Application of a PC analysis of remote sensing spectral reflectance data to determine source regions for dust storms.

The source regions of deserts over which dust is lifted into the air occur on relatively small scales and hve been difficult to characterize geographically. The research programs supporting NASA's terra and aqua satellites have pioneered the development of quality controlled climatologies of surface information, that should include features characteristic of dust source regions. With quality satellite data, patterns can be distinguished that are likely to indicate real geophysical features. Zhou et al. (2005) applied a principle component (PC) analysis to the North African desert that revealed intriguing spatial structures. It represents the spatial patterns with a limited number of PC's that maximize the covariances between spectral bands averaged over space. Such analysis acts as a filter of nonphysical information and should highlight commonly occurring small scale features, some of which are likely connected to dust source regions. We present a new such analysis for the Taklamakan desert of China. The first 2 or 3 modes are removed as being of too coarse a spatial scale to represent source regions. Features of the remaining 3 or 4 patterns of significant amplitude are examined for their possible relevancy to dust uplifting. In particular, patterns that show seasonality of vegetation are deemed unlikely to be so relevant. Distinctions in the spectral patterns and MODIS BRDF kernels of remaining terms are examined for their likely indication of a dust source region.