Making Sense of Large, Complex Datasets: Using MISR's Multiangle and Multispectral Information to Detect Clouds and Aerosols

Traditional remote sensing is performed using a single, nadir-pointing instrument. Detection of cloud and/or aerosols from such an instrument takes advantage of multiple spectral bands and physically or empirically based thresholds using a number of these bands. The Multiangle Imaging SpectroRadiometer (MISR) instrument currently operational on the Terra satellite offers a unique perspective by obtaining radiance data in four spectral bands using nine cameras with viewing directions ranging from 0 to 70.5 degrees. The MISR dataset presents a challenge to scientists, not only because it requires a shift from the traditional, downward-directed way of viewing the world, but also because, in order to take advantage of the full power of the instrument, new ways must be found to combine and interpret the data. We will describe a variety of methods that have been developed to detect clouds and aerosols using MISR. These methods range from traditional, threshold approaches to techniques which geometrically combine information from MISR's cameras to provide a quasi-three-dimensional view of the world. Most powerfully, machine learning techniques have been applied, specifically Support Vector Machines (SVMs), a cousin to neural networks, that show immense promise in exploiting the full potential of the MISR instrument.