Share Email Print

Proceedings Paper

Virtual green band for GOES-R
Author(s): Irina Gladkova; Fazlul Shahriar; Michael Grossberg; George Bonev; Donald Hillger; Steve Miller
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

The ABI on GOES-R will provide imagery in two narrow visible bands (red, blue), which is not sufficient to directly produce color (RGB) images. In this paper we present a method to estimate green band from a simulated ABI multi-spectral image. To address this problem we propose to use statistical learning to train and update functions that estimate the value for the 550 nm green channel using the values that will be present in other bands of the ABI as input parameters. One challenge is that in order to exploit as many bands as possible, we cannot use straightforward non-parametric methods such as a look-up tables because the number of entries in look-up tables grows exponentially with the number of input parameters. Other simple approaches such as simple linear regression on the multi-spectral input parameters will not produce satisfactory results due to the underlying non-linearity of the data. For instance, the relationship among different spectra for cloud footprints will be radically different from that of a desert surface. The approach we propose is to use piecewise multi-linear regression on the multi-spectral input to train the green channel predictor. Our predictor is built from the combination of a classifier followed by a multi-linear function. The classifier assigns each pixel to a class based on the array of values from the simulated (or proxy) ABI bands at that pixel. To each class is associated a set of coefficients for a multi-linear predictor for 550 nm green channel to be predicted. Thus, the parameters of the predictor consist of parameters of the classifier, as well as coefficients defining the approximating hyperplane for each class. To determine these classifiers we will use methods based on K-means clustering, as well as multi-variable piecewise linear approximation.

Paper Details

Date Published: 13 September 2011
PDF: 9 pages
Proc. SPIE 8153, Earth Observing Systems XVI, 81531C (13 September 2011); doi: 10.1117/12.893660
Show Author Affiliations
Irina Gladkova, The City College of New York (United States)
Fazlul Shahriar, The City College of New York (United States)
Michael Grossberg, The City College of New York (United States)
George Bonev, The City College of New York (United States)
Donald Hillger, NOAA/NESDIS/STAR/RAMMB (United States)
Steve Miller, Colorado State Univ. (United States)

Published in SPIE Proceedings Vol. 8153:
Earth Observing Systems XVI
James J. Butler; Xiaoxiong Xiong; Xingfa Gu, Editor(s)

© SPIE. Terms of Use
Back to Top