLO, GPA, LN3 and PE3) and the five-parameter Wakeby distribution. Relative difference and root-mean square difference of the analysis results by the two regionalization methods were assessed with reference to the at-site rainfall depth-duration-frequency. Hosking-Wallis method was recommended here due to its better performance in the empirical assessment and more rigorous theoretical basis, development and adoption in worldwide. Several factors associated to the sample data and procedures of Hosking-Wallis method (varying site record lengths, number of clusters, and distance measure for discordant site detection) were also studied.