Evaluating geostatistical methods along with semi-destructive analysis for forensic provenancing organic-rich soils in humid subtropical climate
Matheus P N E Silva,Carlos C F Guedes,Vander de F Melo,Ricardo de O Mascarenhas,Fábio A S Salvador
DOI: https://doi.org/10.1016/j.forsciint.2022.111508
Abstract:When no samples are available for direct comparisons during a criminal investigation, forensic scientists must resort to georeferenced soil databases in order to find the source of a single questioned evidence. To this end, several authors addressed many methods to infer the origin of soil samples, such as establishing search range intervals or defining statistical similarities. However, little is currently known about the efficiency of these methods when it comes to organic-rich and deep weathered subtropical soils. Therefore, this study attempts to contribute to this subject by evaluating the predictability of a soil database built in a 100 km² area in the Curitiba Metropolitan Region, Brazil, where 232 topsoil (<5 cm), 14 validation, and 38 subsoil samples (B horizon) were collected. To determine their physical, chemical, and colorimetric features, samples were subjected to magnetic susceptibility analysis, gamma-ray spectrometry, color analysis, pXRF, and ATR-FTIR. Two prediction methods were selected: search range intervals (SR) and n-dimensional Euclidean distances (ED). After the descriptive and multivariate statistical analyses, geostatistical models were generated for the 28 obtained variables using the empirical Bayesian kriging (EBK) with regression prediction method, where geological and airborne geophysical data were used as explanatory matrix. For the ED provenancing method, dissimilarity values were calculated and interpolated by applying the inverse distance weighting (IDW) approach. Findings, while preliminary, suggest that magnetic susceptibility, lightness, Fe, Mn, Cu, Ni, and the light elements group (LE) may be the most significant variables to discriminate parent rock and to correlate with subsurface data. The SR method managed to correctly predict the origin of 4 of 14 validation samples, reducing up to 97.5% of the original area. The ED method, on the other hand, accurately estimated the provenance of 8 out of 14 samples, excluding up to 63.4% of the area. Despite the limitations and errors inherent to semi-destructive analyses, the moderate chemical weathering, and the high organic matter content, overall soil provenancing results were suitable, demonstrating that spatial variability of topsoils developed under humid subtropical environments can be modeled and used in forensic contexts, as long as the right methods are applied.