Predicting Apple Yield Based On Occurrence of Phenological Stage in Conjunction With Soil And Weather Parameters

Abstract

Accurate and reliable yield forecasting is required for efficient planning and management of an important crop like apple. Efforts have been made to predict apple yield, mostly through the use of statistical tools with limited indicator parameters. The proposed neural network (NN) based system predicts yield of apple crops in an orchard based on identification, characterization, time of arrival and duration of phenological stages interactively with soil and weather parameters. The task of automatic yield prediction in orchards is challenging. Despite the significant amount of work that has been put into developing automated methods for estimating yields, the majority of methods currently in use are based on fruit counting, which is only useful one to four weeks before harvest. Whereas, in the proposed system, we will be predicting yield, during each phenological phase, among six classes, taking into account time of phenological stage occurrence (i.e. early occurrence, normal occurrence, or delay occurrence), soil parameter, and parameter related to weather conditions. This model will help the growers to timely take decision to execute contingency plans in case of average or low yield. The f1-score of the proposed system is 0.94. It is compared with other popular machine learning (ML) algorithms like Logistic regression, Support vector machines (SVM) and K-nearest neighbors (KNN).