Impact of Multivariate Background Error Covariance Matrix in satellite radiance Assimilation

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Monday, 5 January 2015
Chandrasekar Radhakrishnan, IBM Research, New Delhi, New Delhi, India; and R. Mittal, V. Saxena, T. George, M. Dawar, Y. Sabharwal, J. P. Cipriani, and L. A. Treinish

The main objective of the present study is to improve the precipitation forecast over Brunei region through assimilating direct satellite radiance and Global Telecommunications System (GTS) conventional observational data using domain specific multivariate background error (BE) covariance matrix. In this work, Weather Research and Forecasting (WRF) model and three-dimensional variational (3DVAR) data assimilation system has been used. One of the main focus is to examine the sensitivity of the BE covariance matrix on the simulation of rainfall events. The radiance from Advanced Microwave Sounding Unit-A (AMSU A), AMSU B, Microwave Humidity Sounder (MHS) and Special Sensor Microwave Imager / Sounder (SSMIS) sensors channels are assimilated into WRF 3DVAR. The Community Radiative Transfer Model (CRTM) has been used as observational operator to simulate radiance. Some heavy rainfall events during 2013 - 2014 has chosen over Brunei region and simulated with and without assimilation.

This study has five set of experiments 1) control simulation (without assimilation) 2) assimilation of only GTS conventional observational data. Assimilation of satellite radiance and GTS with 3) NCEP global BE covariance matrix (CV3 option), 4) assimilation with domain specific BE covariance matrix (CV5 option) 5) assimilation with domain specific multivariate BE covariance matrix (CV6 option). The NMC method and a one month forecast difference has been used to generate the both CV5 and CV6 BE covariance matrix matrix.

Also a study has been done to analyze the channel sensitivity and to estimate the standard deviation (SD) and relative mean square error (RMSE) for all channels. These sensitivity studies showed that the channels 1 to 3 in AMSU-A has more SD and these channels are highly sensitive with surface temperature. This is due to more uncertainly in estimating surface emissivity from CRTM emissivity model and land surface temperature from first guess (GFS). Therefore in aforementioned experiments, surface temperature channels (1- 4) in AMSU-A are not assimilated and the radiances from the Tropospheric and Stratospheric temperature channels (5-10) in AMSU-A are assimilated. Similarly the high RMSE and SD channels from the MHS and SSMI sensors are identified and excluded from the assimilation system.

The simulated rain fall has been compared and validated with Automatic Weather Station (AWS) rain gauges data located over Brunei region. The spatial distribution of simulated rainfall has been compared with TRMM 3 hourly accumulated rainfall product (3B42) . The preliminary results showed that the assimilation of both AMSU-A and GTS with global BE covariance matrix (CV3) over estimate the 12 hr accumulated rain fall. It was found that the reduction in the variance and horizontal length scales in the global BE matrices have significant improvements in the simulation of rainfall episodes. Similar findings were obtained for other two BE options, although theses options are less sensitive to the changes in length scales.