The idea of using short hindcasts to identify causes of biases associated with the fast-physical processes in climate models has been around for more than a decade, as evidenced by the evolution of the international model inter-comparison activity called ‘transpose-AMIP’. This method makes use of analyses produced with a different forecast model (e.g. reanalysis data) to initiate weather forecasts with a climate model. Although the method is efficient and straightforward to use, it suffers from the initial shock problem caused by the inconsistency between the reanalysis and the model, and a long spin-up (a couple of days) is often necessary. In addition, this method lacks a good strategy to generate ensembles to estimate hindcast uncertainty.

To address these challenges, we propose to use data assimilation (DA) to initialize the short-term ensemble hindcast simulations with the DOE’s E3SM Atmosphere Model. This approach is based on the Data Assimilation Research Testbed (DART) that uses the ensemble Kalman Filter method to generate initial conditions (ICs) for hindcast simulations. DART uses the model to assimilate available in-situ and conventional satellite observations, along with aircraft/ship reports. The resulting analysis products are used to initialize the hindcast simulations for testing model sensitivities to parameterization changes and revealing model deficiencies. Our preliminary evaluation suggests that the DART-initialized hindcasts show much less initial shock than the ‘transpose-AMIP’ during the first 24h hindcast period. The ensemble hindcasts also provide a good estimate of the uncertainty. A comprehensive assessment of the E3SM-DART analysis and the ensemble-hindcast skill will be presented at the conference.