D-ISN/Collaborative Research: Machine Learning to Improve Detection and Traceability of Forest Products using Stable Isotope Ratio Analysis (SIRA)

University Of Washington, Seattle WA

Investigators

Abstract

The objective of this Disrupting Operations of Illicit Supply Networks (D-ISN) project is to develop new machine learning approaches to help discover and trace illicitly sourced timber products. Specifically, this project leverages Stable Isotope Ratio Analysis (SIRA), a technique that uses the ratios of several elemental stable isotopes within natural products to help trace their geographic origin. By comparing isotope ratios against reference databases built from verified locations, the researchers can impute the origin of suspicious timber products. The project brings together data scientists, analytical chemists, geospatial and remote sensing specialists, and international trade and supply chain experts to develop new data science approaches that will enhance SIRA accuracy and resolution. This project will advance our national ability to counter nefarious and illegal activities by rapidly imputing the source for timber products, helping identify violators of international treaties and regulations, and thus combat natural resource trafficking. The project will develop new machine learning methods to overcome the relative scarcity of labeled data (compared to traditional machine learning applications like computer vision and natural language processing, where datasets might contain millions of labeled examples). Specifically, the PIs will investigate contrastive learning, generative learning, and science-guided machine learning algorithms that can harness prior domain knowledge to combine climate layers with the best available local-scale data, to ensure fidelity to both large-scale patterns and site-specific observations. In addition to location determination from isotope ratios, the project will develop active sampling strategies to “close the loop”, i.e., quantify a model’s uncertainty and determine future sampling regions in order to improve prediction accuracy and resolution. This project is expected to improve geospatial prediction accuracy of product origin and will enable a cost-benefit analysis to minimize future data collection costs and optimize prediction gain. The project will involve partners in industry, non-profit, and government to source samples and to communicate results with relevant enforcement agencies and SIRA analysis labs. The project will also support graduate students who will be exposed to a multi-disciplinary approach to address important societal problems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.