Assessment of Statistical Methods Applied to Geochemical Data for Paleoenvironmental Interpretation: A Case Study from the Al-Azraq Basin, Jordan

Abstract

This study investigates the appropriateness of statistical methods frequently applied to inorganic geochemistry data for indicating past environments. The data examined are two large geochemical data sets of cored lacustrine sediments from the Al-Azraq basin, Jordan. Lacustrine sediments contain valuable records of paleoenvironmental conditions, recording local and regional responses to environmental chang¬¬e. The Al-Azraq basin is a low lying an endorheic basin draining an area across eastern Jordan, southern Syria and northern western Saudi Arabia of approximately 12,700 km2. The basin is bounded to the west and east by silt dunes, and to the northwest by basaltic hills. Two cores, AZ1 and AZ2, are located on the eastern and western margins of the Al-Azraq Basin, respectively. AZ1 is located to the west of basalt fed springs and on the eastern edge of the centrally located oasis. AZ2 is located near the western margin where six major wadis enter the basin. Inorganic geochemical data generated by inductively coupled atomic emission spectroscopy (ICP-AES) show significant changes in elemental concentrations with depth throughout each sediment core. AZ1 cored sediments are described in previous work (Ahmad, 2013). ICP-AES data from the AZ2 sediments indicated periods of humidity and aridity, increased and decreased precipitation, and low and high chemical weathering. This study contributes to understanding the paleoenvironments of the Al-Azraq basin during the mid-Pleistocene period. Methods commonly used with geochemical data included data transformation, data standardization, outlier detection, and replacement. Both graphical and statistical methods examined data normality to assess the assumptions of the parametric requirement of normality, and determine the use of parametric or non-parametric statistical analyses. Both non-robust and robust estimates of location and scale, of which parametric and non-parametric statistical methods are based, respectively, are studied with geochemical data based on Rock (1988). Multivariate statistical methods included correlation analysis, principal components analysis, and hierarchical cluster analysis. Ahrens (1953), Rock (1988) and Reimann and Filzmoser (2000) stated geochemical data is commonly not normally or lognormally distributed. Overall the majority of geochemical elements in Al-Azraq sediment cores AZ1 and AZ2 are neither normal nor lognormal. Different graphical and statistical methods to determine data distribution largely yielded slightly different results. Neglecting that geochemical data are neither normal nor lognormally distributed may lead to biased or erroneous results with the use of parametric statistical methods. Non-robust estimates of location and scale used with AZ1 and AZ2 geochemical data corresponded to higher values and less consistent values compared to their robust counterparts. The results of parametric statistical methods with non-normally distributed data may yield geologically interpretable results; however, these methods are erroneous. The use of non-parametric statistical methods or the proper data standardization and transformation methods prior to the use of parametric statistical methods should be taken into consideration, as this may yield a better interpretation of the data.