Comparison of machine learning approaches and statistical methods for estimation of airborne chemical exposures

Accurate quantitative estimation of exposure using retrospective data has been one of the most challenging tasks in the exposure assessment field. To improve these estimates, some models have been developed using published exposure databases with their corresponding exposure determinants. These models are designed to be applied to reported exposure determinants obtained from study subjects or exposure levels assigned by an industrial hygienist, so quantitative exposure estimates can be obtained. ^ In an effort to improve the prediction accuracy and generalizability of these models, and taking into account that the limitations encountered in previous studies might be due to limitations in the applicability of traditional statistical methods and concepts, the use of computer science- derived data analysis methods, predominantly machine learning approaches, were proposed and explored in this study. ^ The goal of this study was to develop a set of models using decision trees/ensemble and neural networks methods to predict occupational outcomes based on literature-derived databases, and compare, using cross-validation and data splitting techniques, the resulting prediction capacity to that of traditional regression models. Two cases were addressed: the categorical case, where the exposure level was measured as an exposure rating following the American Industrial Hygiene Association guidelines and the continuous case, where the result of the exposure is expressed as a concentration value. Previously developed literature-based exposure databases for 1,1,1 trichloroethane, methylene dichloride and, trichloroethylene were used. ^ When compared to regression estimations, results showed better accuracy of decision trees/ensemble techniques for the categorical case while neural networks were better for estimation of continuous exposure values. Overrepresentation of classes and overfitting were the main causes for poor neural network performance and accuracy. Estimations based on literature-based databases using machine learning techniques might provide an advantage when they are applied to other methodologies that combine `expert inputs' with current exposure measurements, like the Bayesian Decision Analysis tool. The use of machine learning techniques to more accurately estimate exposures from literature-based exposure databases might represent the starting point for the independence from the expert judgment.^

Geochemical composition of surface snow, ice and debris from glaciers in Norway, Nepal and New Zealand

Alpine glacier samples were collected in four contrasting regions to measure supraglacial dust and debris geochemical composition. A total of 70 surface glacier ice, snow and debris samples were collected in 2009 and 2010 in Svalbard, Norway, Nepal and New Zealand. Trace elemental abundances in snow and ice samples were measured via inductively coupled plasma mass spectrometry (ICP-MS). Supraglacial debris mineral, bulk oxide and trace element composition were determined via X-ray diffraction (XRD) and X-ray fluorescence spectroscopy (XRF). A total of 45 elements and 10 oxide compound abundances are reported. The uniform data collection procedure, analytical measurement methods and geochemical comparison techniques are used to evaluate supraglacial dust and debris composition variability in the contrasting glacier study regions. Elemental abundances revealed sea salt aerosol and metal enrichment in Svalbard, low levels of crustal dust and marine influences to southern Norway, high crustal dust and anthropogenic enrichment in the Khumbu Himalayas, and sulfur and metals attributed to quiescent degassing and volcanic activity in northern New Zealand. Rare earth element and Al/Ti elemental ratios demonstrated distinct provenance of particulates in each study region. Ca/S elemental ratio data showed seasonal denudation in Svalbard and Norway. Ablation season atmospheric particulate transport trajectories were mapped in each of the study regions and suggest provenance pathways. The in situ data presented provides first order glacier surface geochemical variability as measured from four diverse alpine glacier regions. This geochemical surface glacier data is relevant to glaciologic ablation rate understanding as well as satellite atmospheric and land-surface mapping techniques currently in development.

Pleistocene and upper Pliocene sapropels in the Tyrrhenian Sea, Mediterranean Sea

A most significant finding of the ODP Leg 107 drilling campaign was the recovery of at least 56 distinct sapropel intervals in upper Pliocene to Pleistocene sediments of six sites drilled in the Tyrrhenian Sea. Except for 3 repots of disturbed organic-rich sediments - recovered in Core 201 of the Swedish Deep-Sea Expedition, in Core 2R-1,107 cm of Site 373 (Leg 13 DSDP) and at Site 373, Core 1-2,O-5 cm of DSDP Leg 42A - sapropels had previously only been described from the eastern Mediterranean and the Black Sea. Scientific deep-sea drilling in the Tyrrhenian Sea during DSDP Legs 13 and 42A apparently missed most of these deposits due to spot coring and rotary drilling techniques; high sedimentation rates may have precluded recovery by conventional gravity coring devices. The recovery of multiple layers of sapropels and sapropelic sediments in the Tyrrhenian Sea demonstrates that oceanographic conditions conducive to sapropel formation were not confined to the Black Sea and eastern Mediterranean, but occurred sporadically and possibly simultaneously in the entire Mediterranean during the Pliocene and Pleistocene. In the light of this finding, previous models of sapropel genesis may need reconsideration. In this paper, we present some initial data on the Tyrrhenian sapropels and suggest some implications of their massive occurrence in the western Mediterranean realm. We end by outlining possible causes for deposition of sapropels in an attempt to revive the interest in sapropels and their paleoceanographic significance.

Grain size and XRD analysis on Site 162-983

The grain size of deep-sea sediments provides an apparently simple proxy for current speed. However, grain size-based proxies may be ambiguous when the size distribution reflects a combination of processes, with current sorting only one of them. In particular, such sediment mixing hinders reconstruction of deep circulation changes associated with ice-rafting events in the glacial North Atlantic because variable ice-rafted detritus (IRD) input may falsely suggest current speed changes. Inverse modeling has been suggested as a way to overcome this problem. However, this approach requires high-precision size measurements that register small changes in the size distribution. Here we show that such data can be obtained using electrosensing and laser diffraction techniques, despite issues previously raised on the low precision of electrosensing methods and potential grain shape effects on laser diffraction. Down-core size patterns obtained from a sediment core from the North Atlantic are similar for both techniques, reinforcing the conclusion that both techniques yield comparable results. However, IRD input leads to a coarsening that spuriously suggests faster current speed. We show that this IRD influence can be accounted for using inverse modeling as long as wide size spectra are taken into account. This yields current speed variations that are in agreement with other proxies. Our experiments thus show that for current speed reconstruction, the choice of instrument is subordinate to a proper recognition of the various processes that determine the size distribution and that by using inverse modeling meaningful current speed reconstructions can be obtained from mixed sediments.