A New Algorithm for Retrieving Hyperspectral Land Emissivity

Surface emissivity over land is an important parameter needed by NWP data assimilation and many other remote sensing applications. At NOAA, 1D-VAR inversion has been developed to retrieve atmospheric parameters as well as surface emissivity at microwave frequencies. In this paper, the 1D-VAR method will be extended to infrared (IR) wavelengths and used to generate the IR spectral surface emissivity.

In 1D-VAR algorithm, a cost function, J(x), is defined as follows

J(x) = [½(x-x0)T * B-1 * (x-x0)] + [½ (Ym - Y(x))T * E-1*(Ym - Y(x))] (1)

which is minimized to find the optimal solutions of the state variables, X. In this equation, Ym is satellite measurements and Y is the simulated radiances from the forward operator. In our study, we derive a new algorithm for getting the first guess of emissivity spectrum using a new set of hinge point measurements from hyperspectral data. Also, a new static hyperspectral data base from JPL is used as background. Since hypespectral channels are typically huge (e.g. AIRS-2378 and IASI – 8461), the retrievals will be performed in EOF space for both satellite measurements and the state variables. Our newly derived products will be compared with several existing data IR emissivity products such as AIRS version 5 and MODIS Level 2.