Study On The Effect Of Different Sampling Methods On Quantile Estimates

Rainstorm is one of the most serious natural disasters. In recent years, the extreme hydrological events, especially flood disasters caused by frequent rainstorms, are of high occurrence, troubling urban flood control departments. Taihu Lake Basin is one of the typical areas that rainstorm disaster occurs frequently in China, with the most advanced economy and the highest level of urbanization. Hence, precipitation analysis in this region is of much significance. This paper explains more about how the frequent events are of high occurrence by comparison with the effect of different sampling methods on quantile estimates, and offers two kinds of reasonable suggestions at last. The major results are as follows:Firstly, introduce the theories of Regional L-moments Analysis (RLMA) method, and its application of hydrological frequency analysis in the Taihu Lake Basin. According to the criterion, buffer area of0.5～1°around the Taihu Lake Basin and8homogeneous regions is eventually divided, and the most fitted distribution for each region are, in order, GEV、GLO、 GEV、 GEV、 GNO、GNO、GEV、 GNO. Consequently, based on RLMA method, the precipitation quantile estimates in the Taihu Lake Basin can be calculated with AMS and AES data, respectively.Secondly, compare the effect of AMS and AES on precipitation quantiles from two aspects of theory and real data check. Theoretically it finds out that the quantiles estimated based on AMS under current conventional computation are underestimated, especially for frequent events. Otherwise the quantiles based on AES are more reliable as it’s in accordance with the definition of return period.In real data check, the empirical frequency of AMS data in the Taihu Lake Basin goes near to its theoretical exceedance probability, that is to say, the quantiles based on the AMS are of little underestimated, contradicting with theoretical analysis. By comparison with the standarded quantiles based on AMS and AES, it shows that different sampling methods make significant effect on rainstorm quantile estimates during2-y to10-y, and the difference is relatively small as with the increasing return period.Thirdly, suggest the possible causes to the difference can be, via a skewness analysis of AMS data in the Taihu Lake Basin, that the stations and the data length used for the study are very limited. Furthermore, it observes only fewer high values available in the high value interval of their AMS histograms for most stations, exhibiting a discontinuation in the AMS histogram. The causes for the discontinuity are unknown. So more studies are needed by collecting large amounts of data in China to further confirm the theoretical analysis.Fourthly, put forward two kinds of reasonable suggestions by using Chow V.T. equation to correct the underestimation of quantiles based on AMS data. Use the PDS (AES) data in combination with the use of the exceedance probabilities0.5,0.2,0.1,0.04,0.02and0.01for return periods of2-y,5-y,10-y,25-y,50-y and100-y. Or use the AMS data in combination with the use of the exceedance probabilities0.3935,0.1813,0.0952,0.0392,0.0198and0.0099for return periods of2-y,5-y,10-y,25-y,50-y and100-y. This paper encourages the use of the latter.Finally, the paper finds that1-y event of AES is equivalent to the1.58-y event of AMS, so1-y rainstorm quantile estimates of each station in the Taihu Lake Basin can be obstained by the conversion metioned above.