P5.13
The Development of Cloud Retrieval Algorithms Applied to GOES Digital Data
Randall J. Alliss, Litton-TASC, Chantilly, VA; and M. E. Loftus, D. Apling, and J. Lefever
Cloud retrieval algorithms are being developed and applied to 10 bit pixel level imager data collected by GOES 8, 9 and 10. The detection of clouds at the pixel level is performed by applying a series of tests to the GOES data and to calculated clear sky backgrounds (CSB).
The CSB can be defined as the radiation emitted by the ground that the GOES sensor receives when no clouds are present. The CSB varies spatially due to terrain height and soil type and diurnally due to changes in the ground temperature and soil moisture. A CSB is estimated for each pixel using GOES imagery by identifying recent clear scenes. The goal is to correctly distinguish and identify true cloud features from the backgrounds by comparing raw imagery to the CSB. The recent clear pixel values are actual measured instances of clear sky backgrounds. Four separate CSB are estimated using the longwave (LWIR) channel, the visible channel, the shortwave reflectivity product (SRP) and fog product. The SRP is calculated during the day only using both the shortwave IR (SWIR) and LWIR channel's and is useful in distinguishing low clouds from snow cover. The fog product is derived at night only by subtracting the SWIR from LWIR and is useful for the detection of low clouds and fog. The four CSB's are calculated for each pixel by estimating the tenth percentile most likely clear scenes of the previous thirty days. For example, the LWIR CSB is estimated using a linear regression model formed from prototypical clear-sky pixels that are found using high-confidence clear tests. Eleven predictors are used to compute the estimated temperature. Coefficients for the predictors are determined from the prototype clear sky pixels. The coefficients for the predictors which have substantial spatial variation are themselves bilinearly interpolated functions of spatial position. This allows the overall regression model to account for variations of the LWIR temperature predictability over larger geographic regions.
The cloud detection tests are a function of time of day. During the daytime, the visible, LWIR, and SRP tests are used to classify pixels as clear or cloudy. During the nighttime, the LWIR and fog tests are used. During twilight times, when there is not enough solar illumination to use the visible or SRP tests and not enough darkness to use the fog tests, only the LWIR test is used. The overall design of the cloud detection algorithms is similar for all the tests. For each pixel, the difference between the LWIR temperature, visible albedo, derived products and the predefined CSB is calculated. Threshold confidence ranges for each test are spatially and temporally defined. This difference is then compared to the threshold range. The classification of a pixel as clear or cloudy is based on where the calculated difference falls with respect to the threshold confidence range. For example, the LWIR temperature for a pixel must be at least 6-8° K colder than the LWIR CSB temperature for that pixel to be classified as cloudy. Together the cloud tests result in a composite cloud assessment for each pixel. Results indicate a high degree of accuracy. Examples will be shown at the conference.
This paper will be most suitable under theme 4) Retrieval of atmospheric profiles and constituents. An electronic poster will be given.
Poster Session 5, Retrieval of Atmoshperic Profiles and Constituents: Part II
Wednesday, 12 January 2000, 3:00 PM-5:00 PM
Previous paper Next paper