Data Analysis Method And Applied Research Of Spatial Autocorrelation

The applied research of spatial statistics is the frontiers and hotspots of statistics.The subject of this dissertation is the applied research of spatial autocorrelation analysis,which is a new focus of this field developed in recent years.Furthermore,we used the spatial autocorrelation theory and the improved spatio-temporal autocorrelation statistic and spatio-temporal autoregressive model in marine ecology for the first time.We firstly introduced the evolution of spatial autocorrelation analysis including the concepts,theories and models;And then we researched the influence of South-Asian tsunami 2004 on the spatio-temporal distribution of Chlorophyll-a(CHL-a) and compared the differences of this distribution during three different huge tsunamis using spatial autocorrelation statistic;Finally,we studied the spatio-temporal effects of four variables(sea surface temperature, normalized water leaving radiance,rainfall and wind speed) on the distributions of Chlorophyll-a before and after South-Asian tsunami 2004 using spatial autoregressive analysis and spatio-temporal models.In modeling CHL-a in relation to other four explanatory variables(sea surface temperature,normalized water leaving radiance,rainfall and wind speed),we conducted an exploration of models in spatio-temporal autocorrelation statistic and spatio-temporal autoregressive models.The application of spatio-temporal models in spatial statistics is still a challenge at present.Based on the model of spatio-temporal autocorrelation statistic just developed by L(?)pez-Hern(?)ndez and Chasco-Yrigoyen(2007),we developed a high-order model and visualized it in three dimensions;And based on the spatial autoregressive models initially developed by Anselin(1988 page:61-72) to model spatial econometric data,in order to consider the autocorrelation of CHL-a in time lags,we developed a new spatio-temporal autoregressive model.We used this spatio-temporal autoregressive model in marine ecology for the first time.The results of our research showed that there was a significant spatio-temporal autocorrelation in CHL-a during the tsunami period.There were considerable variations in spatio-temporal distributions of CHL-a round the occurrence of tsunami.The variation in spatial distributions of CHL-a seemed more anomalous around the time of the tsunami occurrence. From the comparison of spatio-temporal distributions in CHL-a in three different tsunamis,it is evident that:At least in one of the results of spatial autocorrelation indexes(Moran' s I and Pearson autocorrelation),there existed noticeable anomalous fluctuation in spatio-temporal distributions of CHL-a before the occurrence of three different tsunamis.It is obvious that this phenomenon should be a precursory signal of the occurrence of tsunami.Further study is needed to confirm the underlying reason of this phenomenon and if this phenomenon can become a method to forecast the occurrence of tsunami.We suggest that: To conduct tentative forecasting of tsunami occurrence by detecting the anomalous changes of spatio-temporal distribution in CHL-a,Moran' s I and Pearson autocorrelation indexes should be together used to study the autocorrelation analysis of the spatio-temporal distribution in CHL-a in the easy earthquake-occurred sea area.Because of the spatial autocorrelation in the distributions of CHL-a, the ordinary linear regression model is invalid.Spatial error models (including spatial model and spatio-temporal model) that effectively took account of the spatial autocorrelation in CHL-a were applied in the third chapter.The spatial error models revealed that sea surface temperature and rainfall had a negative effect on CHL-a,while normalized water leaving radiance and wind speed had positive effects.The effects of all the four explanatory variables were significant(positive or negative) before the occurrence of the tsunami.Except for normalized water leaving radiance,the other three variables(sea surface temperature,rainfall and wind speed) all showed a certain degree of weakening of effects on CHL-a after the tsunami.In addition,there was a significant autocorrelation in CHL-a before tsunami along forward time lags.And this autocorrelation in CHL-a along forward time lags also showed a certain degree of weakening after the tsunami.We suggest the influence of tsunami caused it.