Prediction of harvest start date in highbush blueberry using time series regression models with correlated errors

Abstract

This work proposes a model for predicting the harvest start date sufficiently far ahead to enable the farmer to make a well-informed plan. The heat unit method is widely used in agriculture as the phenological unit of time, which offers the least variation in date predictions, and heat units have been used to estimate the start of harvesting in various crops. The problem is that the farmer needs to know the number of days and not the number of heat units that are needed until the harvest can begin. It is proposed that the daily maximum and minimum temperature time series be modelled through regression models with errors correlated using a sine curve. Using the requirements reported by Carlson and Hancock (1991) for the start of harvest of 13 varieties of blueberry over 15 years, a model has been developed that allows the requirements of heat units to be translated into days remaining until harvest. The models are estimated at intervals of 3 months, 2 months, 1 month, 14 days and 7 days before the date at which the heat unit requirements are reached. Three months ahead, the error was less than 10 days late, and 7 days ahead, it was 2 days late. A blueberry orchard in Temuco, Chile, was used as a case study and had similar results. All the errors are within the variability of the heat unit models. The models can be used by farmers to predict and plan the blueberry harvest with adjustments for location and variety.