Assessing and mapping vineyard water status variability using a miniaturized nir spectrophotometer from a moving vehicle
Abstract
Context and purpose of the study – In the actual scenario of climate change, optimization of water usage is becoming critical in sustainable viticulture. Most of the current approaches to assess grapevine water status and drive irrigation scheduling are either destructive, time and labour consuming and monitor a small, limited number of plants. This work presents a novel methodology using a contactless, miniaturized, low-cost NIR spectrometer to monitor the vineyard water status variability from a moving vehicle, to provide reliable information towards precision irrigation.
Material and methods – Spectral measurements were acquired using a NIR micro spectrometer, operating in the 900–1900 nm range, from a ground vehicle moving at 3 km/h. Spectra acquisition was carried out on the northeast side of the canopy across six dates in 2021 season and five dates in 2022, in two VSP commercial vineyards of Vitis vinifera L. Tempranillo and Graciano in the Rioja Appellation Board (Spain). Grapevines were monitored at solar noon using stem water potential (Ψs) as reference indicator of plant water status. At each date, 36 and 27 measurements of Ψs were taken in the Tempranillo and Graciano vineyards, making a total of 396 and 297 data respectively. Partial least squares (PLS) regression and the Variable Importance in the Projection (VIP) method were used to build calibration and prediction models using the pooled data from the two seasons for each variety. Multiple Linear Regression (MLR) was also applied to build simplified estimation models using 8 and 10 spectral bands with the highest VIP scores (always >1). Determination of coefficient (R2) and root mean square error (RMSE) were computed to assess model performance.
Results – Remarkable cross-validation models were built using the whole spectrum (117 wavelengths) with R2cv ranging from 0.62 to 0.80, and RMSECV between 0.115-0.138 MPa in Tempranillo and Graciano vineyards, respectively. With the aim of simplifying model building, the 8 and 10 spectral bands showing the highest VIP scores, with values above 1 in all instances, were selected to build MLR cross validation models of stem water potential. In both varieties MLR8 and MLR10 (MLR models built with 8 and 10 wavelenghts only respectively) yielded R2cv ranging from 0.45-0.59 and RMSECV ~ 0.156-0.171 MPa. Although lower performance was achieved with the simplified models they could still be utilized to classify and map the vineyard plots into three different water status zones, susceptible of precise, differentiated irrigation.
DOI:
Issue: GiESCO 2023
Type: Article
Authors
1Department of Agriculture and Food Science, University of La Rioja, Madre de Dios 53, 26007 Logroño, Spain
2Instituto de Ciencias de la Vid y del Vino (Universidad de La Rioja, CSIC, Gobierno de La Rioja) Finca La Grajera, Ctra. Burgos Km 13, 26007 Logroño, Spain
Contact the author*
Keywords
water stress, stem water potential, proximal sensing, partial least squares, multiple linear regression