Study On Statistical Test Of Artificial Precipitation Enhancement Effects

The evaluation of cloud seeding effects remains one of the most important and difficult issues.This dissertation focuses on some modern statistical methods(e.g.,cluster analysis using Mahalanobis distance,periodic analysis,separation of variables in different time scales,bootstrap,etc.),conventional ground meteorological data,etc.Many seeding records and much professional experience has been accumulated in Jilin province where the first cloud seeding happened in 1958.Aimed at evaluating effects of cloud seeding with different targets in Jilin province during April-July,this paper analyzes water resources in the air and precipitation,divides Jilin province into some subareas,and selects proper covariates.At last,statistical numerical simulation is used to analyze the influence of natural variance in testing seeding effects,while data deletion model is built to control the influence and improve the design of operation and evaluation.The main conclusions are:(1)Summary of water resources in the air and on the ground.Atmospheric water in Jilin province is the most abundant in summer,followed by autumn and spring,and the lowest in winter.More than 90%atmospheric water concentrates below 500hPa.Therefore,seeding below 500hPa during April-July has good condition in atmospheric water.Annual precipitation in Jilin province usually has a cycle of 4.2a-7.6a,while monthly precipitation has a single peak(the highest value occurs in July).From the spatial distribution,terrain influences precipitation so significantly that rainfall in the southeast is greater and varies faster than in the northwest.Baicheng and Songyuan have been the important target area for the frequent drought during April-July.(2)Proper division of subareas.Floating control historical regression method,which needs subarea division using cluster analysis at first,is used to test seeding effect when the target area is not fixed.It is concluded that the number of clusters(subareas)should be 6,according to four statistics(R2,pseudo F,pseudo T2 and semi partial R2)and spatial distribution of precipitation in Jilin Province.Cluster analysis using Mahalanobis distance can overcome the influence of units and correlation in Euclidean distance.This dissertation considers the influence of weights and sample categories,and improves the algorithm of covariance matrix in Mahalanobis distance.As a result,the accuracy rate of clustering can be increased by at least 6.63%,and the stations are very similar in the characteristics of terrain and precipitation.(3)Selection of physical covariates.Based on the ground meteorological data,the global NCEP reanalysis data and the large-scale climate indices,covariates of monthly precipitation have been chosen using variable separation,correlation analysis and stepwise regression,while covariates of daily rainfall have been analyzed using simple correlation analysis and canonical correlation analysis.The inter-decadal monthly precipitation has relationship with large-scale climate indices,the global NCEP reanalysis data,and the inter-annual component is related with the above three kinds of data.Wind speed,temperature,vapor pressure and relative humidity are taken as covariates of daily precipitation.The improvement has been verified when proper covariates have been selected,using blank tests and seeding empirical analysis.(4)Influence of natural variability.An improved numerical simulation method is established based on the modern statistical method of "bootstrap," to analyze the natural rainfall variability in Jilin Province.The variation in Subarea No.1-No.6 is in the range of(-20%,20%)."Detectable lower limit" of seeding effect is defined using one-side confidence upper limit of natural variation.A case deletion model has been built when we remove the outliers and select the control data similar to the seeded ones,and shown efficacious using 100 blank tests and 35 experiments.(5)Effects of artificial precipitation enhancement.Calculated seeding effects in Jilin province concentrate in the range of(0,30%),with an average of 11.95%.The effects are due to complex factors and have no significant linear relationship with precipitation itself.However,heavy rainfall gives smaller fluctuation of the seeding effect than light rainfall.Effects vary greatly when precipitation is small and gradually become stable as precipitation increases.