PNNL

Determining the geographic source of agricultural products, such as honey, is important for international trade to help ensure food quality, avoid mislabeling, and enhance economic security. A study by researchers at Pacific Northwest National Laboratory (PNNL) assessed the feasibility of using strontium isotope ratios and an existing machine learning–based model to predict and verify a product’s source—in this case, honey.

“Accurately knowing where an agricultural product is from and assuring authenticity of commodities from a given region is important to prevent intentional mislabeling, where low-value items may be sold at higher prices under false labels for personal economic gain,” said Elizabeth Denis, chemist at PNNL.

The study focused on honey as an example of an agricultural product that can have higher economic value depending on geographic origin or floral sources. Manufacturers or traders may attempt to obscure the geographic origin of a high-value product like honey to avoid costs or to increase a product’s value. The study assessed both the feasibility of using strontium isotope ratios to determine honey provenance and the feasibility of using an existing machine learning–based model to predict the measured strontium isotope ratios in honey.

Specifically, the study measured the strontium isotope ratios of honey samples from the United States, Latvia, and India, and compared the results with published strontium isotope values from surrounding areas and bioavailable strontium isotope values predicted by an existing random forest isoscape model developed by Bataille et al. (2020). To predict bioavailable strontium isotope ratios, the model uses a training dataset consisting of information on plant, soil, and other samples with known strontium isotope ratios. Honey samples were not part of the training dataset. While the model was not originally intended to predict honey isotope values, the PNNL team assessed how well it could.

“Our study was the first to compare measured strontium isotope values in honey to values predicted by a machine learning–based model. We tested the efficacy of this modeling approach for predictive geolocation of an agricultural product, and our study highlights opportunities for improvements and potential applications for the future,” said Megan Nims, chemist at PNNL.

The study proposes that strontium isotopic compositions are appealing for developing a predictive model for the geolocation of honey and other products, which could be used to verify the product’s source. In contrast to older geolocation methods, which are time and cost intensive, predictive frameworks with machine learning could be powerful and cost-effective tools. Additional research on what controls the strontium isotope ratios in honey (e.g., forage crop, water, or dust) could be useful to inform models and strengthen predictions.

The study, “Geographic Source Attribution of Honey by Strontium Isotope Analyses: Latvia and India Measurements Compared to Model Predictions in a Feasibility Study,” was published in the Journal of Food Composition and Analysis. Nims presented this work at the American Chemical Society Fall Meeting in August 2025.