92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Wednesday, 25 January 2012: 4:00 PM
Evaluation of Satellite Precipitation's Spatial Patterns with EOF Analysis and Shannon Entropy
Room 342 (New Orleans Convention Center )
Wei Chu, Univ. of California, Irvine, CA; and X. Gao and S. Sorooshian

Thanks to its capability to cover the mountains, where ground measurement instruments cannot reach, satellite provides a good means of estimating precipitation over mountainous regions. In regions with complex terrains, correct information on high-resolution spatial distribution of precipitation is critical for many important issues, such as flood/landslide warning, reservoir operation, water system planning, etc. Therefore, in order to be benificial to many practical applications, satellite precipitation product should possess high quality in characterizing spatial distribution. However, most existing validation metrics, which are based on point/grid comparison using simple statistics, cannot effectively measure satellite's skill of capturing the spatial patterns of precipitation fields. This deficiency results from the fact that point/grid-wised comparison does not take into account of the spatial coherence of precipitation fields. Furth more, another weakness of many metrics is that they can barely provide information on why satellite products perform well or poor. Motivated by our recent findings of the consistent spatial patterns of the precipitation field over the western U.S., we developed a new metric utilizing EOF analysis and Shannon entropy. The metric can be derived through two steps: 1) capture the dominant spatial patterns of precipitation fields from both satellite product and reference data through EOF analysis, and 2) compute the similarities between the corresponding dominant patterns using mutual information measurement defined with Shannon entropy. Instead of individual point/grid, the new metric treat the entire precipitation field simultaneously, naturally taking advantage of spatial dependence. Since the dominant spatial patterns are shaped by physical processes, the new metric can shed light on why satellite product can or cannot capture the spatial patterns. For demonstration, a experiment was carried out to evaluate a satellite precipitation product, CMORPH, against the U.S. daily precipitation analysis of Climate Prediction Center (CPC) at daily and .25o scale over the Western U.S.

Supplementary URL: