|
Abstract:
|
Remote sensing methods are useful to monitor agricultural crops or other vegetation types. However, remote sensing data from satellites often has a wide spatial resolution that captures multiple classes of ground cover. Each time the satellite collects data over a given region or pixel, the observed signal is a weighted combination of different land covers. Disaggregating the signal into its distinct components is a problem sometimes referred to as "unmixing." Without additional information about the signatures of each component, identifying unique land classes is problematic. However, if prior data is available about the annual pattern of each land type, this information can be combined with the observed data to unmix the satellite signal into distinct latent groups. Using data from the Soil Moisture and Ocean Salinity satellite, we propose a Bayesian parametric model to unmix pixels observed over an intensively cultivated agricultural region in the Midwest. Each pixel is composed of a mixture of corn and soybean. Experimental field data show these two crops exhibit unique signatures as a function of accumulated thermal time or growing degree days. Information is also avai
|
