Assessing Grapevine Nutrient Status from Unmanned Aerial System (UAS) Hyperspectral Imagery

Robert Chancia,Terry Bates,Jan Van Aardt,Justine Vanden Heuvel

doi:10.3390/rs13214489

Robert Chancia, Terry Bates + Show 2 more

Open Access

https://doi.org/10.3390/rs13214489

Copy DOI

Abstract

This study aimed to identify the optimal sets of spectral bands for monitoring multiple grapevine nutrients in vineyards. We used spectral data spanning 400–2500 nm and leaf samples from 100 Concord grapevine canopies, lab-analyzed for six key nutrient values, to select the optimal bands for the nutrient regression models. The canopy spectral data were obtained with unmanned aerial systems (UAS), using push-broom imaging spectrometers (hyperspectral sensors). The novel use of UAS-based hyperspectral imagery to assess the grapevine nutrient status fills the gap between in situ spectral sampling and UAS-based multispectral imaging, avoiding their inherent trade-offs between spatial and spectral resolution. We found that an ensemble feature ranking method, utilizing six different machine learning feature selection methods, produced similar regression results as the standard PLSR feature selection and regression while generally selecting fewer wavelengths. We identified a set of biochemically consistent bands (606, 641, and 1494 nm) to predict the nitrogen content with an RMSE of 0.17% (using leave-one-out cross-validation) in samples with nitrogen contents ranging between 2.4 and 3.6%. Further studying is needed to confirm the relevance and consistency of the wavelengths selected for each nutrient model, but ensemble feature selection showed promise in identifying stable sets of wavelengths for assessing grapevine nutrient contents from canopy spectra.

Highlights

Grapes grown on over 920,000 acres (~372,000 hectares) in the United States for wine, raisins, fresh market, and juice products have a multi-billion dollar production value [1].Macro- and micronutrients are often applied as fertilizers in vineyards to optimize the vine size and fruit quality
The leave-one-out cross-validation (LOOCV) score results for the ensemble and partial least squares regression (PLSR) methods for leaf blade nutrients are shown in Table 1, along with the final number of bands selected in each case
Regardless, this work and additional studies will provide useful benchmarks for comparisons and for reaching the goal of identifying consistent wavelengths for nutrient-specific management protocols and sensors. The results of this Concord vineyard study showed that ensemble feature selection can identify wavelengths providing comparable prediction accuracy to PLSR feature selection for predicting key micro- and macronutrients in vineyard management

Summary

Introduction

Macro- and micronutrients are often applied as fertilizers in vineyards to optimize the vine size and fruit quality. It is commonly applied to vineyards, since it plays a major role in the growth and development of shoots and leaves, which photosynthesize to support the subsequent fruit development. An accurate and precise assessment of grapevine leaf blade or petiole (i.e., leaf stem) contents of nutrients is required to guide the judicious application of fertilizers to optimize both the quantity and quality of grapes. The application of appropriate amounts of N, as well as other nutrients, will allow for minimized detrimental impacts in the vineyard (e.g., soil acidification), as well as off-site impacts (e.g., groundwater pollution and the eutrophication of nearby waterways) [4]

Objectives

Methods

Results

Discussion

Conclusion