Main content area

Improving the spatial prediction of soil Zn by converting outliers into soft data for BME method

Zhang, Chu-tian, Yang, Yong
Stochastic environmental research and risk assessment 2019 v.33 no.3 pp. 855-864
Bayesian theory, data collection, geostatistics, heavy metals, kriging, prediction, probability distribution, remediation, urban soils, zinc, China
Understanding the spatial patterns of heavy metals is important for the protection and remediation of urban soil. Considering that the conventional Geostatistical methods, such as ordinary kriging (OK), are sensitive to dataset outliers, this study converted the identified outliers into a discrete probability density function (PDF). Then, the PDF was used as soft data in the Bayesian maximum entropy (BME) framework to perform a spatial prediction of soil Zn contents in Wuhan City, Central China. By using OK as the reference method, the BME framework was found to produce an overall further accurate prediction, and the PDF of BME predictions was further informative and close to the observed Zn concentrations. An improved BME performance can be expected if soft data with high quality are provided. The BME is a promising method in environmental science, where the so-called outliers that probably carry important information are common.