Correlations between Vineyard Yield Data, UAV-Generated Indices and Physical Properties

A study was undertaken, on several vineyards (6,2 Ha) on the farm Lindenhof, to compare February 2018 yield data, collected by Ackerview, against Normalized Difference Vegetation Indices (NDVI) and physical properties collected by the application of a UAV-borne (Unattended Aerial Vehicle, or drone) system by GeoFlight. Specifically, the spatial distribution of grape yield was compared against maps of NDVI, an indicator of vineyard health, and vineyard plant volume, in order to establish correlations between these data sets. UAV data was refined to show the distribution of relative plant growth and resilience in terms of the vineyard’s health and the volume of plant material, all derived from multispectral NDVI imagery.

Health data and plant volumetric data were filtered, according to plant height, to exclude green growth less than 20 cm off the ground, which typically constitutes weeds or grass. Vineyards were divided into sections of 5m per vineyard, along each row, and a data point was assigned to each 5m section. Each section’s data point contains averages of area coverage of vine canopy, canopy volume, NDVI (health) value and, relative grape yield. Yield is portrayed by relative values as mass calibrations were not available. Mass calibrated yield maps can be seen in this post.

A correlation was found between plant health, plant volume and relative grape yield across most of the surveyed area; however, correlation diminishes in areas where there is exceptionally large or exceptionally small biomass volumes and/or volume changes.

The following deductions were derived from this study:

  • Where poor plant volumes and poor plant health coincides, relatively poor yields may be expected.
  • Where moderate canopy cover and moderate- to high- plant health is evident from the analysis, moderate yields may be expected.
  • Canopy health alone does not necessarily provide an indication of the expected grape yield.
  • Different vine varieties result in different yields.
  • Consistent yield measurements in the North East block resulted in a greater correlation with mapping measurements.
  • Correlation between yield, vineyard health and vineyard size diminishes for sections with very large or very small canopy cover.

The following issues need to be addressed in further studies:

  • Correlations between plant size, health and yield are likely to differ, based on vine variety.
  • The accuracy of standard GPS systems results in a slight offset between different datasets; however, this may be corrected on a case-by-case basis.
  • There is typically a slight lag between grape collection and yield measurement during harvesting, which results in a slight displacement of yield data. This is easily corrected in further post processing.

Chris Anthonissen
22 February 2019