I have recently completed a 6-month project for Oregon State University making a map of the surface roughness of Greenland during 2007 using data from the Multi-angle Imaging SpectroRadiometer instrument (MISR: http://www-misr.jpl.nasa.gov/) aboard NASA's Terra satellite. This satellite carries several different instruments which are dedicated to observing the Earth for scientific, and primarily environmental, purposes. The MISR instrument has the ability to almost simultaneously view the same part of the Earth from 9 different angles, in 4 different wavebands. I developed a technique that uses images from 3 of these cameras to create several image-space data models that could together predict the roughness (in RMS (Root Mean Square) deviation from a local plane) of each pixel in a range of from about 3 cm to 10 m, even though the resolution of the spacecraft images was 275 m. The different data models were calibrated with LIDAR (LIght Detection and Ranging) data from ATM (Airborne Topographic Mapping) aircraft that flew over Greenland during May of 2007 and could measure the surface roughness to a high degree of accuracy, but only for the points under their flight paths. Using the satellite images, roughness could be estimated almost everywhere on Greenland.
Measuring surface roughness in this way is important for at least 3 reasons:
As you should know, Greenland is a large island situated near the north rotational pole of the Earth. The following is an image from Google Earth, and a plot of the Scambos & Haran 2002 DEM in meters.
The Terra Spacecraft and MISR instrument fly in a nearly polar orbit that is locked to the sun throughout the year (i.e. the orbit turns with the orbit of the Earth), looping over the north pole 'down' to the south pole many times a day (99 minute period). As the Earth turns under this orbit, Greenland is overflown by the spacecraft in 124 out of 233 different 'paths' which intersect the island at various angles. Because of its northerly latitude, quite often the spacecraft encounters Greenland near to the terminator, around sunset, when the sun is very low in the sky, which makes image analysis more difficult. Shown below are paths 1, 11, 21, and 31.
Corresponding image 'swaths' for these paths during April, 2007:
Although the surface of Greenland has a high albedo (reflectance) and appears very bright in sunlight, it is actually very low-contrast. When the average brightness is removed, images of the ice and snow surface look quite grey, and must have their contrast enhanced in order to show details. Images from either the forward-looking (Cf) or backward-looking (Ca) MISR cameras look pretty much the same, with some differences due to forward-scattering vs. back-scattering of light depending on the material properties of the surface. In general, there is expected to be more forward-scattering from smooth surfaces like ice, and more back-scattering from rough surfaces like snow dunes or heavily crevassed glaciers.
This effect has been used by combining 2 of the MISR camera views (Cf and Ca) to create the 'normalized difference angular index' (NDAI), which is just (Ca - Cf) / (Ca + Cf) for each pixel in the image. The NDAI can bring out considerable detail in images of Greenland. (For more information about the NDAI, see the research log link above.) The rightmost image below is the same orbit as the figure above.
These images, or the NDAI, can be projected onto the Greenland DEM, although to be useful they must first be masked for clouds:
The roughness (in RMS centimeters) of parts of Greenland have been measured by aircraft using lasers which scan the surface under the flight path (ATM data). The data from these flights can be correlated with satellite images acquired a few days before or after the flight. Shown below is an NDAI mosaic for an ATM flight made on May 2, 2007. Color is the measured surface roughness, from <= 5.75 cm (blue) to >= 15.64 cm (red).
Although there is a clear positive correlation between the NDAI and surface roughness (because rougher surfaces produce more back-scatter and a greater NDAI), it is unclear just what that relationship is, in a mathematical sense, or how to calibrate it numerically with respect to the ATM data, since the NDAI from different orbits often produce different regressions. Many of these are very good (i.e. have high r values), but some are very noisy. This is because the NDAI is a relative quantity, not an absolute, which can vary from orbit to orbit, with surface orientation, and with changes in illumination.
However, another technique tested with another ATM flight showed more promise.
More on next page...
©Sky Coyote 2009