HR: 08:00h
AN: A11E-01 INVITED    [Abstracts]
TI: Fusing measurements statistically: combining aerosol data from MISR and MODIS
AU: * Cressie, N
EM: ncressie@stat.osu.edu
AF: The Ohio State University, Department of Statistics 1958 Neil Avenue Room 404, Columbus, OH 43210-1247, United States
AU: Braverman, A
EM: Amy.J.Braverman@jpl.nasa.gov
AF: Jet Propulsion Laboratory, California Institute of Technology Mail Stop 306-463 4800 Oak Grove Drive, Pasadena, CA 91109-8099, United States
AU: Nguyen, H
EM: hnguyen@stat.ucla.edu
AF: University of California, Los Angeles, Department of Statistics 8125 Math Sciences Bldg Box 951554, Los Angeles, CA 90095-1554, United States
AB: We are interested in producing an aerosol data set that provides 1) the best possible representation of aerosol properties, given information from the MISR and MODIS instruments, and 2) quantitative measures of uncertainty associated with that representation. Uncertainties are due to instrument measurement errors, aggregations over space and time arising from different sampling characteristics and footprints, and incomplete data. Our approach is to consider this as a statistical estimation problem. That is, using all information available, find the best statistical estimate of the quantity of interest, say aerosol optical depth (AOD), as a function of location and time. We do this in two steps. First, we use geostatistical smoothing (GS) to estimate the true values of AOD using each instrument's data individually, on a reference grid of locations and times. GS, also known as kriging, is a spatial analog of simple linear regression that accounts for and exploits spatial autocorrelation to produce optimal estimates, or predictions, of unobserved values. Estimates from GS are routinely accompanied by the kriging variance, a formal measure of estimation uncertainty. In the second step, which we call Bayesian Data Fusion (BDF), we form linear combinations of instruments' smoothed estimates at each point of the reference grid. The coefficients for these linear combinations are derived from a statistical model for the relationship between the smoothed data and the true but unobserved values of AOD. BDF not only combines the individual instruments' information in a statistically optimal fashion, but also propagates their uncertainties through the fusion step to produce the desired data set. Both GS and BDF have been used successfully for many years in social- and physical-science applications; their combination in this context offers a coherent way to make inferences with NASA data in the presence of uncertainty.
DE: 0305 Aerosols and particles (0345, 4801, 4906)
SC: Atmospheric Sciences [A]
MN: 2007 Fall Meeting