Skip to main content
ARS Home » Midwest Area » Madison, Wisconsin » Vegetable Crops Research » Research » Publications at this Location » Publication #327723
Title: Evaluation of hyperspectral reflectance for estimating dry matter and sugar concentration in processing potatoes

Author
item FREDERICK, CURTIS - University Of Wisconsin
item COUTURE, JOHN - University Of Wisconsin
item Bethke, Paul

Submitted to: American Journal of Potato Research
Publication Type: Proceedings
Publication Acceptance Date: 3/29/2016
Publication Date: 6/1/2017
Citation: Frederick, C., Couture, J., Bethke, P.C. 2017. Evaluation of hyperspectral reflectance for estimating dry matter and sugar concentration in processing potatoes. American Journal of Potato Research. 94(3):211-250. doi: 10.1007/s12230-017-9581-5.

Interpretive Summary:

Technical Abstract: The measurement of sugar concentration and dry matter in processing potatoes is a time and resource intensive activity, cannot be performed in the field, and does not easily measure within tuber variation. A proposed method to improve the phenotyping of processing potatoes is to employ hyperspectral radiometry for the estimation of these traits in fresh potato tissue. This technology is already in use in many crop breeding and research programs for traits that require labor-intensive wet-chemistry analysis. In order to implement this technology, a calibration equation relating the trait values to the reflectance values of the hyperspectral radiation must be developed on a diverse set of potato genotypes. In addition, optimizing the method used to collect spectral samples can often increase the accuracy of trait estimation. We developed and analyzed computational models for estimating dry matter and sugar content on tubers of 96 individuals from a bi-parental population. Each individual had spectral and physical samples taken from tissues including the face of a tuber cut in half, a 1.1 mm chip slice, and a 10 mm slab. Models were constructed between spectral samples and physical trait values. Each data set combination was cross-validated 70:30 over 500 iterations of a Partial Least Squares regression model. Our results indicate that dry matter is best estimated out of these traits with a 1.1 mm chip slice sampling procedure. The mean validation R2 was 0.68 and the mean validation RMSE was < .1% of mean dry matter. Sugar estimations were less accurate but managed reasonable prediction accuracy for simple genotype selection. These results indicate that this could be an accurate tool for rapidly measuring within tuber dry matter content in field, storage, or lab settings.