Detection of outliers in spatial data by using local difference

Abstract

Detection of outliers in spatial data is different from h e a r model of ordinary and independent sample data because there is strong correlation among the data. Most of spatial outliers detection method are based on standardized residuals of Kriging prediction and are easy affected by masking and swamping affection when multiple outliers are present. Mean while due to the limitation of Kriging linear model, the method usually can't correctly estimate the difference between outliers and their neighboring data. In this paper, we suggested a new criterion using local difference to estimate the difference. Because of using .positional relation to replace correlation of spatial data, the method can avoid masking and swamping problems, and can correctly estimate the difference of spatial data. The validity of the method is shown in example based on several simulated spatial dataset. Index termlrpatial data, Kriging prediction model, Outliers detection, Local difference