Towards Ensemble Land Data Assimilation at ECMWF

In this talk we will outline the current status of the ECMWF Land Data Assimilation System (LDAS) and discuss progress towards a more ensemble focused DA technique. The current technique employed at ECMWF is a flavour of the Simplified Extended Kalman Filter (SEKF). However, the linearised observation operator of the SEKF is derived from the output of an Ensemble of atmospheric 4DVar Data Assimilations (EDA), in place of the traditional finite difference method. We will show how increasing the utilisation of the EDA in order to diagnose flow-dependent background errors in the LDAS improves forecasts for surface variables. Although the spread derived from the EDA is very beneficial in land data assimilation experiments a number of land surface variables remain under-dispersed. In order to address this, we have begun experiments perturbing land surface parameters across the EDA ensemble. We will demonstrate how this can alleviate issues of low ensemble spread and outline possible future directions.