A probabilistic approach to estimating timber harvest location

Abstract

Determining the harvest location of timber is crucial to enforcing international regulations designed to protect natural resources and to tackle illegal logging and associated trade in forest products. Stable isotope ratio analysis (SIRA) can be used to verify claims of timber harvest location by matching levels of naturally occurring stable isotopes within wood tissue to location-specific ratios predicted from reference data (“isoscapes”). However, overly simple models for predicting isoscapes have so far limited the confidence in derived predictions of timber provenance. In addition, most use cases have limited themselves to differentiating between a small number of predetermined location options. Here, we present a new analytic pipeline for SIRA data, designed to predict the harvest location of a wood sample in a continuous, arbitrarily large area. We use Gaussian processes to robustly estimate isoscapes from reference wood samples, and overlay with species distribution data to compute, for every pixel in the study area, the probability of it being the harvest location of the examined timber. This is the first time, to our knowledge, that this approach is applied to determining timber provenance, providing probabilistic results rather than a binary outcome. Additionally, we include an active learning tool to identify locations from which additional reference data would maximize the improvement to model performance, allowing for optimisation of subsequent field efforts. We demonstrate our approach on a set of SIRA data from seven oak species in the United States as a proof of concept. Our method can determine the harvest location up to within 520 km from the true origin of the sample and outperforms the state-of-the-art approach. Incorporating species distribution data improves accuracy by up to 36%. The future sampling locations proposed by our tool decrease the variance of resultant isoscapes by up to 86% more than sampling the same number of locations at random. Accurate prediction of harvest location has the potential to transform worldwide efforts to enforce anti-deforestation legislation and protect natural resources.