News & Updates
January 4-8, 2015
JCSDA News and Newsletters
GSI Goes Operational at the Air Force Weather Agency
From the June 2014 issue of the JCSDA Quarterly
The Air Force Weather Agency (AFWA) has been utilizing the Weather Research & Forecasting (WRF) model as its numerical weather prediction model of choice since 2007. From the first operational WRF model run, AFWA has employed the three-dimensional variational analysis (3DVAR) data assimilation (DA) system developed at the National Center for Atmospheric Research (NCAR) to produce the initial conditions for the forecast model. In 2012, AFWA began preparing its infrastructure to transition its DA system to the Gridpoint Statistical Interpolation (GSI) DA system developed at the National Centers for Environmental Prediction (NCEP) and used operationally in their suite of models. AFWA's Initial Operational Capability (IOC) was met when the first operational domain began using GSI in July 2013. By the end of 2013, all of AFWA's regional modeling efforts had transitioned to GSI.
Figure 1. GO index plot of AFWA's operational hemispheric domain, using GSI with full radiance assimilation, compared to the North American Model over the contiguous United States. In this plot, a GO index value > 1 indicates that the use of GSI is better. More info on the GO index can be found here: http://www.dtcenter.org/eval/meso_mod/afwa_test/go_index/GO_Index_Description.pdf (Courtesy Mike McAtee, The Aerospace Corporation)
At the time of IOC, AFWA began assimilating data from AMSU, HIRS, MHS, IASI, C-NOFS, and COSMIC, in addition to conventional observations. AFWA was able to directly assimilate radiance data for the first time. These data are critically important to meet AFWA's requirement to support its global mission. The transition to GSI immediately allowed AFWA to increase the number of observations assimilated by as many as 5-10 times, depending on the domain. Additionally, in a March 2014 GSI implementation, AFWA added the capability to assimilate CrIS and ATMS from Suomi-NPP, AMSU, MHS, and ASCAT from MetOP-B, as well as TerraSAR-X. As it currently stands, AFWA assimilates approximately 5 million radiance observations per day. Over the next several months AFWA will continue to add the capability to assimilate additional data.
Figure 2. FSO results averaged over our Southwest Asia for 1-20 March 14 broken out by platform showing the relative importance of each observation type (left) and the number of each type of observation (right). (Courtesy Mike McAtee, The Aerospace Corporation)
The requirement of AFWA to provide timely and accurate weather intelligence to support our unique and complex global mission precludes the reliance on external models such as the Global Forecast System model, the United Kingdom Meteorological Office Unified Model (UM), or the North American Model to meet our specific needs. At AFWA, a number of regional domains are available with principal theaters running 4 times per day and others running 2 times per day. These domains vary in horizontal resolution from 45 km to 1.67 km and are positioned over specific areas of interest across the globe, many of which are in data sparse locations. In addition to our static theaters, we have several theaters that can be moved as needed to meet rapidly changing demands. The WRF model is initialized using the UM. The UM model is run 4 times per day and pre-processed for initialization conditions to each WRF grid. The WRF model also incorporates surface characterizations from the Land Information System (LIS) and the Navy Sea Surface Temperature analysis. DA is run 8 times per day for the core theaters and 4 times per day for the others (i.e., twice per model cycle run, once for a 6-hour "spin-up" run and once for subsequent "free forecast"). This is done to incorporate the maximum number of available observations. Surface, upper air and aircraft observations are the primary conventional observations used, while Global Positioning Satellite-Radio Occultation, satellite cloud drift wind, radiance observations, and sea surface wind speed are especially important in regions where these conventional data are limited.
Using the General Operations (GO) index to as a measure of relative performance, we have seen a slow but steady improvement in our 20 km hemispheric domain since our most recent implementation (Figure 1).
Figure 3. FSO results averaged over our Southwest Asia for the 1-20 March 2014 period broken out by sensor.
Additionally, AFWA uses the adjoint-based Forecast Sensitivity to Observations (FSO) tool developed at NCAR to assist us in monitoring and tailoring our DA system to maximize its utility. A look at the results for our Southwest Asia domain for a few weeks in March shows that the assimilation of radiance data provides a significant reduction in the 12 hour forecast model error (Figures 2 and 3).Jason Martinelli, AFWA