AIRS Radiance Assimilation Background
One of the primary mission goals of the Atmospheric Infrared Sounder (AIRS) is to improve weather forecasting. The instrument provides high spectral resolution measurements of the thermal infrared spectrum, providing 2378 spectral channels from 3.74 to 15.4 µm. While other work at SPoRT focuses on the assimilation of retrieved profiles of temperature and moisture, work to assimilate direct radiance measurements has also been performed, eliminating the retrieval error from the total error of the observation and thus strengthening the impact of the observation on the analysis.
The impact of the assimilation of AIRS radiances in the framework of the National Centers for Environmental Prediction / Environmental Modeling Center (NCEP/EMC) operational North American Mesoscale (NAM) model at SPoRT, with cooperation and resources from the Joint Center for Satellite Data Assimilation (JCSDA) and NCEP/EMC, was investigated. Though the operational NAM runs to 84h, the focus of verification has been on the short-term (0-48h) forecasts as per the mission of SPoRT. The JCSDA has effectively shown that the use of AIRS measurements within an assimilation system can significantly improve medium range forecasts (Le Marshall et al. 2006) within the NCEP operational Global Forecast System (GFS).
The research performed investigated forecasts, run four times daily, from 9-16 April 2007. A control run is performed using all data operationally assimilated in the NAM data assimilation system. For the AIRS experiment, AIRS radiances were used in addition to that of the control. It is noted that the Advanced Microwave Sounding Unit (AMSU) onboard Aqua was not assimilated in either run. Assimilation is performed using the Gridpoint Statistical Interpolation (GSI) three dimensional variational (3D-Var) assimilation suite, which acts as the operational assimilation suite for both the GFS and the NAM at NCEP.
Results from the addition of AIRS to a system mimicking the operational NAM were positive. The incorporation of AIRS measurements resulted in the improved characterization of the troposphere in data void regions. The measurements were capable of detecting small-scale features in temperature and moisture in regions that are otherwise sparsely observed. By improving the initial analyses, the corresponding forecasts integrated from these initial states were also improved.
Figure 1. Height anomaly correlations for the control (black) and the AIRS experiment (red) at 500 hPa (top) and 1000 hPa (bottom) for forecasts spawned during 9-16 April 2007.
In considering the 500 hPa height anomaly correlations in Figure 1, a forecast improvement of 3h was observed by the addition of AIRS data to the data assimilation system. This improvement is defined as the time difference between the correlation of the AIRS forecasts to the corresponding analyses and the time at which the control has an equal correlation value. For all forecasts spawned in the experiment, forecasts were improved consistently at 48 hours through the troposphere, as also shown at 1000. These height anomaly correlations were performed over the continental United States.
The impact of including AIRS radiance measurements on precipitation forecasts was also considered. At 25mm/6h, which is roughly an inch of rain in a 6 hour period, the bias and the equitable threat scores were improved by 8% and 7% over the control, respectively, showing that the AIRS data were improving the forecast of the most extreme precipitation events. Though the AIRS experiment tended to have an increased bias towards the occurrence of precipitation below 25 mm per 6 hours, the equitable threat scores over these thresholds were improved at thresholds of 11 mm/6h and greater.
A CO2 sorting technique was developed and implemented to determine cloud contamination. Cloudy radiances were not assimilated in this work because the background fields and radiative transfer could not properly account for the effects of the cloud emission in the scene and the discontinuous nature of the cloud fields. The sorting technique was based on that developed by Holz et al. (2006). Initially, it was used to classify cloud top pressure, but in the current application, the tropospheric AIRS channel brightness temperatures in the 15 µm CO2 absorption region are used to identify channels not affected by clouds (SPoRT % Cloud Free Channels Product). AIRS channels which sense emission from the lower part of the troposphere will be affected by the presence of a mid-level cloud and measure colder temperatures than a cloud free spectrum of the same environment. The separation point between channels that are uncontaminated and ones affected by clouds is where the two sorted spectra diverge. Thus, channels colder than this point are not affected by the presence of clouds in the observed field of view. The separation point occurs at lower brightness temperatures for higher clouds, thus providing fewer channels uncontaminated by clouds, while low-level clouds will separate at higher brightness temperatures. The magnitude of the separation of the cloudy and clear sorted spectra, however, is a function of the effective cloud fraction, which is the product of the cloud emissivity and the physical cloud fraction of an instantaneous field of view (IFOV). Thus, the tuning of the algorithm to detect the separation point incorporates more advanced approaches than a simple brightness temperature separation threshold approach.
The method had been developed previously utilizing the entire AIRS spectrum of the 15 µm absorption continuum. The method, however, had to be adjusted to the 281 channel subset available in near-real-time for assimilation purposes. The technique, which is implemented within the GSI system, utilized a forward radiative transfer algorithm to determine the clear-sky IFOV. The implementation of the technique showed similar results to the cloud screening inherent in the GSI, but did not require the use of tangent linear or adjoint calculations.
