Finding a Way to Predict How Good a Cup o’ Joe Will Be

Lynn M. Savage

The difference between a great-tasting cup of coffee and a bland one depends on many factors, including the coffee plant species, the steps taken to produce the plants and, perhaps primarily, the roasting process. During roasting, coffee beans undergo a large number of complex reactions that ultimately shape the drink’s sensory qualities — taste and smell. Caffeine content also can be used to judge the quality of the beans, although it has little effect on the final flavor.

Researchers have provided the means to develop near-IR studies the effects of roasting and caffeine content on the quality of coffee beans.

Colorimetry can be used to gauge the extent to which beans have been roasted, and spectroscopy can be used to ascertain the chemical compounds residing within them. Unfortunately, little research has been performed to determine which processes best predict roasting color or caffeine content in any given group of coffee beans. However, investigators at the University of La Rioja in Spain and at Università di Genova in Italy have developed a strategy for modeling and predicting these attributes, which one day may lead to effective analysis of coffee beans on the production line.

The investigators used a near-IR spectrophotometer from Foss NIRSystems Inc. of Laurel, Md., and a colorimeter from Dr. Bruno Lange GmbH & Co. KG of Düsseldorf, Germany, respectively, to obtain roasting color and caffeine content data. They tested 83 samples of Arabica and Robusta beans of various origins both separately and in 108 blends of three types each. Roasting times varied from 12 to 20 min at 200 to 230 °C; roasting color corresponds directly with the effect of these parameters.

The researchers split the complete data set into two subsets: a calibration set with 115 samples and a test set with 76 samples. They constructed a partial least squares model for each roasted coffee parameter using the 1100- to 2200-nm-wavelength range, then tested several variable-selection techniques by using each one to select a subset of significant near-IR absorption bands. They used these bands, in turn, to develop a simplified regression model for predicting roasting color and caffeine content.

They found that one of the selection techniques — called stepwise orthogonalization of predictors, or SELECT — provided the best regression model for analyzing coffee samples because it derives a minimum number of informative predictors from the near-IR data.

The scientists note that their study does not provide a definitive near-IR calibration model for monitoring roasting color and caffeine content. They believe, however, that their strategy will provide other investigators with the tools to develop a more exhaustive and specific collection of calibration samples.

Journal of Agricultural and Food Chemistry, Sept. 5, 2007, pp. 7477-7488.