Predictive geographical authentication of green tea with protected designation of origin using a random forest model

Reliable origin authentication methods are critical for protecting high-value food products with designated geographical origins. A total of 623 tea samples were collected from important green tea production regions around China from 2012 to 2016. A Random Forest model (RF) with 19 input predictors (e.g., delta C-13, Mg-24, Rb-85, and Pb-206/Pb-207) was developed. Our RF model not only discriminated Westlake Xihu Longjing green tea (XHLJ) from other regions with an accuracy of 97.6%, but also correctly identified green tea from surrounding regions with an accuracy of 97.9%. The geographical discrimination of tea subsequently harvested in the following years also showed good reliability. Predictive accuracies were higher than 91%. Rb-85, Mg-24, delta C-13 and K-39 were the most important geographical proxies for determining geographical origin of tea with a relative contribution of 20.6%, 12.5%, 12.1% and 7.4%, respectively. This RF model showed higher classification accuracy than other commonly used chemometrics models and provides a new insight into the use of predictive models utilizing historical data for geographical authentication of agricultural products with Protected Designation of Origin (PDO).