Mapping soil Pb stocks and availability in mainland France combining regression trees with robust geostatistics

Maps of lead (Pb) stocks in soils and estimates of its availability are needed to assess risks of contamination. Stocks in soils of total and ethylenediamine tetraacetic acid (EDTA) extractable Pb, as well as Pb availability, assessed by EDTA/total Pb ratio, were measured and calculated to a depth of 30 cm with the French soil monitoring network at sites defined by a regular 16 × 16 km grid. Setting aside punctual anomalies by winsorizing, these properties were mapped using linear mixed models (LMM). LMMs combined conditional partitioning trees upon 5 predictors (pH, texture, parent material, land use, population density) with robust geostatistics to avoid distortion due to outlying values. Rather than selecting the fixed effects according to expert-knowledge, regression trees were used to account for explanatory variables in a single classification. This original method stressed both the necessity for a geostatistical component to complement regression tree models when spatial correlation is evident, and the usefulness of these trees to interpret maps. Pb stocks varied widely with peak concentrations and availability in densely populated areas. Lithology, texture and forestation also affected total Pb stocks. With regards to availability, forestation and pH appeared as key factors.

► We map soil Pb stock and availability based on a French monitoring network dataset.
► We use a novel model combining regression trees with robust geostatistics.
► Trees classifications do not require expert knowledge and ease map interpretation.
► Densely populated areas display the highest stocks and high availability.
► Despite low stocks, availability is greatest under forests foremost when acidic.