California’s many colors analyzed from the air
In 2013 and 2014, researchers from the Jet Propulsion Laboratory (JPL) flew back and forth over long swaths of California aboard NASA’s ER-2 aircraft to take pictures of the land, and occasionally the ocean. The flights carried them at an altitude of 20 km over forests and mountains around Lake Tahoe and Yosemite National Park; urbanized coasts near San Francisco, Santa Barbara, and Los Angles; farmlands inland from those cities; and the upper reaches of the Sonoran desert near the Salton Sea. Taken with JPL’s Airborne Visible/Infrared Imaging Spectrometer (AVIRIS), the 20-m-resolution images are no ordinary aerial photographs. Each pixel contains a spectrum that spans the wavelengths of sunlight, from visible to shortwave-IR (380 nm–2500 nm), reflected by Earth’s surface. To broadly examine how much information the images carry, JPL’s David Thompson and his colleagues performed a statistical analysis of the nearly 3 × 109 spectra that make up the images. From that analysis they have estimated the intrinsic dimensionality, the number of independent variables required to describe a data set. As shown in the example flight line below, the dimensionality of individual 20 km segments ranges from the low 20s to the high 40s. For the combined data set, the researchers find a dimensionality of about 50. The number of distinguishable spectral signatures—for example, of forest canopy chemistry or mineral deposits—grows exponentially with dimensionality. And the dimensionality sets the minimum number of spectral bands that need to be measured to capture the full spectral diversity. For context, NASA’s Landsat 8 satellite measures eight noncontiguous bands in the visible to shortwave-IR range; AVIRIS measures 224 contiguous bands. Thompson and his colleagues expect their analysis will inform scientists and engineers who are planning future satellite-based imaging spectrometers and will give a preview of the wealth of data that will come from those instruments. (D. R. Thompson et al., Opt. Express 25, 9186, 2017 .)
