Containing words
NIR Spectroscopy to Map Quality Parameters of Sugarcane
M. N. Ferraz, J. P. Molin
University of São Paulo

Precision Agriculture aims to explore the potential of each crop considering the differences within the field. One information that is considered the most important is the yield or the obtained income in the field. However, in the case of sugarcane, quality will also directly influence farmer’s income. Several studies suggest harvester automation aiming to monitor yield, but few consider the quality analysis in the process. Among the existing methods for measuring sugar content the one that is best suited to this need is the spectroscopy. Therefore, it is necessary to assess the feasibility of its use and to investigate the spatial variability of sugarcane quality attributes in a field. Georeferenced samples were obtained from a 16.5 ha field for quality analysis (total solid content, sucrose content, fiber content, purity and total recoverable sugar) and the results were correlated with spectra from NIR (near-infra-red) region obtained using two different field spectrophotometers: Veris Vis-NIR Spectrophotometer (Veris Technologies, Inc., Salina, KS, EUA), and the AgriNir (Dinamica Generale, Pogio Rusco, Italy). The spectra were obtained in both juice and fibrate sample forms. For the correlations, multivariate analyzes using Partial Least Square Regression and leave-one-out cross validation were used. Maps were made using laboratory results and predictions made by the spectral measurements. It was found spatial dependence among samples; the variograms suggest that two samples per ha would be enough to map these parameters in the field.  A coefficient of variation of 5% was obtained in the field, which could justify local management actions. Spectra from fibrate samples were better correlated with sugar content than those from juice samples. The R² obtained for total recoverable sugar were 0.22 for juice samples, 0.42 for fibrate samples using Veris spectrophotometer and 0.65 for fibrate samples using AgriNir. Overall, spectrometry showed good potential to predict quality and was effective in mapping regions with different levels of quality attributes over the field. The correlations  between  the normalized maps of the predictions and the laboratory results for total recoverable sugar was 0.64 for juice samples,0.68 for fibrate samples using Veris spectrophotometer and 0.83 for fibrate samples using AgriNir.

Keyword: Precision agriculture; Sugarcane quality; Geostatistic; Multivariate analysis