Ecological risk source distribution, uncertainty analysis, and application of geographically weighted regression cokriging for prediction of potentially toxic elements in agricultural soils

A resilient environment is essential for society's long-term viability. Receptor models have evolved into an excellent tool for detecting pollution sources and evaluating each source's empirical contributions based on ecological datasets. One hundred and fifteen soil sample were collected from the district of Frydek Mistek in the Czech Republic and the concentration of arsenic (As), cadmium (Cd), copper (Cu), chromium (Cr), manganese (Mn), nickel (Ni), lead (Pb)and zinc (Zn) measured inductively coupled plasma-optical emission spectrometry. The results suggested that the hybridized receptor models ER-PMF and PMF identified the following geogenic, steel industries, vehicular traffic, and agro-based activities such as pesticide and fertilizer applications as the primary sources in the source distribution. The ER-PMF source pollution identification efficiency ranged from R2 0.872-0.970, RMSE 0.128-17.344 and MAE 0.085-10.388, whereas the PMF R2 ranged from 0.883 to 0.960, RMSE 0.246-79.003 and MAE 0.145-49.925. The overall assessment of the efficiency of the receptor models suggests that the ER-PMF appears to yield more efficient results in pollution source identification compared to PMF. The PTEs mapping using geographical weighted regression (GWR) and a hybridized regression approach, geographical weighted regression cokriging (GWRCoK), revealed that GWRCoK had a higher goodness of fit in the spatial prediction maps than GWR. According to Hakanson's risk index classification, the ecological risk level in the study area was moderate to high (risk level = 51 observed locations out of 115, or 44.35%); however, Chen's risk index reclassification indicated that the toxicity level in the study area was moderate to extremely high (risk level = 113 observed locations out of 115, or 98.26%). However, the uncertainty assessment results indicated that the DISP interval ratio of the hybridized ER-PMF model was lower than that of the parent PMF model. However, it was clear that the random error that could occur in the DISP based on the DISP interval ratio was likely to be lower in the ER-PMF receptor model than in the parent model. The assessment of PTEs in soil has been widely published, but this study recommends using a pollution assessment-based receptor model (ER-PMF), which has been shown to be reliable and practical in estimating distribution sources.