Stochastic Simulation Of Precipitation In China

The weather generator can well produce long sequences of climate data,which can be used to simulate the climate in areas where data is missing and has been widely used to evaluate the effects of climate change.The simulative results will be different because of different models for its different parametric methods.Furthermore,the weather generator may also produce significant differences due to the dissimilarities of varied climate characteristics.It is therefore of urgent need to evaluate the applicability of more weather generators so as to choose simulation models reasonably.This study chose daily precipitation data at 743 weather stations from different climatic regions of China,based on the analysis for spatiotemporal distribution regulation of Chinese precipitation,four models were assessed for its simulate effect of precipitation occurrence.In addition,the performances of five amount models were compared to determine the applicability to simulate precipitation of China.Especially,we also compared the simulation results of five amount models which simulated the extreme precipitation indices.The main conclusions are as follows:（1）The spatial distribution of annual precipitation,precipitation days and most extreme precipitation indices in China,basically decreased from the southeast to northwest,while the change rate of annual precipitation days and the maximum days of continuous drought（CDD）were gradually increased over the last 5 decades.Seen from the trend of inter annual change,the annual average precipitation and precipitation days in arid-areas showed an increasing trend,while precipitation in other areas showed a decreasing trend.The variation tendency was not-significant.Four extreme precipitation intensity indices（PTOT、R95P、R99P、SDII）showed an increased trend in general,while four extreme precipitation frequency indices（CDD、CWD、R10、R20）was on the decline.（2）The weather generator can simulate precipitation probability parameters.In detail,Markov Chain with first order and two order（MC1 and MC2）well simulated the average monthly precipitation days,while the two and three order Markov Chain（MC2 and MC3）well simulated the standard deviation of precipitation days.The applicability of the models was great influenced by the climate.In different seasons and climates,MC1 and MC2 both performed better in simulating precipitation days,however,MC3 existed large deviation in summer and autumn.So as to SE,deviation existed on the whole year except winter.（3）The Mixed exponential distribution（MEXP）and the Webb distribution（WBL）were superiors for precipitation simulation on different time scales.Generally,model error increases with the increase of precipitation.In the semi-arid and arid-areas or less rainfall season,MEXP simulated well.In the opposite area or season,SKEW performed better.For this reason,when selecting the optimal precipitation model,seasonal and regional factors need to be considered,especially for the areas with high precipitation amount and precipitation intensity.（4）Comparing the simulation results of the extreme precipitation indices,the annual precipitation（PTOT）,percent of annual precipitation exceeding the 95^th（R95P）and 99^th（R99P）percentile were well simulated,CDD was relatively bad simulated.The average of extreme precipitation indices was well simulated relatively to the standard deviation for all the models,among them,SE was considered as the superior model.The temporal change of extreme precipitation indices was not well simulated by the five models.