The Statistical Characteristics of the Precipitation Variable in a Global Model and Satellite Observations from the Point of View of Ensemble Data Assimilation

Satellite remote sensing has provided a great source of global precipitation estimates. For example, the TRMM Multisatellite Precipitation Analysis (TMPA) has provided a 3-hourly high-resolution gridded precipitation rate data in near real time. The Global Precipitation Measurement (GPM) is a follow-up and aims to provide more advanced precipitation measurements. Many attempts to directly use those precipitation observations in numerical models through data assimilation methods have been made, but important issues related to the characteristics of the precipitation variable have made such precipitation assimilation less effective, especially in improving forecasts at the end of the assimilation.

Using a local ensemble transform Kalman filter (LETKF), together with a variable transformation technique and a model background-based observation selection criterion, we have demonstrated promising effective precipitation assimilation based on observing system simulation experiments (OSSEs) (Lien et al. 2013). With a more realistic global model and real satellite observation data, additional issues emerge, including the nonlinearity and the non-Gaussianity of both the model background and the observations, the serious resolution-dependent biases between the model and the observations, and the unknown error statistics of the satellite precipitation data. Some of these issues have been discussed within the variational data assimilation framework, but there are also points that should be highlighted within the ensemble data assimilation. We compute from the point of view of the ensemble data assimilation, several statistics of the precipitation rate variable in both a series of the NCEP Global Forecasting System (GFS) 9-hour forecasts and the TMPA data. A proper interpretation of these statistical results would be essential to develop the strategy of ensemble precipitation assimilation with real observations.