Fig. 1.

The study area in the north of France with the land-use classes.

 


Fig. 2.

The PCB-187 concentration data from the 105 sample locations.

 


Fig. 3.

The potential covariates for explaining and mapping the spatial distribution of PCB-187 concentrations: (a) measured soil organic carbon (SOC), (b) predicted SOC from model of Jones et al. (2005), (c) predicted SOC from model of Meersmans et al. (2012), (d) elevation, (e) land-use classes from CLC database, (f) observed land-use classes at RMQS data locations, (g) population density, and (h) road density.

 


Fig. 4.

Optimized log-likelihood with population density and road density as covariates, defined as the average of the population densities (per km2) within a radius of k km.

 


Fig. 5.

Predicted PCB-187 concentrations across the study area from (a) imputed data approach with constant mean (IO), (b) censored data approach with constant mean (CO), (c) imputed data with {XM, κM}, the Akaike Information Criterion (AIC)–selected fixed effects for mapping (IM), and (d) censored data with AIC-selected fixed effects (CM). Predictions in urban areas are masked.

 


Fig. 6.

Estimated probability of PCB-187 concentration >2 μg kg−1 from (a) imputed data approach with constant mean (IO), (b) censored data approach with constant mean (CO), (c) imputed data with {XM, κM}, the Akaike Information Criterion (AIC)–selected fixed effects for mapping (IM), and (d) censored data with AIC-selected fixed effects (CM). Predictions in urban areas are masked.

 


Fig. 7.

Mean of the (a, c) logarithmic loss function (MLLF) and (b, d) quadratic loss function (MQLF) from cross-validation, plotted against the concentration threshold, zT (μg kg−1). All plots show results from S: simple nongeostatistical method. (a) and (b) compare methods built on imputed (IM) and censored (CM) data with fixed effects selected for mapping (population density). (c) and (d) use the censored data approach to compare the fixed effects functions: no fixed effects (CO), fixed effects selected for mapping (population density, CM) and fixed effects selected for explaining PCB data (population density and soil organic carbon, CE).