Prediction Software using OFPE and Machine Learning

The increasing accessibility of precision sensors capable of consistently and continuously gathering data from agricultural fields provides an opportunity to develop models that predict critical response variables, including but not limited to crop yield and protein content. Leveraging various machine learning (ML) algorithms, these models are trained to integrate data from On-Farm Precision Experimentation (OFPE), generating site-specific insights into field management responses. In this training session, we will present the prediction tool we have developed as part of the Analytic Engine of the Data-Intensive Farm Management (DIFM) project. This tool incorporates recent and well-known ML algorithms, including Hyper3DNetAQD, convolutional neural networks with late fusion (CNN-LF), random forests, generalized additive models (GAMs), Bayesian linear regression, and standard linear regression. The session outlines the configuration, training, and storing of these models, along with the creation of prediction shapefile maps. Furthermore, we will discuss the quantification of prediction uncertainty through the generation of uncertainty maps based on estimated prediction intervals.

Jan 8, 2024 —