Sequential synthetic aperture radar (SAR) images from the ERS 1 spacecraft were routinely used as input to the Geophysical Processor System at the Alaska SAR Facility from 1992 through 1994 to produce sea ice displacement vectors on a 5-km grid. We have combined some 5000 of these products, consisting of nearly 700.000 displacement vectors, into a set of 914 strips or mosaics of ice motion that span (mostly) 3-day intervals. For each strip we compute the opening and closing of leads based on change in area of the 5-km grid cells. We also compute the area-averaged deformation of strain invariants of the motion. The small-scale (several kilometer) lead activity can be parameterized fairly well in terms olf the large-scale (several hundred kilometer) strain invariants. Sea ice models implicitly contain such a parameterization through their constitutive equations. The well-known model of Hibler (1979) that uses a viscous-plastic rheology and an elliptical yield curve is in good agreement with our data. The variance in the data about this theoretical relationship is less than that suggested by a random model of sea ice motion. The data also indicate that shearing deformation contributes to the opening and closing of leads. Models with two categories or levels of ice thickness generally do not take shearing deformation into account in the evolution of the ice concentration, nor do they redistribute ice as a result of ridging. We show how to add these features to two-level models. However, we argue that three-level models consisting of open water, thin ice, and thick ice are much better suited to the proper treatment of open water production, and we present such a model. ¿ American Geophysical Union 1995.