Network Optimization-based Approach for Identification of Illegal Trade in the Global Timber Supply Chain

Abstract

Forest ecosystems play a crucial role in the global carbon cycle and provide numerous regional environmental and economic services. Thus, it is essential to limit their degradation due to human exploitation and the risk of climate change. To effectively regulate the production and trade of timber and to ensure that ambitious sustainability and legality targets are met, enforcement agencies need to reliably monitor the flows of timber products circulating in the global supply chain. However, available reported data on the trade of timber-based products at the global level is typically subject to a range of irregularities, including misreported and inconsistent data. Current estimates of these irregularities -- particularly, illegal trade -- analyze reported flows at the level of individual trade partners and do not account for the structure of global trade network. In this thesis, we attempt to address these limitations by imposing nodal volume balance across the entire network and develop a general framework to identify multiple link-level irregularities in trade data.

Specifically, we present an optimization-based approach to model and identify data irregularities in the global timber supply chain. We evaluate the ability of this approach to recover flows of timber products volumes from perturbations to reported data on multiple links -- this is called the reconstruction problem -- and identify flows that are most likely to involve irregularities -- which is called the identification problem. These reconstruction and identification tasks essentially rely on the use of network optimization techniques. In this context, we explore both classic optimization formulations and matrix scaling-based algorithms. We extend the well-known formulation of matrix scaling algorithms to include prior knowledge of the reliability of the data. We propose a link-specific weighted iterative scaling algorithm (WIS) and a node-specific weighted iterative scaling algorithm (NSWIS). In doing so, we extend the current literature on matrix scaling algorithms by expanding the scope of their practical application to supply chain data correction problems.

For the type of perturbations studied in the evaluation procedure, the WIS algorithm shows a strong ability to correctly reconstruct data, even under limited prior information on data reliability. Moreover, the combinations of the WIS with threshold-based identification models obtain satisfactory results on the identification phase (True Positive Rate of more than 75% for a False Positive Rate of less than 30%) even under limited prior information on data reliability. We evaluate the reconstruction and identification performance of the RIMs both on synthetic and real data. Our results support the relevance of a principled approach to network flow modeling and optimization for correcting and identifying irregularities in timber trade and timber production data.