Sie befinden Sich nicht im Netzwerk der Universität Paderborn. Der Zugriff auf elektronische Ressourcen ist gegebenenfalls nur via VPN oder Shibboleth (DFN-AAI) möglich. mehr Informationen...
•Soil arsenic level estimation by pXRF is promising.•Mapping soil arsenic levels requires stochastic simulation techniques.•Feature selection can improve soil arsenic estimation.•Multivariate methods are essential in soil science.
Rapid, inexpensive, and equally reliable estimates of potentially toxic elements are a necessity; portable X-ray fluorescence (pXRF) spectrometry is a handy tool to help achieve such. The current study sought to compare multiple linear regression with three regularized regression models [Ridge, Lasso, and ElasticNet (ENET)] for the estimation of total arsenic (As) using pXRF datasets in polluted temperate floodplain soils of Příbram, Czech Republic. A total of 158 surface (0–25 cm) floodplain surface soil samples were collected from a specific site in Příbram. Models were evaluated separately and compared based on mean absolute error (MAE), root mean squared error (RMSE) and the coefficient of determination (R2). All four models were able to predict As with good accuracy (MAE and RMSE values of 0.02 and 0.03, respectively, and R2 values ranging from 0.94 to 0.95). As measured via pXRF as well as predicted via the four regression models produced similar spatial variability as shown by the standard laboratory-measured As using ordinary kriging and Conditional Gaussian Simulations (CGS), although the latter produced more details of As spatial distribution in floodplain soils. Future research should include other auxiliary predictors (e.g., soil physicochemical properties, other various sensor data) as well as cover a wider range of soils to improve model robustness.