A New Outlier Detecting Model for Geospatial-Autocorrelation Data

This paper presents a new approach for outlier handling and analysis of geospatial-autocorrelation data. We improved a new outlier detecting model named comparative method of unit matrix (CMUM) based on standard-deviation rule (SDR) and estimating neighborhood method (ENM) and compared this model with SDR, ENM and influencing coefficient method (ICM), and evaluate all four methods with respect to their advantages and limitations by geostatistics analysis and visualization software. Examples of visual displays are provided for data on soil nutrients in Pingba county, Guizhou Province. We suggest a combination of SDR with CMUM for analysis and integration of spatial variability data. Emphasis is put on combination of traditional statistical, geostatistical, visualization techniques and software for geospatial autocorrelation data exploration. The combination improved the precision and efficiency of geostatistic analysis, enhances the information obtained and gives more complete description of the distribution and associations.