Abstract: A typical ecological unit of Dongting Lake plain was selected as experimental site. Approximate grid approaches were employed for the sampling scenario in 2004 with 651 Global Position System (GPS) established spots sampled in topsoil (0-20cm). The purpose of this paper is to evaluate various data processing methods, including logarithmic transformation, robust statistics, excluding outliers and Box-Cox transformation for evaluating total soil P content with normal distribution. The result showed that Box-Cox transformation was applied in order to achieve normality in the data set and to dampen the effect of outliers. Geostatistical analyses were carried out, including calculation of experimental variograms and model fitting. The best theoretical model for semivariogram of total soil P were spherical model. The ordinary kriging estimates of total soil P concentration were mapped. The integrative comparisons of semivariogram parameters with different trend of the kriging prediction errors of total soil P indicated that the 2-order trend effect was preferable. Kriging standard deviations (KSD) were regarded as the standard deviations of the interpolated pixel values and provided valuable information for which will increase the accuracy of the total P mapping. Spatial distribution and hazard assesment of total soil P in the typical area of Dongting Lake plain were investigated using geostatistics and geographic information system (GIS) techniques such as probability kriging on the basis of the software ArcGIS Desktop. Probability distribution of total soil P at different levels will be helpful to conduct hazard assessment, optimal fertilization and develop management practices to control the non-point sources of P pollution.

Key words: total soil P, geostatistics, spatial variability, probability kriging, Dongting Lake plain area