Albedo bias and the horizontal variability of clouds in subtropical marine boundary layers: observations from ships and satellites
Robert Pincus, Sally A. McFarlane, and Stephen A. Klein<
Journal of Geophysical Research, March 1999.
Volume 104, pages 6183-6192.

Abstract

The scientific problem: Clouds are one of the most important parts of the climate system, and also one of the hardest to simulate correctly. Clouds are important because they affect the radiation that flows between the Earth and outer space. The planet is warmed by incoming sunlight and cooled by the infrared radiation emitted by the planet and its atmosphere, including the clouds embedded in it. The amount of infrared radiation is dependent on the temperature of the earth-atmosphere system, and in the long term the temperature adapts so that amount of cooling adapts to match the amount of heating. Low clouds in particular dramatically change the amount of sunlight reflected back to space, so they act to cool the planet.

Clouds are hard to simulate because they vary enormously over very small distances. The current generation of climate and weather models keeps track of the average amount of water and ice in clouds at each point in the model, but the distance between the points is hundreds of times larger than the size of the clouds themselves. This means that the model results have to be thought of as an average over all the clouds between the points. This averaging causes a serious problem, however. The amount of light returned to space changes nonlinearly according to how much water is in the clouds. The result of this nonlinear relationship is that the amount of light reflected from a group of clouds containing varying amounts of water differs systematically from the amount of light reflected from a cloud containing the average amount of water. This difference is called the "albedo bias", and it can be large enough to be quite worrisome.

Combining observations from ships and satellites: Several mathematical tricks have been developed to remove the albedo bias from the radiation calculations in weather and climate models. These methods need a way to predict how much the clouds vary between the model points, though, and no one knows what factors control the variability. We decided to see how variable clouds really are, and try to discover what that variability is related to. We combined two sets of data. One is a collection of logs taken by sailors on ships sailing in the Pacific Ocean during summer, where reflective low clouds are very common. These reports contain the sailors' estimates of how much of the sky is obscured by clouds, and a note of the most common type of low cloud, such as isolated and puffy cumulus or large sheets of uniform stratus. The second data set is a collection of images taken by weather satellites over the same area. We created matching pairs of observations: for every ship report, we created a satellite sub-image containing only the area around the ship that was visible to the observer. This area, extending about 60 km from the ship, is about the same size as the distance between the points in a climate model, and contains about 16000 measurements of individual clouds. We used a well-known technique to estimate the amount of water in the clouds based on how much light each reflected back to space.

How variable are the clouds? Our first step was to ask if we could describe the 16000 measurements of cloud water in each scene more succinctly. We found that several different formulae which depend on only two numbers did a good job of representing the variability of the 16000 measurements. These two numbers represent specific ways of describing both the average amount of water in the cloud and the amount of variability. The statistical summaries did a very good job, nearly eliminating the albedo bias, which means that large scale models would need to predict only two numbers rather than 16000.

Is the variability related to the type of cloud? We then looked to see if the cloud variability, as measured by the statistical summaries, is related to the quantities that might be available in a climate model. Of all the quantities we examined, cloud type is most strongly tied to the amount of variability. Stratus and mixtures of stratus and cumulus tend to contain more water than cumulus clouds, but the cumulus are generally more variable. Fog is usually very uniform and not particularly thick. There is a lot of overlap in the amount of variability between different cloud types, but each type is distinct. We pointed out that, since information about cloud type can be inferred from large scale models, the relationships we saw would provide a useful way to predict the amount of variability in cloud water in climate models, which would let us remove the albedo bias.

The abstract from the technical journal is available.