Objective wine grape crop estimation model using actual spatial and production data sets
Objectives & Rationale
A project titled “Objective wine grape crop estimation model using actual spatial and production data sets” was initiated in 2016. The project was conceptualized within the context of current wine grape crop estimation methods which often provide yield projections merely based on historic data. The accuracy of the estimates are often influenced by inconsistent and often subjective sampling approaches and inputs, which tend to frequently adjust towards historical yield estimates. Typically the yield predictions consider historical yield data and weather indices in combination with manual measurements in the vineyard. This generally involves harvesting whole segments of vines or randomly sampling inside the vineyard, weighing bunches and combining average bunch weight with the number of bunches per vine, in order to infer the yield of the entire vineyard, which makes this approach time consuming and costly.
The new era of data capturing on computers and tablets means more databases exist, containing a wealth of information related to a farm, production unit and/or vineyard. Available technology has improved significantly in recent years, allowing for the automatic (effective), repetitive (frequent) and unbiased capturing of growth information in space and over time. Specifically in the Western Cape of South Africa, a wealth of information is captured within the wine grape industry. Many of the datasets extend over numerous years and provide the opportunity for use in statistical analyses and machine learning exercises.
This led to a proposal to Winetech to investigate the use of available big datasets for wine grape crop estimation. The overall objective of the study was to leverage these big datasets as a basis for developing and researching a new, consistent and reliable wine grape yield estimation model. In this project, two specific aspects were investigated, namely the:
(a) relationship between spatial datasets available through FruitLook (FL) and vineyard block yield information, through statistical analysis and machine learning (ML) approaches; and
(b) feasibility of using these relationships to estimate wine grape yield in a season.