A smoother introduced earlier by van Leeuwen and Evensen is applied to a pr
oblem in which real observations are used in an area with strongly nonlinea
r dynamics. The derivation is new, but it resembles an earlier derivation b
y van Leeuwen and Evensen. Again a Bayesian view is taken in which the prio
r probability density of the model and the probability density of the obser
vations are combined to form a posterior density. The mean and the covarian
ce of this density give the variance-minimizing model evolution and its err
ors. The assumption is made that the prior probability density is a Gaussia
n, leading to a linear update equation. Critical evaluation shows when the
assumption is justified. This also sheds light on why Kalman filters, in wh
ich the same approximation is made, work for nonlinear models. By reference
to the derivation, the impact of model and observational biases on the equ
ations is discussed, and it is shown that Bayes's formulation can still be
used. A practical advantage of the ensemble smoother is that no adjoint equ
ations have to be integrated and that error estimates are easily obtained.
The present application shows that for process studies a smoother will give
superior results compared to a filter, not only owing to the smooth transi
tions at observation points, but also because the origin of features can be
followed back in time. Also its preference over a strong-constraint method
is highlighted. Furthermore, it is argued that the proposed smoother is mo
re efficient than gradient descent methods or than the representer method w
hen error estimates are taken into account.