A multiresolution approach to representing MODIS aerosol optical depth data

Satellite observations of the Earth often contain excessive noise and extensive data voids. Aerosol measurements, for example, can be obscured and contaminated by clouds and are possible only on the sunlit side of the globe. Moreover, the inversion algorithms are sensitive to measurement conditions and physically can change abruptly both spatially and temporally. These features make aerosol filtering and data assimilation a challenging problem.

Multiresolution (wavelet) techniques have found widespread applications in data analysis, with applications ranging from financial time series analysis and prediction to speech pattern recognition and audio signal compression. These techniques are well suited to problems which exhibit local features. Based on recently developed local representations of functions on the sphere, we propose a mutliresolution strategy for scattered data interpolation on the sphere. We present initial results from applying the technique to Moderate Resolution Imaging Spectroradiometer (MODIS) level 3 aerosol data. While the specific results are for aerosol data, the technique is general and can be applied to other data sets.