Incorporating Raw Signal Processing and Machine-Learning Techniques for Chemical Fate and Transport Investigations, Chemical Threat Detection and Environmental Forensics
Abstract
Approved for public release The big idea in this multi-year multi-institution interdisciplinary effort is to create a dynamic, scalable and science-interpretable database of soil signatures with different types of contaminants undergoing different degrees of degradation, either by photo-oxidation or due to other environmental weathering factors (e.g. microbial degradation and hydrolysis). This will allow on-the-field measurements to be matched against an existing and growing library of soil samples and use signal processing and machine cognition to access, interpret and recognize patterns of known contaminants in the in-field samples. Accordingly, we will harness powerful signal processing techniques (to separate contaminant signatures and peaks within the raw instrument signal) and machine cognition with domain knowledge in environmental chemistry. Specific learning algorithms will be chosen from a suite of suitable architectures as appropriate to the type of soil signatures being considered. The scientific goal will be to gain enhanced understanding of fate and transport of toxic complex mixtures that will render such a dynamic soil signature repository possible. A related goal is to enable early discovery, forensics and surveillance of chemical threats in the environment. In particular, we focus on studying degradation of toxic chemicals by photooxidation, with particular attention to the fate of complex mixtures in different soils exposed to sunlight under a variety of circumstances as well as the changes in signals of the soil overtime when exposed to sunlight. The scientific as well as technical novelty of the project lies in our interdisciplinary ambition to quantify, beyond a few target chemicals, how hundreds, if not thousands, of known and unknown compounds undergo photooxidation and other forms of environmental degradation in a compound-cognizant expert-interpretable setting. This will be achieved by sophisticated signal processing techniques which are markedly different from traditional chemometrics, e.g. principal component, correlation and regression-based analysis, that study statistical trends without the means to offer interpretation at the level of individual compounds. A further novel aspect is to quantitatively disambiguate, across the full raw instrument signal at large, how environmental degradation of the soil itself is separable from that of the complex mixture in the soil. This will be achieved by novel graph-based signal separation and machine learning algorithms that will learn the degradation profiles of different types of soil and toxic mixtures from systemic laboratory studies as well as experimental field data. Our algorithms will be deployable at several instrument settings (e.g. full-scan vs. selective modes) as well as across different types of analytical hardware. All computational algorithms will be provided to the academic and military community at large on an open-source basis. This will allow longer-term and larger scale independent validation of our methods above and beyond the scope of the proposed basic research effort.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 22, 2022
- Source ID
- W911NF2210272
Entities
People
- Ananya Sen Gupta
Organizations
- Army Contracting Command
- United States Army
- University of Iowa