51����

Data modelling is an important part of the wider research around the environment and human health. Statistical modelling techniques of soil property data are used to:

• look at differences in the concentration of potentially harmful substances (PHS) in different cities
• develop normal background concentrations (NBCs) of soil contaminants for use in human health risk assessment
• predict the mobility at different spatial scales (urban and rural scales; flood catchment)
• identify anthropogenic and geogenic signatures to assess the past impact of anthropogenic activity and urban land use 

The team use samples and data from the extensive 51���� G-BASE project as well as those from individual, site-specific studies. We apply a variety of statistical modelling approaches to laboratory and field-based measurements to identify different physical and chemical properties of soil e.g. near infra-red; total element analysis; soil pH, etc. The range of statistical modelling approaches includes least squares linear regression, multiple linear regression, stepwise linear regression, principal component analysis and machine learning (random forest modelling) to explore the relationships between geology and geochemistry, providing robust predictions of spatial contamination and its health impacts. 

Spatial modelling techniques can be carried out on a relatively small number of samples to predict element mobility on wider urban or regional scales. Application of our modelling techniques on spatial datasets allows us to create urban and regional maps of bioaccessibility and understand the relationship between soil geochemistry and measurements of health deprivation.