Atmospheric Sciences Research Spotlight

Insights into Long-Standing Bias in Cloud Property Retrieval

A new framework provides a more comprehensive view of how subpixel variations can create biases in a commonly used method of analyzing cloud properties with satellites.

Source: Journal of Geophysical Research: Atmospheres


Clouds play an incredibly important role in shaping the Earth’s climate, both reflecting solar radiation back out to space and acting like an insulating blanket, trapping heat near the surface. The net warming or cooling effect depends on several properties, including cloud optical thickness and droplet size. But unlike other facets of climate science, clouds are ethereal and transient, making them difficult to model despite their importance.

One of the most commonly used methods for analyzing cloud properties is known as the bispectral method, which measures the reflectance of two different wavelengths of light to glean information about a cloud’s optical thickness and droplet radius. These measurements, usually gathered by orbiting satellites, rely on breaking an image down into pixels and performing a pixel-by-pixel analysis.

Although useful, this sort of approximation can also lead to significant biases because it ignores subpixel variation and doesn’t account for the fact that each pixel is continuous with—and thus constantly influencing—its neighbors. This problem can be exacerbated because many current analyses treat optical thickness and droplet radius as two independent, noninteracting variables. In reality, they influence each other.

In an effort to understand and quantify how these subpixel variations can bias the bispectral method, Zhang et al. developed a new mathematical framework that uses a Taylor expansion to account for the fact that droplet radius and optical thickness variables are mutually dependent on one another.  The framework allows researchers to estimate how the two variables change from one application of the bispectral method to the next, especially when using different satellites, which often have different pixel resolutions.

The team then applied their framework to real-world data acquired from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS), as well as data from a synthetic cloud field created using a large-eddy simulation model called DHARMA. MODIS collects data at several different spatial resolutions, i.e., different pixel sizes. This allowed the researchers to explore whether their framework could predict the variations observed in the reported optical densities and droplet radii for the different resolutions. The team reports that their framework was quite useful in predicting the biases caused by subpixel variability in both the large-eddy simulation and real-world MODIS data.

The researchers suggest that their mathematical framework could be useful for understanding the statistical differences observed at different spatial resolutions when using the bispectral method to resolve cloud properties, thus providing future researchers with more accurate tools to analyze our planet’s clouds and their impact on climate. (Journal of Geophysical Research: Atmospheres, doi:10.1002/2016JD024837, 2016)

—David Shultz, Freelance Writer

Citation: Shultz, D. (2016), Insights into long-standing bias in cloud property retrieval, Eos, 97, doi:10.1029/2016EO054799. Published on 28 June 2016.
© 2016. The authors. CC BY-NC-ND 3.0