Getting a handle on measurement error
Measurement error is a given in science. Multiple measurements of any subject will vary, either because of variation of the subject or variation in the measurement method. The field of statistics has helped scientists to come to grips with this issue, allowing them to separate out and quantify various types of errors. A group at Oxford has developed new techniques for analysing measurement error associated with satellite-based measurements of atmospheric constituents from space. The Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) aboard Envisat makes spectral measurements of the Earth's atmosphere. Following a complex processing procedure, concentrations of greenhouse and other gases can be inferred from the spectra. Error, deriving from the instrumentation itself, its calibration, the assumptions upon which the data processing models are based and so on, can distort the results. In their approach to the problem, the Oxford team applied advanced mathematics. New techniques, such as the micro-window selection algorithm, were created. The intricate relationships between the different types of error are managed according to theories that have been verified with experimental data. The tricky part is understanding how these errors propagate through the system of forward radiative transfer models and subsequent inversion. For this a new posterior error analysis technique was developed: Residual and Error Correlation (REC) Analysis. REC Analysis provides information about the number and magnitude of errors in the resulting vertical profiles of the various gaseous species. Collectively, these results will facilitate scientists worldwide working with data collected by MIPAS and similar instrumentation.
