Theoretical Research Of Semiparametric Statistical Approach And Its Application To Climate Analysis

In Recent Centuries, the weather of the global and China has Significant change, which is caused by the nature and Human activities. The Climate warming has been a focus. This paper analyze the weather and the weather change-point based on the partially linear model and structural changes problems. The main content of this paper as the follows:Firstly, the research status of the partially linear models,the varying-coefficient partially linear models and the structural changes problems,the methods of detecting and estimating the change-points was given.Secondly, local linear estimation in varying-coefficient partially linear models was given. When the error sequence of the model is not i.i.d., a simulation shows that this estimation is efficient. The measure standard is the Relative efficiency ρ=MISE1/MISE2.And we got that the Relative efficiency decreases as the sample size increase.Thirdly, based on the un-homogenization temperature data of552weather stations from the year1961to2010, the annual average temperature, the annual maximum temperature and the annual lowest temperature variation were analyzed. By the method of least squares, we calculated their climate tendency rate. Considering that the temperature decreases as the latitude increases. So this paper proposed three models based on the relationship of the highest temperature of110weather stations during year1961to2000and the longitude, latitude. And a simulation shows that the partially linear model is more efficient than the other two models.Finally, A series of methods of detecting the change-points was listed, including sliding t-inspection method, Cramer method, Mann-Kendall method, Yamamot method and BG method. In this section, the change-points of the annual average temperature, the annual maximum temperature and the annual lowest temperature from the year1873to2009of Shanghai are detected by the methods of sliding t-inspection method, BG method and CUSUM method. The solution shows that there are no change-points when using BG method; and the test results detected by the method of sliding t-inspection are that there are change-points of the annual lowest temperature and the annual average temperature; under five different y values respectively when using the CUSUM method, we have evaluated the change-points of the annual maximum temperature, the annual average temperature and the annual lowest temperature. And the results of the above three methods shows that the result evaluated by the method of the CUSUM is more accurate.