Estimation of degree brix in grapes by proximal hyperspectral sensing and nanosatellite imagery through the random forest regressor
Abstract
Context and purpose of the study – The assessment of physiological parameters in vineyards can be done by direct measurements or by remote, indirect methods. The latter option frequently yields useful data, and development of methods and techniques that make them possible is worthwhile. One of the parameters most looked for to define the quality status of a vineyard is the degree Brix of its grapes, a quantity usually determined by direct measurement. However, other ways may be possible, and presently Brix estimations in vineyards using as data sources field radiometry, localized Brix measurements and satellite imagery are reported.
Material and methods – The investigation was developed in a commercial vineyard in south Brazil at two stages of the 2017/2018 vegetative cycle. Brix degree was measured twice: using a spectroradiometer which measured reflectance from 350nm to 2500nm, and a refractometer. Brix estimates were derived using a machine learning model, the Random Forest Regression (RFR) algorithm, applied on data from images of PlanetScope satellites.
Results – Results produced coefficients of correlation between observed and predicted degrees Brix as high as 0.89. Analysis of an importance parameter, the Gini index, suggested that spectral data at ultraviolet, visible, and near-infrared wavelengths and the vegetation indices TGI and NDVI are the most important variables used for the predictive model. This methodology is potentially useful for the derivation of vineyard quality parameters at situations when specific vineyard conditions, as rugged terrain and large variations in soils, turn direct measurements a difficult task.
DOI:
Issue: GiESCO 2023
Type: Poster
Authors
Remote Sensing Center, Universidade Federal do Rio Grande do Sul, Av. Bento Goncalves 9500, CEP 91501-970, Porto Alegre RS, Brazil
Contact the author*
Keywords
degree Brix, hyperspectral data, Random Forest Regression