Development of a prototype remote sensing data assimilation system for improving land products
Hongliang
Fang, University of Maryland, fanghl@geog.umd.edu
(Presenting)
Shunlin
Liang, University of Maryland, sliang@geog.umd.edu
John
R.
Townshend, University of Maryland, jtownshe@geog.umd.edu
Robert
E.
Dickinson, Georgia Institute of Technology, robted@eas.gatech.edu
The NASA Earth Observing System (EOS) Program is routinely producing high-level land products from multiple sensors. However, there exist a series of generic issues: 1). multiple sensors have not been used effectively; 2). products are not continuous in both space and time; 3). most products are generated by one instrument algorithm regardless of many algorithms developed by the remote sensing community; and 4). most algorithms have not taken advantage of temporal signatures and incorporated a prior knowledge objectively. As a result, almost all products continue to have large uncertainties that have not been well characterized, and many products are not physically consistent.
To address these issues, we have been funded to reformulate the analysis of EOS data by developing a prototype remote sensing data assimilation system. After more than two years of work, significant progress has been made. Specifically, we have 1). developed methods for generating the spatially and temporally continuous land climatology as the first guesses; 2). conducted extensive validation for determining the accuracies of the existing products; 3). developed several new algorithms for producing different estimates of variables (e.g., aerosol optical depth, leaf area index, broadband albedo) that can be integrated through a data fusion algorithm, and 4). evaluated and developed different assimilation algorithms (e.g., ensemble Kalman filter, variational optimization with the adjoint method, neural networks).