Dielectric spectroscopy for evaluating dry matter content of potato tubers

• Dry matter prediction from dielectric response of potato tubers is examined.
• Origin of prediction errors is investigated with agar-gel model systems.
• Predictions for uniform model systems was possible with high precision.
• Spatial inhomogeneous dry matter was the major source of errors specific for potatoes.
• Variation in chemical composition and air-space tissue was other possible sources.

The present study investigated the application of dielectric spectroscopy as a method for evaluating the dry matter content of potato tubers. Sample specific factors determining the precision of this application were investigated by studying the prediction of the dry material content in agar gel based model systems with known dry matter content. Dielectric spectra were measured with a large custom-made open-ended coaxial probe in the frequency interval from 0.01 GHz to 3 GHz. Both univariate linear models motivated by a two phase mixture model and cross-validated multivariate partial least squares regression (PLSR) models were applied to predict the dry matter content.Results showed that the PLSR models gave markedly better prediction of the dry matter content from the dielectric response in most of the investigated systems compared to the univariate linear models. The highest precision of the predicted dry matter content was observed in chemically and spatially uniform systems, with a root mean square error (RMSE) of the predicted dry-matter content of 0.64 percentage points observed in agar gels containing refined potato starch. A marked decrease in precision is observed in model systems which include chemical variations between potato tuber samples. The added dry material content was predicted with a RMSE of 0.94 percentage points in agar gels with added dried material extracted from separate potato tubers. The local dry matter content from a region within 2 cm of the center location of the dielectric measurement on potato tubers was predicted with a RMSE of 1.21 percentage points. The dry matter content of total potato tubers were predicted with a RMSE of 1.93 percentage points. The difference between the precision of the local and total potato tuber dry matter prediction indicated that spatially non-uniform dry matter distributions were the largest contribution to the observed prediction errors. Variations in chemical composition and air space tissue was also found to be possible factors contributing to the prediction error for potato tubers.