AIRS Profile Assimilation Background
The SPoRT center seeks to improve short-term weather forecasts, especially in the data-sparse regions, by the use of satellite-based observations. The Atmospheric Infrared Sounder (AIRS), aboard the EOS polar-orbiting Aqua satellite, provides near-radiosonde-quality atmospheric temperature profiles with the ability to resolve some small-scale vertical features in both clear and partly cloudy scenes.
AIRS profile assimilation began at SPoRT in 2004 using the Local Analysis and Prediction System (LAPS) to initialize the MM5 but has since evolved to using the Advanced Research Weather Research & Forecasting (ARW) system and its analysis component, the WRF-Var. The evolution of the work and some lessons learned can be seen in Chou et al. (2006) and Zavodsky et al. (2007). Two key lessons learned from previous work have been applied for use in the WRF-Var system: 1) effectively use the quality indicators to select only the highest quality observations and 2) assimilate the over-land and over-water soundings separately with different error characteristics to take into account emissivity issues that lead to degraded soundings over-land. Currently, SPoRT uses the latest version of JPL's AIRS retrieval algorithm (version 5) for this work. An overview of these profiles can be found at the AIRS Profile Background page. To separate the AIRS profiles into over-land and over-water soundings, changes to the WRF-Var source code were made to add AIRS-Water and AIRS-Land data set with observation errors based on estimates cited in Tobin et al. (2006).
Sample three-dimensional distribution of AIRS profile data. The black points represent the highest quality data. Each colored point denotes the maximum pressure level corresponding to the level above which data are of good quality. The red rectangle denotes the bounds of the WRF domain.
The WRF forecast domain covers the contiguous United States, western Atlantic Ocean, and Gulf of Mexico. Each WRF forecast is initialized from a cold start using the North American Mesoscale (NAM) analysis at 0000 UTC, and its boundary conditions are updated every 3 hours using the NAM forecasts. The forecast/assimilation cycle runs several hours until the AIRS overpass time is reached (usually around 07-09Z for the morning pass). The WRF forecast is then used as the background field for the WRF-Var analysis. The forecast continues with the AIRS-enhanced WRF-Var analysis as the new initial state for a total forecast up to 48 hours.
Besides the observations and background field, one of the major components in the WRF-Var system is the background error covariance matrix (B matrix). Correct use of the B matrix is important in determining the appropriate weighting between the background field and observations as well as how information contained in observations is spread horizontally and vertically. Optimal analysis configuration desires background errors that are consistent with the domain/grid spacing, the model used as the background, and the season. A B matrix was calculated using the National Meterological Center (NMC) method (Parrish and Derber 1992), which takes differences between multiple 12- and 24-hour forecasts to determine model error. Within the WRF-Var system, the B matrix is generated using the "gen_be" program. In our application, short-term WRF forecasts for a two week period (17 to 31 January 2007) were used to run the NMC method. For more information on the B matrix used for this work, see Chou et al. 2008 (LINK WRF Workshop).
Analysis impact of AIRS on 700 hPa temperature. The difference between the AIRS and the background field is shown in b) resulting in the analyses in c) and d). Part c) shows the analysis increment with the original length scale that has obvious bull's eyes and streaking, while d) shows the impact of tuning the length scale to remove some of those smaller scale features.
The figure above shows preliminary results for 17 January 2007. Two swaths of AIRS profiles are used: one along the East Coast and the other over the Midwest. Figure b) depicts the temperature difference between the AIRS profile and model background at 700 hPa and shows that AIRS is cooler than the background over Florida and the Great Lakes and warmer over the Southeast US. The analysis increment (difference between the analysis and background) in Figure c) shows a similar pattern but with bull's eyes and stripping features, especially over KS and MO. The way that the analysis draws tightly to each observation indicates that the original horizontal length scale may be too small. Tests were conducted using a WRF-Var tuning factor, which adjusts the spread of analysis variables by multiplying the length scale by a prescribed value. Subsequently, it was determined that increasing the length scale by 50% led to an optimal configuration that smoothed the bull's eyes and stripping features without compromising analysis fidelity. Figure d) shows the magnitude and horizontal spread of the AIRS observations on the 700 hPa temperature analysis using the new length scale. Similar tests have been conducted for the moisture analysis, and it was determined that doubling the moisture length scale yields a satisfactory result.
Profiles of temperature (solid) and dew point (dashed) near Greensboro, NC (GSO) radiosonde location for 0800 UTC 17 January 2007. The background (black) and WRF-Var (red) profiles are for the nearest grid point. The AIRS profile (blue) is for the highest-quality retrieval closest to the grid point. The radiosonde (green) is a linear interpolation of the 0000 and 1200 UTC soundings to 0800 UTC.
The impact of the AIRS profiles on the WRF-Var analysis was also examined by comparing collocated soundings profiles of the short-term WRF-forecast background, AIRS profiles, and WRF-Var analysis near several radiosonde stations. In general, the temperature and moisture soundings of the AIRS-enhanced analyses lie between those of the background and AIRS profiles as it should for proper data assimilation. The inclusion of AIRS also produces a superior analysis to the background when compared to the radiosonde. Results indicate that AIRS profiles produce an analysis closer to in situ observations than the background field, which should lead to improved initial conditions and better forecasts when used to initialize a model forecast. Future work will focus on conducting model simulations using the AIRS-enhanced initial conditions for short-term (0-48h) regional ARW runs. These forecasts will be verified against in situ observations, and if superior to control forecasts will be transitioned to SPoRT's WFO partners for their local WRF runs.
