Conversion to mechanical management in vineyards maintains fruit

Abstract

Context and purpose of the study – Current environmental, ecological and economic issues require a better vineyard production management. In fact, a poor use of fertilizing could lead to harmful impact on environment. Another issue concerns the cultures themselves which couldn’t use fertilizers efficiently, leading to a loss of income or too much expense for farmers. Presently, estimation of fertilization’s needs is realized by the laboratory analysis of leaves selected through a random sampling. The present study aims at optimizing fertilization’s management by using a map of biophysical parameters estimated from satellite images.

Material and methods – Since 2016, experiments are carried out in three vineyard regions of France on three grapevine varieties (Merlot, Cabernet Franc and Merlot). The objective is to test if biophysical parameters or vegetation indices could be used to manage fertilization. Around ten plots in each region were studied. Leaves were sampled around the veraison period. Laboratory analysis were made to determine various parameters such as nitrogen, phosphorus and potassium content of leaves. Spot and Sentinel 2 satellite images were taken during the same period with a spatial resolution from 1.5m/pixel to 20m/pixel. A radiative transfer model was used to calculate biophysical parameters, including leaf area index (LAI), green cover fraction (Fcover), and chlorophyll content estimated in leaf (CHL). First, principal component analysis (PCA) were made to better understand the data distribution. Then, links between leaves components and biophysical parameters or vegetation indices were determined using simple and multiple linear regression.

Results – Differences were observed between each region. This could be due to different varieties, soil, climate and grapevine management (row spacing, pruning…). Models were also founded to predict nitrogen content of leaves using the biophysical parameter CHL (2016: R²=0,64, 2017: R²=0,59). These promising results still need to be confirmed with 2018 data. To improve accuracy further work will be carried out with other innovative methods such as machine learning.