Information constraints in variational data assimilation

Journal article, 2018

Data assimilation of indirect observations from remote-sensing instruments often leads to highly under-determined inverse problems. Here a formulation of the variational method is discussed in which (a) the information content of the observations is systematically analysed by methods borrowed from retrieval theory; (b) the model space is transformed into a phase space in which one can partition the model variables into those that are related to the degrees of freedom for signal and noise, respectively; and (c) the minimization routine in the variational analysis is constrained to act on the signal-related phase-space variables only. This is done by truncating the dimension of the phase space. A first test of the method indicates that the constrained analysis speeds up computation time by about an order of magnitude compared with the formulation without information constraints.