|
Abstract:
|
Satellite remote sensing measurements provide global coverage of Earth and its atmosphere within a matter of days, which helps scientists characterize and understand the spatio-temporal distribution of environmental processes. Because satellite data are obtained based on reflected energies from Earth's surface, the retrieved measurements usually have large variability that potentially contains biases. Calibration/validation is often achieved by modeling a linear relationship between the satellite data (Y) and other data sources (X). In this talk, we propose an errors-in-variables model for capturing this linear relationship, where errors are present in both X and Y. Consistent estimators of regression coefficients rely on unbiased estimating equations, which rely on statistical-dependence models from which the errors' variances can be calculated. We illustrate our calibration method through both simulated examples and an application to atmospheric column-averaged CO2 measurements (Y) from the Orbiting Carbon Observatory-2 (OCO-2). Here, the other data source (X) comes from ground-based measurements obtained from the Total Carbon Column Observing Network (TCCON).
|
