Earth-observation satellites produce images of the Earth that are stored on
board and then downloaded when the satellite can communicate with a
ground reception station. The use of sophisticated onboard compression
algorithms makes the amount of data resulting from an observation very variable.
Until now, timed data download plans have been built on the ground, sent to the
satellite and executed on board without any change. The current way of dealing
with the data volume uncertainty is to consider maximum volumes. That makes the
data download plans always consistent but also very sub-optimal since
data volumes are often lower than maximum.

The data download problem that we consider is a hard scheduling problem with
scheduling constraints ranging from unsharable resources to time-dependent
processing times and specific constraints such as data encryption. The goal is
to schedule file downloads during visibility windows, meeting the constraints
while minimizing information age and promoting the fair sharing of the satellite
usage between users.

The goal of this work is to share the decision-making process between ground and
board by producing partially-instantiated plans on the ground which are then
completed on board. On the ground, we have enough computing power and time but a
high level of uncertainty about file volumes; on board, we have limited
computing power and time but the uncertainty is far less prominent since many
acquisitions have been already performed and their volumes are known.

We present several levels of flexibility in which decisions are distributed
between ground and board and we show their advantages both from the algorithmic
and operational (performance, predictability) points-of-view.