Spatial data mining for predicting of unobserved zinc pollutant using ordinary point Kriging

Abstract

Due to pollution over many years, large amounts of heavy metal pollutant can be accumulated in the rivers. In the research, we would like to predict the dangerous region around the river. For study case, we use the Meuse river floodplains which are contaminated with zinc (Zn). Large zinc concentrations can cause many health problems, for example vomiting, skin irritations, stomach cramps, and anaemia. However there is only few sample data about the zinc concentration of Meuse river, thus the missing data in unknown regions need to be generated. The aim of this research is to study and to apply spatial data mining to predict unobserved zinc pollutant by using ordinary point Kriging. By mean of semivariogram, the variability pattern of zinc can be captured. This captured model will be interpolated to predict the unknown regions by using Kriging method. In our experiments, we propose ordinary point Kriging and employ several semivariogram: Gaussian, Exponential and Spherical models. The experimental results show that: (i) by calculating the minimum error sum of squares, the fittest theoretical semivariogram models is exponential model (ii) the accuracy of the predictions can be confirmed visually by projecting the results to the map.