Recent progress in the field of land surface data assimilation has been extensive. Operational systems for the off-line integration of land surface models with observed meteorological forcing data are already in place and are producing regional and global estimates of land surface conditions for a variety of clients. These systems are obvious starting points for the development of true data assimilation systems, i.e., systems that use comprehensive error analysis to incorporate radiances or geophysical retrievals of land surface conditions (e.g., soil moisture, snow, surface temperature, and vegetation state) into the state estimates produced by the land surface models. Indeed, the time is ripe for the development of such systems for operational applications, given that land state information is now provided with unprecedented coverage in time and space by satellite sensors. Prototype global systems already exist and show the desired improvements in the estimation of land states.

Land surface data assimilation, however, comes with many challenges. The penetration depth of the satellite signal is limited; soil moisture estimates, for example, can only be retrieved by satellite for the top few millimeters of soil when vegetation is sparse and cannot be retrieved at all under dense vegetation. The land surface boasts an extensive heterogeneity relative to that of the ocean and atmosphere, complicating the interpretation of the signal. Furthermore, satellite retrievals and land model variables are generally not directly comparable and sometimes exhibit strikingly different climatologies; they are thus not easily combined. These difficult issues have been the subject of much research in recent years. In the present talk we will discuss the current state of land surface data assimilation, with some emphasis on the supporting research.