New Technique Successful for Measuring Thickness of Broken Clouds

Marshak, A., NASA - Goddard Space Flight Center

Cloud Distributions/Characterizations

Cloud Properties

Marshak, A, Y Knyazikhin, K.D. Evans, and W.J. Wiscomb, (2004): The "RED versus NIR" Plane to Retrieve Broken-Cloud Optical Depth from Ground-Based Measurements, Journal of Atmospheric Sciences , 61, 1911-1925.


In the "lookup table," vertical lines within the curves show calculated values of cloud optical depth. Observed data points show actual RED and NIR values; the cloud cover and optical depth are read from the overlaid lines.


In the "lookup table," vertical lines within the curves show calculated values of cloud optical depth. Observed data points show actual RED and NIR values; the cloud cover and optical depth are read from the overlaid lines.

Cloud optical depth (or thickness) is a fundamental property for calculating the amount of solar radiation entering and leaving earth's atmosphere. Current techniques to measure this property work well for completely overcast skies, but leave much to be desired for broken-cloud skies. Working under a grant from the DOE's Atmospheric Radiation Measurement (ARM) Program, researchers at NASA's Goddard Space Flight Center developed a new approach for retrieving cloud optical depth from broken-cloud skies using ground based radiometers. As described in the Journal of Atmospheric Sciences (August 2004), the technique takes advantage of the reflective qualities of surface vegetation, which absorbs 90% of the incident red light but reflects most of the incident near-infrared light back to space. (Near-infrared radiation is emitted by the sun but at wavelengths not visible by human eyes.)

For overcast skies, the amount of solar radiation reaching the ground is a relatively simple function of the optical thickness of the overlying cloud deck. In skies with broken clouds, however, the amount of solar radiation reaching the ground is a combination of direct sunlight shining through the spaces between the clouds and sunlight scattered both through and from the sides of clouds. Thus, the resulting transmitted radiation is simultaneously a function of the fraction of the sky covered by clouds, the 3-D shape of the clouds, the location of those clouds relative to the sun and the ground-based instruments, and the average optical depth of the clouds. Due to the lack of a one-to-one relationship between solar zenith radiance and cloud optical depth, determining the optical depth from radiance measured by a one-channel upward-looking radiometer is nearly impossible.

Because chlorophyll in leaves absorbs 90% of the red light in the infrared (approximately 648-680 nm) range, yet a leaf reflects 90% of the light in the near-infrared (wavelengths greater than 700 nm), the researchers theorized that a spectral ratio approach used to calculate vegetation surface area (often referred to as a vegetative index) might be modified for use in determining cloud optical depth. The essence of the approach is that a radiometer pointing straight-upward looking at a cloud base measures radiation from two sources - direct solar radiation incident on cloud top and transmitted through the cloud and diffuse solar radiation that is reflected from the surface to the cloud and then back from the cloud to the ground. The researchers measured the down-welling solar radiance in two narrow spectral bands in the NIR (0.87um) and RED (0.67 um) spectral regions. They then used these two radiances to construct a "normalized difference cloud index" or NDCI, representing the ratio between the difference and the sum of the two normalized radiances. Because the radiation transmitted through the cloud is roughly the same for both wavelengths, but the radiation reflected from the surface (and then from cloud base) is very different due to the differing reflective properties of vegetation, this approach largely removed the double-valued relationship between radiance and cloud optical depth for broken cloud fields.

To improve the solution of the NDCI equation for cloud optical depth, the researchers developed a "lookup table" approach by creating a graph using the normalized RED and NIR wavelengths as the axes. They then used measurements of the local ground reflection at these wavelengths to calculate and plot curves on the graph that show expected normalized radiance values for cloud covers (ranging from zero to 1, in steps of 0.2), and for a range of cloud optical depth values. Data points measured by the upward-looking radiometer are plotted on the RED-NIR graph, and cloud cover and cloud optical depth can then be read directly from the overlying lines on the graph.

To test the validity of their approach, the researchers used radiance simulations from a fractal model of horizontally inhomegeneous and broken clouds, as well as comparisons to satellite, aircraft, and surface data from the ARM Program's Southern Great Plains site in Oklahoma. In comparison to surface albedo measurements, it was found that the acuracy of satellite-based surface albedo was sufficient for the proposed retrievals of cloud optical properties. In simulations where the true optical depth was known, 75% of the retrieved optical depths (out of about 10,000 pixels with a 25 m resolution) had absolute errors smaller than 3; averaging over 200 m substantially improved the retrieval.