This paper analyzes MODIS 1 km albedo kernels of 7 spectral bands over Northern Africa and the Arabian Peninsula and through these kernels develops a new high quality dataset that provides a simple statistical method to scale up spectral and broadband albedos from pixel to arbitrary coarse resolution grid square for use in climate models. This dataset significantly improves characterization of spatial and spectral variability and solar zenith angle dependence of soil albedo relative to simple grid means from MODIS data. The statistical method based on minimum noise fraction rotation transforms is able to not only successfully capture most of the MODIS albedo variance but also to extract large-scale spatial structures of albedo patterns from the original MODIS data while improving the data quality and reducing the number of parameters needed to represent the data.