Data assimilation has traditionally been employed to provide initial c
onditions for numerical weather prediction (NWP). A multiyear time seq
uence of objective analyses produced by data assimilation can also be
used as an archival record from which to carry out a variety of atmosp
heric process studies. For this latter purpose, NWP analyses are not a
s accurate as they could be, for each analysis is based only on curren
t and past observed data, and not on any future data. Analyses incorpo
rating future data, as well as current and past data, are termed retro
spective analyses. The problem of retrospective objective analysis has
not yet received attention in the meteorological literature. In this
paper, the fixed-lag Kalman smoother (FLKS) is proposed as a means of
providing retrospective analysis capability in data assimilation. The
FLKS is a direct generalization of the Kalman filter. It incorporates
all data observed up to and including some fixed amount of time past e
ach analysis time. A computationally efficient form of the FLKS is der
ived. A simple scalar examination of the FLKS demonstrates that incorp
orating future data improves analyses the most in the presence of dyna
mical instabilities, for accurate models and for accurate observations
. An implementation of the FLKS for a two-dimensional linear shallow-w
ater model corroborates the scalar analysis. The numerical experiments
also demonstrate the ability of the FLKS to propagate information ups
tream as well as downstream, thus improving analysis quality substanti
ally in data voids.