Cadmium in topsoils

Cadmium has one of the strongest adverse effects on all biological processes of plants, animals and humans. Globally, the average Cd concentration in soils is within the range of 0.1–0.5 mgˑkg⁻¹. In agricultural soils, Cd can be mobile and become available for uptake by crops, which becomes a food contamination issue in cases of high Cd concentrations. Using the 21,682 soil samples from the LUCAS soil survey, we aim to estimate the spatial distribution of the concentration of Cd in the European Union (EU) and UK topsoil. Out of the total, 72.6% of the samples have Cd values <0.07 mg kg⁻¹, 21.6% in the range 0.07–1 mg kg⁻¹ and the remaining 5.5% higher than the threshold of 1 mg kg⁻¹, which is generally considered the limit for risk assessment. Applying an ensemble of machine learning models supported by a variety of environmental descriptors, we created maps of Cd distribution at a resolution of 100 m. The ensemble approach included five models and increased the prediction accuracy to R2 of 0.45 (an increase of 0.1 compared to best single model performance). The approach used resulted in a high predictive power for the general Cd distribution, while also identifying hotspots of Cd contamination. The models considered as candidates in building the ensemble were: Cubist regression trees, Ordinary Least Squares (OLS) regression, xgbTrees, Elasticnet regression,Gaussian Process Regression (GPR). In this investigation, we consider a number of variables that could potentially influence the distribution of Cd in the topsoil. Variables may be categorised according to their impact on Cd distribution as following: land cover, climate, topography, geology and anthropogenic sources of Cd (mining, coal combustion, zinc smelters, fertilizers in soils). The high resolution dataset of Cadmium can be download here.

Considering a value of 1 mgˑkg⁻¹ as the risk threshold for definition, we mapped the 1191 points above this threshold.  The point size reflects the cadmium level with 821 LUCAS samples in the range 1–2 mg kg−1. In absolute numbers, Poland, Finland and France have 236, 180 and 170 samples with Cd higher than 1 mgˑkg−1. However, in relative terms, approximately 14 % of the samples in Slovenia and Poland, are higher than 1 mgˑkg−1 while in Ireland >50 % are above this threshold

The models within the learning ensemble were fitted by performing an extensive parameters tune (excluding OLS that has no tuning parameter) using cross-validation in order to avoid model overfitting. After the models fitting, a meta-model based on the stacking of the five candidate models was created. The meta-model takes the estimates of the target property (Cd) and pools them using a machine learning algorithm (in this case GPR was used) in order to obtain a final estimate. It should be noted that this procedure is meaningful only if the meta-model yields better estimates than each single model. In the present study, the ensemble produces more accurate results with an average increase in R2 of about 0.1 over the single models.

The models within the learning ensemble were fitted by performing an extensive parameters tune (excluding OLS The ensemble model was applied on a set of raster dataset in order to predict Cd concentrations for each pixel of the area sampled in the LUCAS 2009 soil survey. The figure below depicts the spatial distribution of Cd concentrations in the topsoil as predicted by the ensemble model. Overall, predicted values range from the detection limit to 2.24 mgˑkg−1. In general, the ensemble does not predict values exceeding this threshold, so values above this limit can be considered as outliers. The spatial distribution of Cd shows an overall relatively low concentration of this element in the EU topsoil with some hotspots. In general, concentrations rarely exceed the 1 mgˑkg−1 threshold. The only exceptions are in some part of eastern Ireland where the geological condition (sedimentary rocks) results in higher than normal values and in Poland where a combination of intense use of coal for power production and fertilization might have increased the natural background level.

Data: You can download the Cd data

The distribution of cadmium in the EU topsoil depends on several factors. Among others, significant sources of natural Cd is rock weathering, volcanic emissions and wildfire ash. Natural factors are related to the background level, due to the mineralogical composition of the soil parent material, the accumulated amount of leaching the soil was subject to and the possible enrichment from external natural sources (volcanic dust). In general, young soils, developed on substrates relatively rich in heavy metals, such as mafic and ultramafic rocks, will have higher concentrations of cadmium. The concentration of cadmium in soils is not depending of a single variable. Overall the most important variables influencing Cd distribution are climate, distance from mines and smelters, distance from coal power plants, soil pH and clay content, surface temperature and vegetation. The Cd concentration in soils is negatively correlated with temperature as it is well known from other studies.

Cadmium content in EU agricultural soils is of concern and the exposure to cadmium hotspots should be further limited. Compared to before the 1990's, cadmium input to soils is declining as both fertilization rates are decreasing while cadmium emissions to the atmosphere are reduced by 60 %. In addition, the cadmium contamination has not the same extent across all Member States or regions. Therefore, the degree of concern differs from Member State to Member State and further studies should focus in the hotspots. The main cadmium input in soils is fertilization and the main output is leaching. The data findings justify the EU Commission proposal for setting lower Cd limits in phosphorus fertilisers. Particular focus should be given in hotspots with Cadmium higher than 1 mg kg−1 and agricultural areas with intense fertilization.

Reference: Ballabio, C., Jones, A. and Panagos, P., 2024. Cadmium in topsoils of the European Union–An analysis based on LUCAS topsoil database. Science of The Total Environment, 912: 168710.