Using Multiview Polynomial Learning to Estimate the Planting Dates of Crops

This study presents a novel approach to predicting crop planting dates by integrating ground-based Leaf Area Index (LAI) measurements with satellite images through a method we term Multiview Polynomial Learning. The research leverages precise time-series LAI data. Third-degree polynomials are used to describe each year's crop growth. Due to the scarce availability of ground LAI data, synthetic polynomial curves are created to mimic a third-degree polynomial space representing any crop growth.Since ground LAI data collection is not feasible, we turn to the abundant satellite images. To connect satellite information with LAI, we use Orthogonal Canonical Correlation Analysis (OCCA), which maps satellite data to LAI by finding optimal linear transformations that maximize the correlation between these two data views. This OCCA-based mapping creates a consistent and robust dataset, unifying ground and satellite data for subsequent analysis. A neural network model is then trained on the augmented polynomial data, employing 18-fold cross-validation to ensure the model’s robustness and generalizability.The multiview OCCA mapping, combined with our trained neural network based on polynomial spaces, is referred to as Multiview Polynomial Learning. This approach not only applies to predicting planting dates but may also offer a framework that can be adapted to other domains where data from multiple sources must be integrated for predictive modeling.