2012 Annual Report
1a.Objectives (from AD-416):
Statistical analyses of 7-year database from Trellis Tension Monitors (TTM) in Washington state and three-year database from TTM installations in California, to determine.
1)spatial distribution needed for yield estimation with TTMs;.
2)accuracy of yield estimation;.
3)accuracy of detecting key grapevine phenological stages from the continuous tension trace; and.
4)accuracy of yield estimation under physically constrained and open-ended TTM systems.
1b.Approach (from AD-416):
Complete field data collection; compute classic univariate descriptors; define linear and nonlinear models; complete geostatistical characterization of yield at research sites; and model yields from dynamic trellis tension values.
This research was conducted in support of objective 1B of the parent project. We were trying to solve the problem among grape growers of how to estimate yield by mid-season, more accurately and more cost-effectively. Anticipated yields must be estimated so that wineries and juice processors can prepare their processing equipment for the expected tonnage and most efficiently run their processing plants. The decades-old practice of yield estimation in vineyards is labor intensive: field "scouts" travel from vineyard to vineyard before the fruit begins to ripen. They collect fruit that is weighed and counted. From these numbers they try to predict the final weight, or yield, at harvest, when the fruit is both ripe and larger than at the time of estimation. This work used a sensor in the grapevine trellis to estimate the weight of the crop on a daily basis from the date at which the fruit was formed until the date it was harvested. Growers currently have only one or two dates of crop weight per season. The sensor also was used to estimate the weight of the crop at harvest, much like the growers do, but the sensor data was used to model the daily increase in weight. After building models for several years to develop average expectations of yield from these sensors, growers will be better able to predict yield while reducing or eliminating the very costly and sometimes inaccurate practice of estimating yield by hand.