We introduce basic ideas and methods of data assimilation in meteorology and oceanography, and illustrate their progress from numerical weather prediction to ocean and climate studies. Novel applications to space physics will be covered as well.
We start with the best linear unbiased estimation of scalar temperature from several observations and show how to extend this to the typical situation of interest, in which one of the current “observations” is a prediction based on a dynamic model and past observations. We thus proceed from the scalar to the vector situation, and study information transfer from observed to unobserved variables. These simple steps are the basis of the sequential approach to combining information from models and observations.
We next extend the models being used from systems of linear differential equations (Kalman, 1960) to systems of nonlinear partial differential equations governing planetary flows, and study advection of information from data-rich to data-poor regions. The duality of estimation and control — and thus of sequential and variational methods — is discussed, along with parameter estimation vs. state estimation.
Data assimilation for weather vs. climate is presented in discussing filters vs. smoothers (Wiener, 1949). Difficulties are emphasized, including strong nonlinearities; data availability, accuracy and diversity; as well as computing power and storage limitations.
Finally, we address the recent use of data assimilation (DA) for the detection and attribution (D&A) of weather- and climate-related events to anthropogenic and other causes (DADA). Bright perspectives for the future conclude the talk.
This introduction to data assimilation is based on joint work with many students and colleagues over several decades; see <link https: dept.atmos.ucla.edu tcd _blank>dept.atmos.ucla.edu/tcd.
Ghil, M., and P. Malanotte-Rizzoli, 1991: Data assimilation in meteorology and oceanography, Adv. Geophys., 33, 141–266.