Analyzing the impact of forecast errors in the planning of wine grape harvesting operations using a multi-stage stochastic model approach

TL;DR

This paper examines how forecast errors in yield and rain probabilities affect wine grape harvest planning and losses, using a multi-stage stochastic model to improve decision flexibility and reduce product losses.

Contribution

It introduces a multi-stage stochastic optimization model that accounts for forecast errors and allows dynamic adjustment of harvest planning decisions.

Findings

Underestimations of yields cause more significant losses than overestimations.

Flexibility in revising hiring decisions has limited impact when yields are overestimated.

The model effectively reduces losses in high-quality grapes, which are crucial for income.

Abstract

Forecasts and future beliefs play a critical role in the harvest labor hiring planning, especially when errors in them entails fixing previous made decisions, which can carry extra costs or losses. In this article, we study the effect that errors in the forecast/belief can have in the wine grape harvest planning process and the losses of the product. Errors are reflected in the prediction of yields and in the estimation of rain transition probabilities have on the value and losses of product. Also, using a multi-stage stochastic optimization model we can study the effect that second stage decisions have on the ability fix the planning decisions, reduce product losses and generate value. In a first step, we develop a multi-stage stochastic model which considers grape growth uncertainty given a belief in future events. The model decisions variables are: hiring, firing and maintaining harvest labor through periods, and also the harvested quantities in each period and block. Once the model defines the plan for the coming epoch, some decisions are implemented and a deviation in the forecast is revealed and the decision maker can adjust future decisions and beliefs. Results indicate that the effect of the errors in yield determination is not symmetrical; underestimations of the yields have a more significant negative effect on the objective function, while overestimation does not. Flexibility to revise hiring decisions does not make a significant difference if the yields are overestimated. The model significantly reduces losses of the better-quality grapes, since they correspond to a significant proportion of the income and account for the largest portion of income loss. Last, grapes that have an early improvement of their quality give the decision-maker an extra level of flexibility to adjust the harvesting plan.