Spatial Prediction for Massive Datasets

Remotely sensed spatio-temporal datasets on the order of megabytes to terrabytes are becoming more common. For example, polar-orbiting satellites observe Earth from space, monitoring the Earth’s atmospheric, oceanic, and terrestrial processes, and generate massive amounts of environmental data. The current generation of satellites, such as the National Aeronautic and Space Administration’s (NASA) Earth Observing System (EOS) Terra and Aqua satellites, generate about 1.5 terrabytes of data per day. In the USA, there are remote-sensing projects in preparation that will dwarf even these datasets. NASA, the National Oceanic and Atmospheric Administration (NOAA), and the Department of Defense (DoD) have created the National Polarorbiting Operational Environmental Satellite System (NPOESS) to provide long-term systematic measurements of Earth’s environmental variables beginning about 2009. The precursor of this NPOESS mission, the NPOESS Preparatory Project (NPP), serves as a bridge between NPOESS and NASA’s EOS program and is scheduled to launch in Fall 2006. Scalable statistical methods are needed to process and extract information from these massive datasets. Of particular interest here is Total Column Ozone (TCO) data from the Total Ozone Mapping Spectrometer (TOMS) instrument (http://toms.gsfc.nasa.gov). Flying on NPP is a whole suite of sensors, including the Ozone Mapping and Profiler Suite instrument used in obtaining measurements of TCO. This will be the next generation of the TOMS instrument that has flown on three satellites since 1978 (Nimbus-7, Meteor-3, and Earth Probe). In spite of a satellite’s regular polar orbit, remotely sensed data yield datasets that are spatially (and temporally) irregular and on occasions are missing whole swaths. Hence, further processing of these data is required to yield a dataset that is regularly