Predicting crop phenology: a simple logistic regression model approach

Abstract

Crop yield prediction plays a central role in the agricultural planning and decision-making processes. In this paper, we analyze the phenology as a crucial aspect of this topic. We propose a simple model to predict phenology groups on maize and wheat crops at the field-level in Argentina. Our model uses logistic regression and includes photoperiod as an explanatory variable, which is very simple to calculate taking into account latitude and date as input. A large number of data records are used to obtain accurate results. Our model has been tested with over 77% accuracy for both crops. It was also benchmarked with Random Forest, which gives comparable results. However, our study shows that a very simple approach could be used with logistic regression, with very little loss of performance. Our model obtains phenology groups and also performs well with certain critical phenology stages for both crops. Our study aims to provide a simple and effective method for predicting phenology, which can be an aid to crop prediction and for farmers to make accurate decisions. Our work emphasizes the simplicity of the model, the use of a large number of data records, and the inclusion of the photoperiod as an input variable.