Impacts of Satellite Observation Data on the WRF 3D-Var System

Abstract

There have been many improvements in modeling (in terms of dynamics, physical parameterizations, and higher spatial resolution) and data assimilation (DA) techniques, aimed at improving the performance of numerical weather predictions (NWPs) but solving numerical model initialization problems in NWP has been an important issue for many years. Improving DA techniques and assimilating more observations, particularly from satellite platforms can help solve initialization problems. However, despite its importance, analysis and prediction remains a challenge in NWP because of the lack of traditional meteorological observations over the oceans. The optimal use of satellite data is crucial for improving the performance of NWP models. In this study, a total of 64 cases including control runs were designed and simulated using initial fields assimilated with Global Telecommunication System (GTS), Global positionaling system radio occultation (GPSRO) and Satellite Radiance observation data including Advanced Microwave Sounding Unit A (AMSU-A) and Microwave Humidity Sounder (MHS) in order to the impact of assimilation with the weather research and forecasting model (WRF) 3-dimensional variational method (3D-Var). Initial fields taken from 12h forecasting field on cyclic runs were used for the assimilation and control runs. Between 2,600 and 6,700 GTS observations were available for assimilation, and between 20% and 48% were used in each case. The spatial distribution of the GTS observations was based on the land. Between 3,000 and 6,200 GPSRO observations were available for assimilation, and between 6% and 12% were used in each case. The GPSRO observations were well distributed but the observational points were sparse. Between 10,500 and 17,500 satellite radiance observation were available for assimilation, between 26% and 33% were used in each case. The satellite radiance observations were well distributed over the whole research area. It was necessary to the correct observational bias before assimilating the satellite radiance observations because most satellite data are biased relative to background. Variational bias correction was used to achieve this in the present study. As a result, the satellite radiance observation were matched to the background fields without bias. The cost and gradient functions were minimized in assimilations in each case. The initial 500 hPa geopotential height assimilated run and control run fields were compared, which showed that Da had a great effect on initial fields in the model over the whole domain. Initial fields from the assimilation of GTS observations showed that the North Pacific high and the jet stream were stronger in each experiment than in the control run. Initial fields were higher with data assimilated only from GPSRO observations than in the control runs around the Pacific. Initial fields with data assimilated from GTS and GPSRO observations were similar to assimilated from GTS observations. Initial fields with data assimilated from satellite radiance observations made the jet stream extend strongly. Initial fields with data assimilated from GTS and satellite radiance observation had patterns similar to initial fielda with data assimilated from GTS observations, making the 500 hPa geopotential height decrease slightly over the whole domain. Root mean square error (RMSE) analysis showed that the forecast fields assimilated with GTS observations; GTS and GPSRO observations; GTS and satellite radiance observations; and GTS, GPSRO, and satellite radiance observations generally performed better than the control run forecast fields. However, the forecast fields assimilated with only GPSRO observations or satellite radiance observations performed more poorly than the control run forecast fields. In detail, the forecast fields assimilated with GTS and GPSRO observations performed better than those with GTS observations, except for over 12 h, and the forecast fields assimilated with GTS and satellite radiance observations performed better than the forecast fields assimilated with GTS observations after 24 h. This means that GPSRO and satellite radiance observation data are useful in DA when used together with GTS observations. The forecast field performance was better in an experiment assimilating the GTS, GPSRO, and satellite radiance observations than in assimilated runs with GTS and GPSRO observations after 12 h, and also better than assimilated runs with GTS and satellite radiance observations up to 24 h. Therefore, forecast fields assimilated with GTS, GPSRO, and satellite radiance observations were useful regardless of the forecasting time.