A Parametric Approach to Unmixing Remote Sensing Crop Growth Signatures

Remote sensing data are often measured or reported over wide spatial footprints with heterogeneous ground cover. Different types of vegetation, however, have unique signatures that evolve throughout the growing season. Without additional information, signals corresponding to individual vegetation types are unidentifiable from satellite measurements. In this paper, we propose a parametric mixture model to describe satellite data monitoring crop development in the US Corn Belt. The ground cover of each satellite footprint is primarily a mixture of corn and soybean. Using auxiliary data from multiple sources, we model the aggregate satellite signal, and identify the signatures of individual crop types, using nonlinear parametric functions. Estimation is performed using a Bayesian approach, and information from auxiliary data is incorporated into the prior distributions to identify distinct crop types. We demonstrate our parametric unmixing approach using data from the European Space Agency’s Soil Moisture and Ocean Salinity satellite. Lastly, we compare our model estimates for the timing of key crop phenological stages to USDA ground-based estimates. Supplementary materials accompanying this paper appear online.