Assimilation of F-16 Special Sensor Microwave Imager/Sounder Data in the NCEP Global Forecast System

Banghua Yan NOAA/NESDIS/Office of Satellite and Product Operations, Camp Springs, Maryland

Search for other papers by Banghua Yan in
Current site
Google Scholar
PubMed
Close
and
Fuzhong Weng NOAA/NESDIS/Center for Satellite Applications and Research, Camp Springs, Maryland

Search for other papers by Fuzhong Weng in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

The Special Sensor Microwave Imager/Sounder (SSMIS) on board the Defense Meteorological Satellite Program (DMSP) F-16 satellite is the first conically scanning sounding instrument that provides information on atmospheric temperature and water vapor profiles. The SSMIS data were preprocessed by the Naval Research Laboratory (NRL) using its Unified Preprocessor Package (UPP) and then distributed to the numerical weather prediction centers by the Fleet Numerical Meteorology and Oceanography Center (FNMOC). This dataset was assimilated into the Global Forecast System (GFS) using gridpoint statistical interpolation (GSI). The initial assimilation of the SSMIS data into the GFS did not improve the medium-range (5–7 days) forecast skill. The SSMIS bias (O-B) still changes with location and time after the GSI bias-correction scheme is implemented. This bias characteristic is related to residual calibration errors in the correction of the SSMIS antenna emission and warm target contamination. The large O-B standard deviation is probably due to the large instrument noise in the SSMIS UPP data. The large O-B and its standard deviation for several surface sensitive channels are also caused by uncertainty in surface emissivity. In this study, a new scheme is developed to remove regionally dependent bias using a weekly composite O-B. The SSMIS noise is reduced through a Gaussian function filter. A new emissivity database for snow and sea ice is developed for the SSMIS surface sensitive channels. After applying these algorithms, the quality of the SSMIS low-atmospheric sounding (LAS) data is improved; the surface-sensitive channels can be effectively assimilated, and the impacts of SSMIS LAS data on the medium-range forecast in the GFS are positive and similar to those from Advanced Microwave Sounding Unit-A (AMSU-A) data.

Corresponding author address: Dr. Banghua Yan, NOAA/NESDIS/Office of Satellite and Product Operations, 5200 Auth Road, Room 510, Camp Springs, MD 20746. E-mail: banghua.yan@noaa.gov

Abstract

The Special Sensor Microwave Imager/Sounder (SSMIS) on board the Defense Meteorological Satellite Program (DMSP) F-16 satellite is the first conically scanning sounding instrument that provides information on atmospheric temperature and water vapor profiles. The SSMIS data were preprocessed by the Naval Research Laboratory (NRL) using its Unified Preprocessor Package (UPP) and then distributed to the numerical weather prediction centers by the Fleet Numerical Meteorology and Oceanography Center (FNMOC). This dataset was assimilated into the Global Forecast System (GFS) using gridpoint statistical interpolation (GSI). The initial assimilation of the SSMIS data into the GFS did not improve the medium-range (5–7 days) forecast skill. The SSMIS bias (O-B) still changes with location and time after the GSI bias-correction scheme is implemented. This bias characteristic is related to residual calibration errors in the correction of the SSMIS antenna emission and warm target contamination. The large O-B standard deviation is probably due to the large instrument noise in the SSMIS UPP data. The large O-B and its standard deviation for several surface sensitive channels are also caused by uncertainty in surface emissivity. In this study, a new scheme is developed to remove regionally dependent bias using a weekly composite O-B. The SSMIS noise is reduced through a Gaussian function filter. A new emissivity database for snow and sea ice is developed for the SSMIS surface sensitive channels. After applying these algorithms, the quality of the SSMIS low-atmospheric sounding (LAS) data is improved; the surface-sensitive channels can be effectively assimilated, and the impacts of SSMIS LAS data on the medium-range forecast in the GFS are positive and similar to those from Advanced Microwave Sounding Unit-A (AMSU-A) data.

Corresponding author address: Dr. Banghua Yan, NOAA/NESDIS/Office of Satellite and Product Operations, 5200 Auth Road, Room 510, Camp Springs, MD 20746. E-mail: banghua.yan@noaa.gov

1. Introduction

The Special Sensor Microwave Imager/Sounder (SSMIS), which measures the thermally emitted radiation from the earth at 24 channels from 19 to 183 GHz (see Table 1), is the first conically scanning microwave sensor to provide temperature and water vapor sounding information. Today, there are three SSMIS instruments flown aboard the Defense Meteorological Satellite Program (DMSP) F-16, F-17, and F-18 platforms. In the next decade, there will be two more SSMIS instruments flown on the F-19 and F-20 satellites, which will be launched in 2013 and 2015, respectively. The low-atmospheric temperature sounding (LAS) channels in SSMIS are similar to those of the cross-track scanning Advanced Microwave Sounding Unit-A (AMSU-A) instrument on board the National Oceanic and Atmospheric Administration (NOAA) and Meteorological Operations (MetOp) satellites. It is thus expected that the SSMIS LAS data would have positive impacts on forecast skills similar to those of the AMSU-A data. At NWP centers [e.g., the Met Office, European Centre for Medium-Range Weather Forecasts (ECMWF), and Naval Research Laboratory (NRL)], a series of assimilation experiments were made on F-16 SSMIS LAS data with a neutral-to-small positive impact on the medium-range forecast being found in the Southern Hemisphere and the neutral impacts in the Northern Hemisphere (Bell et al. 2008). At the National Centers for Environmental Prediction (NCEP), the Global Forecast System (GFS), an assimilation experiment was conducted on F-16 SSMIS LAS data in early 2007 and a neutral impact was also found on the global medium-range forecast (Kazumori 2007). Recently, F-18 SSMIS data were assimilated into the U.S. Navy 4D Variational Analysis System with a significant positive impact on NRL global forecast (Swadley et al. 2010). This positive impact is primarily due to several factors: 1) improved calibration of the LAS data from the F-18, 2) use of the data at several upper-atmospheric sounding (UAS) channels (i.e., channels 19 ~ 24 in Table 1), and 3) the 4D variational system. This effort more accurately represents the best accumulated impacts of SSMIS data from both LAS and UAS channels thus far. The NCEP GSI is a 3D variational analysis system (Parrish and Derber 1992; Derber and Wu 1998) and its upper level is not high enough to assimilate the SSMIS data at all the UAS channels. Today, the independent impact of SSMIS LAS data in GFS and other NWP models has not been clearly demonstrated due to some other issues such as a lack of a bias-correction scheme and a preprocessor for SSMIS data.

Table 1.

Channel characteristics of the F-16 SSMIS sensor (Poe et al. 2001).

Table 1.

The F-16 SSMIS observations at the LAS channels display radiance anomalies due to the presence of antenna emission and solar intrusion into the calibration target. The main reflector can emit radiation at 50–60 GHz as high as 3 K in equivalent brightness temperature. Solar intrusion into the calibration target can cause a 1.5-K uncertainty (Swadley et al. 2005; Weng et al. 2005b; Bell et al. 2005; Kunkee et al. 2008; Yan and Weng 2009). Currently, the anomaly corrected F-16 SSMIS data are released to the NWP community by applying the NRL–Met Office Unified Preprocessor Package (UPP; Bell et al. 2008). The quality of F-16 UPP data has been significantly improved compared to the original measurements. However, residual errors still exist over some specific areas. The F-17 SSMIS observations display even larger anomalies than those from F-16 due to the reflector emission at the LAS channels and solar array shadow (Swadley et al. 2010). The F-18 SSMIS observations have better quality due to improved antenna and calibration subsystems (Yan et al. 2010).

In addition to a regionally dependent bias, the noise of the F-16 SSMIS UPP data is also significant and can be as high as 0.5 K. Prior to 14 August 2008, a noise reduction algorithm (Bell et al. 2008) was developed and applied to F-16 data but it was soon removed from the UPP process. The noise reduction process was not further applied to F-17 and F-18 SSMIS radiance data and there remains a similar level of noise in all three SSMIS instruments. A proper bias correction and noise reduction must be applied to SSMIS radiance data from F-16 to F-18 satellites prior to the data assimilation process.

In satellite data assimilation, microwave emissivity models are important for the assimilation of surface-sensitive channels. With the microwave emissivity models used in forward calculations, more microwave satellite data are being assimilated into global medium-range forecast systems (Wiesmann and Mätzler 1999; Weng et al. 2001; Andreadis et al. 2008; Wójcik et al. 2008). For instance, the Microwave Land Emissivity Model (MELM) developed by Weng et al. (2001) has been used with the NCEP GFS through the Joint Center for Satellite Data Assimilation (JCSDA) Community Radiative Transfer Model (CRTM; Weng et al. 2005a; Han et al. 2006, 2009). The introduction of the MLEM into the NCEP assimilation system has significantly increased the utility of satellite microwave data over most land conditions (Yan and Weng 2011). However, the MLEM model displays a large uncertainty (about 0.05) over snowy surfaces. Due to the lack of a reliable sea ice emissivity physical model, a constant of 0.9 was typically used over sea ice for SSMIS applications in GFS. The actual sea ice emissivity could depart significantly from this default value (Hewison and English 1999). An uncertainty of 0.05 may cause an SSMIS bias (O-B) ranging within several degrees Kelvin for surface-sensitive channels depending on frequency (Yan and Weng 2011). Such a difference can lead to the rejection of many useful measurements in the data assimilation processes. Improved snow and sea ice emissivity models are required in the radiance calculations.

To further improve the assimilation impacts of the SSMIS data, new algorithms are developed for the correction of regionally dependent bias, radiance noise reduction, and snow and sea ice emissivity calculations for assimilation of F-16 UPP LAS data. The quality control scheme in GFS is also revised for SSMIS data assimilation. A series of numerical experiments using the GFS is carried out to test the performance of these new algorithms for assimilating F-16 UPP data at the LAS channels. Specifically, the use of SSMIS LAS data at several surface-sensitive channels and the impact of the LAS data on global medium-range forecasts are assessed.

2. Satellite data assimilation procedure

a. Variational approach

Given a set of observations from satellites, radiosondes, and ground-based measurements, a 3D or 4D variational system can produce a “best” analysis of the atmospheric state at desired resolutions in a statistically “optimal” way by assimilating them into an NWP model. Under the assumptions that the observation errors are nonbiased and follow Gaussian distributions, the best analysis X (state of the atmosphere) can be obtained by minimizing a cost function of J(X) (Parrish and Derber 1992; Ide et al. 1997; Derber and Wu 1998):
e1
where Xb is the background, is the background error covariance matrix, H the forward model (e.g., radiative transfer model for satellite radiance), Y0 the observations, the instrument error covariance matrix, the representativeness error covariance matrix, and Jc the constraint term. An important computation involves simulating satellite-measured brightness temperatures [i.e., H(X)] at various channels with surface emissivity information through the radiative transfer model (RTM). The differences between the simulated and observed brightness temperatures [i.e., H(X) – Y0] contribute to the solution of the analysis X through Eq. (1). Note that the differences are also affected by the uncertainties in the satellite-observed and RTM-simulated brightness temperatures. For brightness temperatures at window and surface-sensitive channels that are characterized by surface properties, the errors in simulated brightness temperatures are significantly affected by errors in surface emissivity. Thus, any uncertainty in satellite radiance data and surface emissivity can further degrade the quality of the analysis variables, X.

b. SSMIS bias characteristics in GFS

A powerful approach for monitoring the quality of satellite brightness temperatures is to compare observed radiances with simulations {see the item [H(X) – Y0] in Eq. (1) above}. Ideally, this difference is determined by the RTM simulation error (e.g., model uncertainty and atmospheric information uncertainty) and satellite observation error (e.g., random noise and calibration anomaly in the data). At microwave frequencies, brightness temperatures at 54–59 GHz under clear atmospheres can be well simulated since the main gaseous absorption is caused by atmospheric oxygen whose concentration is very stable. For example, an error of 1 K in the physical temperature can result in an error of 0.1 K in the simulated brightness temperature (Poe et al. 2001). In this study, the Community Radiative Transfer Model (CRTM; Weng et al. 2005a) is used to simulate the brightness temperatures at SSMIS channels. The parameters input into CRTM include the temperature and water vapor profiles, which are obtained from the NCEP Global Data Assimilation System (GDAS) analysis. The root-mean-square error of the GDAS temperature profiles against radiosonde measurements below 10 mb is smaller than 2 K, which results in an uncertainty of about 0.2 K in simulations from 54 to 59 GHz. Note that the errors in simulated brightness temperatures are less affected by the errors in the GDAS water vapor profile. Also, the effects of cloud and rain are not taken into account in these simulations since cloud-contaminated data are not used because of a lack of information in the GDAS data. Thus, any bias beyond an uncertainty of 0.2 K at these four channels can be attributed to the other error sources such as instrument calibration.

Figures 1a and 1b display distributions of brightness temperature differences (ΔTB) on 5 August 2008 for F-16 UPP at 54.4 and 55.5 GHz, respectively. For a comparison, the biases at these two frequencies in the data from the AMSU-A onboard the Meteorological Operational Satellite Programme (MetOp) satellite are also shown in Figs. 1c and 1d, respectively. We can see that the biases from the SSMIS UPP data are dependent on region (e.g., latitude and satellite orbit node). The biases are persistently observed on different days. Note that the results in the figures are generated after applying the original GFS bias-correction (BC) algorithm (Derber and Wu 1998). This implies that the above regional biases in F-16 SSMIS data cannot be removed by the GFS BC algorithm. An additional BC algorithm must be developed.

Fig. 1.
Fig. 1.

Distribution of brightness temperature differences on 5 Aug 2008 for F-16 UPP at (a) 54.4 and (b) 55.5 GHz, and for MetOp-A AMSU-A at (c) 54.4 and (d) 55.5 GHz, which are calculated using the data passing the current GFS quality control test after the GFS bias correction (Derber and Wu 1998) is applied. Here, std. represents the standard deviation of the brightness temperature differences.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

c. F-16 UPP data noise characteristics

The SSMIS instrument oversamples the upwelling radiance in its along-track direction (Bell et al. 2008). All SSMIS channels are sampled every scan and a sampling period of 4.22 ms is used for each field of view (FOV). This integration in time has reduced the data noise significantly but it is not long enough for data applications in NWP models. To reduce the noise, an averaging algorithm consisting of a 2D Gaussian weighting approach was developed by Bell et al. (2008) and applied to the F-16 UPP data. The ΔTB uncertainty (the standard deviation of ΔTB) of the SSMIS LAS data in Figs. 1a and 1b is only about 0.2 K. This uncertainty is determined by the standard deviations due to random noise in the radiance data, short-range forecast errors, RTM uncertainty, and the remaining calibration anomaly in the radiance. At the 54.4-, 55.5-, 57.3-, and 59.4-GHz channels, the error in the short-range forecasts can result in an uncertainty of up to 0.2 K in radiance simulations (see section 2b above). This makes it difficult to quantify the magnitude of the random noise in the accumulated ΔTB uncertainty of 0.2 K. It is believed that the random noise of the UPP LAS data on 5 August 2008 may be lower than 0.2 K.

The noise reduction processing in the UPP was stopped after 14 August 2008. Figures 2a and 2b show the distributions of ΔTB for F-16 UPP data on 28 August 2008 for 54.4 and 55.5 GHz, respectively. There is no noise reduction applied to this dataset. Compared with the results in Figs. 1a and 1b, the ΔTB uncertainty of the data after 14 August 2008 increases due to a lack of noise reduction processing. For example, the standard deviations of brightness temperatures at these channels are approximately 0.4 K. Also, the values of the standard deviations at 57.3 and 59.4 GHz are more than 0.5 K (figures not shown). Table 2 lists the standard deviations of the UPP data for ΔTB at the seven LAS channels, which are computed using the data from 16 August through 31 December 2009. The results marked NR in Table 2 indicate there was a noise reduction, which will be discussed in section 3b. The statistical results in Table 2 are similar to those we derived from the data on 28 August 2008. The errors at the 54.4- and 55.5-GHz channels are the smallest compared to those at other LAS channels. The larger errors, at 50.3 and 52.8 GHz compared to those from 53.6 to 57.3 GHz, are due primarily to the uncertainty in the surface emissivity. The errors at 57.3 and 59.4 GHz are similar to or large than those at 52.8 and 53.6 GHz, which is primarily caused by their larger calibration (residual) anomalies (Yan and Weng 2009). Normally, it is expected that radiances with a noise magnitude on an order of 0.3 K or better can improve the analysis fields and produce better forecasts (Bell et al. 2008, 2010), which may be observed in the MetOp-A and NOAA-18 AMSU-A tropospheric temperature sounding channels (see Figs. 1c and 1d). Therefore, the uncertainty in the UPP LAS data needs to be reduced to an acceptable level through a proper noise reduction processing.

Fig. 2.
Fig. 2.

F-16 UPP brightness temperature difference maps on 28 Aug 2008 at (a) 54.4 and (b) 55.5 GHz, where no noise reduction algorithm is applied. As in Fig. 1, std. represents the standard deviation of the brightness temperature differences.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

Table 2.

Standard deviations of the brightness temperature differences (ΔTB) from channels 1 to 7 between backgrounds from the CRTM and satellite observations without and with NR processing. The results for channels above 52.8 GHz are computed using the data over global areas; the results at 52.8 GHz are computed using the data under free rainy-clouds (see section 4a) and over global areas above 700 mb; the results at 50.3 GHz are computed using the data under rain-free conditions and over oceans.

Table 2.

d. Microwave land and sea ice emissivity models

The Microwave Land Emissivity Model (MLEM) was developed using a two-stream approximation that characterizes the emission and scattering processes of various land surfaces such as snow cover, desert, and vegetation (Weng et al. 2001). Currently, this model has demonstrated some significant impacts on the assimilation of various satellite microwave data in GFS, especially over nonscattering surfaces. For a constant emissivity value, about 20% of satellite microwave data at window and surface-sensitive channels over land are used. With the land emissivity model, 30% more satellite data can be assimilated. However, the current MLEM has a large bias over snow conditions. Our analysis shows that the MLEM emissivity simulations at the LAS channels over snow have a mean error of 0.04 (figure not shown), which could cause an uncertainty of a few degrees Kelvin in the simulated brightness temperatures. Over sea ice conditions, the emissivity model has not been developed for the assimilation of microwave data. In the past, a constant (0.9) was used for sea ice emissivity for all microwave data in GFS. In this study, snow and sea ice emissivity models are further developed. To measure the impacts of the surface emissivity model on our data assimilation system, we use a ratio of the data passing a series of quality controls to the total volume of data ingested into the assimilation system, which is called the utilization rate.

3. Improvements in SSMIS data quality and new developments

a. Bias-correction algorithm

For the current UPP data, the biases are persistently high at some regions and do not change on a weekly base. Thus, we generated the weekly composite O-B biases in terms of the nodes and latitudes (θ) of the satellite observations, . At a given time and location, the brightness temperature will be corrected through
e2
where represents the original UPP brightness temperatures. The O-B bias represents the latitudinal-mean O-B biases during a week. Since some of the LAS channels (e.g., 50.3 and 52.8 GHz) are affected by clouds and/or surface emissivity, a quality control procedure must be developed. For the LAS channel at 52.8 GHz, which is affected by raining clouds, a threshold approach is used to detect cloud contamination, depending on surface type (see section 4a). This channel is also sensitive to surface emissivity, especially for high-elevation terrain. Thus, is derived from all the data over the land where the surface pressure is greater than 700 mb.

Figures 3a and 3b display the variations in the daily averaged biases at channels 4 (54.4 GHz) and 5 (55.5 GHz) from 5 to 11 August 2008. The major features of the O-B biases vary slowly with day, including the locations and magnitudes of the maximum and minimum biases, and the magnitude of the longitudinal average of the O-B biases with latitude. The slow change in ΔTB with time is related to the pattern of the original calibration anomaly in F-16 (Swadley et al. 2008; Yan and Weng 2009). As a result, the features of weekly averaged do not significantly deviate from those of daily averaged biases within that week. Figures 3c and 3d display time series of the weekly averaged ΔTB (green color) at the same channels for a few weeks from 5 August to 30 September. Although the regionally dependent biases exist at each channel during 2 months, their locations and magnitude have a small change from week to week. Therefore, the weekly averaged is used to correct the residual biases for each day. Using the correction in Eq. (2) to the original UPP data, the time series of the longitudinal-mean ΔTB at 54.4 and 55.5 GHz is replotted in Figs. 3c and 3d and the bias is more uniform than the original data. After this analysis is applied to other LAS channels except for 50.3 GHz, the same conclusion is obtained (figure not shown). Note that the 50.3-GHz channel is not analyzed here due to a large uncertainty in the simulated brightness temperature caused by surface emissivity uncertainty.

Fig. 3.
Fig. 3.

Time series of the daily or weekly averages of the longitudinal-mean brightness temperature (TB) differences at 54.4 and 55.5 GHz. (a) Daily average of longitudinal-mean TB differences at 54.4 GHz for 7 days from 5 to 11 Aug 2011. (b) As in (a), but for 55.5 GHz. (c) Weekly average of longitudinal-mean TB differences at 54.4 GHz for 8 weeks from 5 Aug to 30 Sep, which include the results with (dark gray) and without (light gray) the newly developed BC algorithm. (d) As in (c), but at 55.5 GHz. Here, all TB differences have been plotted by adding the integer number of each overlapping dashed line for clarity.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

b. Noise reduction algorithm

Our algorithm for noise reduction is based on a 2D Gaussian weighting proposed by Bell et al. (2008) and Yan and Weng (2009) and is used to produce the new sensor data record (SDR) brightness temperatures from the original F-16 SSMIS brightness temperatures; that is,
e3
where . In Eq. (3), is the noise-reduced or resampled brightness temperature at the location indexed by scan line l and scan position p, wi(p) is a weight assigned to the neighborhood i to the scan position p, TB(l + δi, p) is the brightness temperature at the position indexed by scan line (l + δip) and scan position p, N is the number of the nearest neighbors included in the domain size for the average, and σ represents an average scale of the footprint area. More details are provided in Bell et al. (2008). In this study, σ = 25 km; the domain size is selected as 4 × 4; that is, the maximum number in the latitudinal and longitudinal directions is four pixels. Thus, the largest N value is 16. This is because the pixels outside this domain do not contribute to the radiance according to Eq. (3). Also, increasing domain size (also N) would help reduce the noise of the data but would increase the computational costs. Most significantly, the spatial averaging over a large domain could smooth out useful information, especially over inhomogeneous surface and atmospheric conditions.

Figures 4a and 4b display the distributions of the F-16 UPP biases (ΔTB) on 28 August 2008 for 54.4 and 55.5 GHz, respectively, with the new noise reduction algorithm incorporated. Compared with Fig. 2, the biases are reduced to 0.23 K for these two channels. Table 2 provides the standard deviation of the brightness temperature differences at the LAS channels from 50.3 to 59.4 GHz after incorporating the noise reduction algorithm (see the column with NR).

Fig. 4.
Fig. 4.

As in Fig. 2, but for application of the noise reduction algorithm.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

c. Snow and sea ice emissivity regression algorithm

The snow and sea ice emissivity regression algorithms were developed early in 2004 (Yan et al. 2004). A lookup table (LUT) of snow and sea ice emissivity at a few discrete microwave windows and several discriminator indices (DIs) was derived a priori (see appendix C in Yan et al. 2008a). The LUT is derived from AMSU snow and sea ice emissivity retrievals under clear-sky conditions, ground-based snow emissivity measurements from Mätzler (1994), and aircraft-based sea ice emissivity measurements from Hewison and English (1999). The LUT includes surface emissivity spectra for 16 snow types and 13 sea ice types at the local zenith angle (LZA) of 50°, covering from 5 to 150 GHz for snow and 6.7 to 157 GHz for sea ice. In Fig. 5a, the five new spectra are labeled as being radiometric snow (RS) type, which implies a distinct emissivity spectrum but cannot be directly linked to a physical snow type, while in Fig. 5b, the six new spectra are labeled as radiometric sea ice (RS).

Fig. 5.
Fig. 5.

(a) Microwave snow emissivity as a function of frequency across the range between 4.9 and 150 GHz, where the five new spectra are labeled with RS, which implies a distinct emissivity spectrum but cannot be directly associated with a physical snow type while the other snow types followed the classification of snow types in Mätzler (1994). (b) Microwave sea ice emissivity as a function of frequency across the range between 6.7 and 157 GHz, where the six new spectra are labeled with RS and have similar meanings as in (a) but for sea ice while the other sea ice types followed the classification of Hewison and English (1999).

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

The DIs in Yan et al. (2008b) are used to identify a unique emissivity spectrum that should mimic the spectral feature of the realistic snow or sea ice emissivity from the LUT. The emissivity at the same LZA for a frequency not included in the LUT is directly interpolated from LUT based on the selected spectrum with a proper offset. The offset is defined as the emissivity difference between the DI value and the emissivity calculated using the recognized spectrum from the LUT. The emissivity at other viewing angles is interpolated from the angular-dependent relationship derived from the physical emissivity model such as MLEM. Currently, this methodology has been successfully applied to data assimilation of microwave sensors such as AMSU-A/B, the Microwave Humidity Sounder (MHS), SSMI, and the Advanced Microwave Scanning Radiometer (AMSR-E) in GFS. In this study, the same methodology is used for the SSMIS sensor to identify three SSMIS-derived DIs (see the appendix).

The performance of the above snow and sea ice emissivity algorithm is tested using the satellite-retrieved emissivity through RTM under clear-sky conditions as described in Yan and Weng (2011). The emissivity retrieval data in 2008 are selected every other week and the standard deviations of the emissivity between the satellite retrievals and the regression algorithm are smaller than 0.02 for the AMSU-A/B window channels from 23.8 to 150 GHz. This value for the regression algorithm is much smaller than the standard deviation of the emissivity model, which is 0.05.

4. Assimilation of SSMIS F-16 LAS data in GFS

a. Improved quality control

A QC scheme with a series of criteria is applied to ingested satellite data for detecting cloudy radiances, uncertainty in forward calculations, gross error, and the weighting factor of the data, as described for AMSU-A observations in Yan and Weng (2011). Detecting cloud-affected data is most important. In the current GFS, the cloud liquid water path (CLW) algorithm in Weng and Grody (1994) is used to estimate a CLW value for the SSMI observations over the oceans. The threshold used to detect the cloud-affected data is 0.2 kg m−2, which is similar to that used for other microwave observations in GFS. Note that this threshold can only remove those data that are highly affected by nearly precipitating clouds. A smaller threshold should be used if all data affected by clouds need to be rejected. To apply the SSMI algorithm to the SSMIS observations, the SSMIS brightness temperatures at the seven window channels from 12 to 18 (see Table 1) need to be remapped to the SSM/I channels. In this study, the mapping coefficients in Yan and Weng (2008) are used.

b. Utilization rate of UPP LAS data

As described in section 2, the accuracy of the surface emissivity affects the utilization rate of the data at the window and surface-sensitive channels. It is important to examine if the new emissivity algorithms increase the amount of SSMIS LAS data used in the NWP systems. This is performed by comparing SSMIS LAS data utilization rates between current and new emissivity calculation algorithms over snow and sea ice surfaces. In the current CRTM, the land emissivities at microwave frequencies can be derived from the MLEM while a constant of 0.9 is assumed for SSMIS observations over sea ice surfaces. This approach is used as a benchmark. The snow and sea ice emissivity regression algorithms in section 3c are used as a new approach in the assimilation of the SSMIS UPP data.

Figures 6a and 6b display the utilization rate of the UPP LAS data at three LAS channels from 50.3 to 53.6 GHz over snow and sea ice surfaces using current and new emissivity approaches, respectively. The results in Fig. 6 are computed based on the data covering a 2-month period from 1 August to 31 September 2008. As shown in Fig. 6, the utilization rate of the data at 50.3 GHz is below 20% when using the current emissivity approach, while it can exceed 40% when using the new emissivity algorithm. At 52.8 and 53.6 GHz, the improved data utilization rate is observed primarily over snow surfaces. This is because the brightness temperatures at 52.8 and 53.6 GHz are sensitive to surface emissivity, primarily at high-elevation surfaces that are covered by snow. It is thus concluded that the utilization rate of the UPP LAS at 50.3–52.8 GHz is increased by using the new emissivity approach.

Fig. 6.
Fig. 6.

Utilization rates of the UPP LAS data at three LAS channels from 50.3 to 53.6 GHz over (a) snow, where the utilization rates are calculated by using the MLEM and the newly derived (SNOWEM) emissivity approaches, and (b) sea ice, where the utilization rates are calculated by using a constant of 0.9 and the newly derived (ICEEM) emissivity approaches.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

c. Impact of UPP LAS data on GFS forecast skill

A key indicator for demonstrating the impact of the data on forecast skill is the anomaly correlation (AC) coefficient, which represents the correlation between the anomaly fields of the forecast and analysis from GFS (Lahoz 1999; Zapotocny et al. 2007). In the following, the AC coefficients over the Northern Hemisphere (NH) and Southern Hemisphere (SH) are calculated for a series of control and experimental runs in GFS. The heritage BC algorithm in Derber and Wu (1998) is used in the following control and experimental runs. The “coefficient(s)” in “AC coefficient(s)” is omitted hereinafter for clarity.

1) Impact of SSMIS LAS data and their comparison with AMSU-A

Since the frequencies of the SSMIS LAS channels are similar to many of the AMSU-A channels, it was expected that the SSMIS LAS data would produce similar effects as AMSU-A when they are used in NWP systems. Also, the SSMIS is a conically scanning instrument and has a constant viewing angle, so the bias should be independent of scan position. Using simulated data, Rosenkranz et al. (1997) showed the retrieval accuracy from a conically scanning instrument is better than that from cross-track scanning data. However, assimilation of real satellite data involves a number of other issues, such as bias characterizations and corrections, quality control criteria, etc., which are very different from the uses of simulated data. Therefore, it is necessary to compare the impacts on forecast skill from conically (e.g., SSMIS LAS) and cross-track (e.g., AMSU-A) scanning data in our NWP model. Here, the control run is the data assimilation without any satellite data. Two SSMIS experiments are conducted in which SSMIS UPP LAS data from 52.8 (channel 2) to 59.4 (channel 7) GHz are assimilated with and without the new BC. Two AMSU-A experiments are carried out, assimilating AMSU-A data from channel 4 (52.8 GHz) to channel 9 (57.3 GHz). In the two SSMIS experiments, the data cover one data period with the UPP noise reduction (e.g., from 1 to 14 August) and the other data period without the UPP noise reduction (e.g., from 14 August to 30 September). It is worth remembering that the new noise reduction and emissivity algorithms obtained in this study are not used in these SSMIS experiments since the experiments were performed before these new algorithms were finalized. The 50.3-GHz channel in both the AMSU-A and SSMIS datasets is not used in the experiment since the SSMIS UPP data at this channel have a large uncertainty.

Figures 7a and 7b show the AC for 500-mb geopotential height in the NH and SH, respectively, for a 2-month period from 1 August to 30 September 2008. Both SSMIS LAS and AMSU-A have positive impacts on global medium-range forecasts in both hemispheres. The impact of the satellite data, including the SSMIS LAS data, in the NH is smaller than that in the SH. More importantly, the impact of the SSMIS LAS data with the new BC is comparable to that of the AMSU-A data from NOAA-18 and MetOp-A over both the NH and SH, while the impact of the SSMIS data without the new BC displays a smaller impact on forecast skill than that of the AMSU-A over the SH. This demonstrates that the impact of the data can be reduced if the satellite data have geographically dependent biases. To further infer the relevance/importance of the anomaly correlation differences, the standard deviations of the anomaly correlations are also plotted in Fig. 7. Generally, the AC standard deviation increases with forecast length in all experiments. Among the five experiments, the AC standard deviations of the three experiments (NOAA-18 AMSU-A, MetOp-A AMSU-A, and F-16 UPP LAS with the new BC) are smaller than those of the control experiment and the SSMIS experiment without the new BC. This suggests that the new BC improves the impact of the UPP LAS data on assimilations and forecasts. The ACs at 1000 mb (figure not shown) yield the same conclusions.

Fig. 7.
Fig. 7.

Anomaly correlations at 500 mb over the (a) NH and (b) SH for one control and four experimental simulations (see solid lines vs left vertical axis), which cover the period from 1 Aug to 30 Sep 2008. Here, Cntrl Exp. is a satellite data denied experiment where only operational conventional in situ data are included. AMSUA Exp1. and AMSU-A Exp2. are two experimental runs with the addition of the AMSU-A data from NOAA-18 and MetOp-A, respectively, to the control dataset. SSMIS Exp1 and SSMIS Exp2 are two experimental runs that add the UPP LAS data without and with, respectively, the new BC. Note that the standard deviations of the anomaly correlations (see dash lines vs right vertical axis) are also plotted.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

2) Impact of SSMIS LAS data on GFS operational forecasts

It is also important to assess the impact of the SSMIS LAS database on the GFS operational dataset, which includes the conventional data, HIRS sounder radiances, AMSU-A/B and MHS radiances, Geostationary Operational Environmental Satellite (GOES) sounder radiances, SSM/I ocean surface wind speeds, Moderate Resolution Imaging Spectroradiometer (MODIS) winds, etc. For this purpose, a new control run and two experimental runs are designed. The control run uses all GFS operational data during the period from 1 August to 15 September 2008 (note that no SSMIS data are used). Two SSMIS experimental runs are conducted by adding the SSMIS UPP LAS data with and without the new BC, where the new snow and sea ice emissivity algorithms are used for emissivity calculations over snow and sea ice conditions. Note that the SSMIS LAS data only includes channels from 52.8 (channel 2) to 59.4 (channel 7) GHz. The channel at 50.3 GHz is not used in the experiment since the SSMIS UPP data at this channel have a large uncertainty.

Numerical results show that in the NH, the impact of SSMIS LAS data on forecast skill at both 500 and 1000 mb is neutral (figures not shown). In the SH, the SSMIS LAS data with the new BC algorithm produce a slightly positive impact on forecast skill at both 500 and 1000 mb (Figs. 8a and 8b). The limited impact of the SSMIS LAS data here is due primarily to the use of many other satellite data sources in the control run. This is also demonstrated by the fact that the AC standard deviations in the SSMIS experiments are similar to those in the control experiment. Note that the impact of the snow and sea ice emissivity regression algorithms is not independently assessed in this study. A major reason is that the regression algorithms for SSMIS follow the same methodology as we did for other microwave sensors such as AMSU-A/B and MHS. The impact of the snow and sea ice emissivity regression algorithms in AMSU and MHS assimilations displays a positive impact on forecast skill (e.g., Yan et al. 2008a).

Fig. 8.
Fig. 8.

Anomaly correlations over the SH at (a) 500 and (b) 1000 mb for one control and two experimental runs (see solid lines vs left vertical axis), which cover the period from 1 Aug to 15 Sep 2008. Here, Cntrl Exp uses all operational data, including conventional in situ data, and all microwave sensors except for SSMIS. The two experiments further add the SSMIS UPP LAS data from channels 2–7 to the Cntrl Exp dataset, where one is for the original UPP LAS data (SSMIS Exp1.) and the other one is for the UPP LAS data with the new BC (SSMIS Exp2). The new snow and sea ice emissivity algorithms in section 3 are applied to the both experimental runs for the assimilation of SSMIS LAS data. Note that the standard deviations of the anomaly correlations (see the dashed lines vs the right vertical axis) are also plotted.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

The above impacts are assessed using the UPP data for which the noise reduction algorithm of Bell et al. (2008) was turned on for the data before 14 August 2008 and turned off for the data after 14 August 2008. To assess the impact of the noise reduction algorithm developed in this study, the data are selected with a different period from 1 September to 30 October 2009 when the noise reduction algorithm of Bell et al. (2008) was totally turned off. The assimilation experiments are designed as follows. The control run uses the conventional data plus all the sensor data except for data from microwave sounding instruments. Two experimental runs adding the SSMIS LAS data at 52.8–59.4 GHz are carried out with and without the new noise reduction. Similarly, the neutral impact of the data on forecast skill is observed over the NH at both 500 and 1000 mb (figures not shown). Over the SH, the slightly positive impact is observed from the SSMIS experiment with the new noise reduction algorithm incorporated (Figs. 9a and 9b). Note that the impact from the SSMIS LAS data without the new bias-correction algorithm is negative. This further indicates the significance of the new bias-correction algorithm. In addition, the AC standard deviations at the SSMIS experiments are similar to those of the control experiment, which implies that the UPP LAS data impact on forecast skill is small when many satellite data sources are used in the experiments.

Fig. 9.
Fig. 9.

Anomaly correlations over the SH at (a) 500 and (b) 1000 mb for one control and two experimental runs. Here, Cntrl Exp. is the control run having operational conventional in situ data and all satellite sensors except for microwave sounding data; two experimental runs add the SSMIS LAS data from 52.8 to 59.4 GHz to the control run dataset but one uses the original UPP data (no noise reduction processing) (SSMIS Exp1.) and the other one uses the noise reduction processing (SSMIS Exp2.). Note that the standard deviations of anomaly correlations (see dashed lines vs right vertical axis) are also plotted.

Citation: Weather and Forecasting 27, 3; 10.1175/WAF-D-11-00062.1

5. Summary and conclusions

Several algorithms for the assimilation of the SSMIS LAS data into the GFS are developed and tested. A bias-correction algorithm is used for removing the geographically dependent biases in the UPP LAS data. The noise reduction algorithm is used to improve the quality of the SSMIS data observations. The regression algorithms for snow and sea ice emissivity are derived using the emissivity indices and an LUT consisting of 16 snow and 13 sea ice type emissivity spectra. It is shown that the new bias correction improves the quality of the UPP LAS data. The noise reduction algorithm decreases the random component of the UPP LAS data by approximately 50%. The snow and sea ice emissivity algorithms reduce the uncertainty in the surface emissivity calculations. As a result, much of the SSMIS data at surface-sensitive channels can be assimilated.

The new algorithms further improve the impacts of the F-16 UPP LAS data on forecast skills according to a series of experimental runs based on different types of control run datasets. When the LAS data are added to the control run with satellite-denied data, the positive impact of the data with the new BC on forecast skill over the SH is larger than that without this new BC. The impact of SSMIS data becomes comparable with that of the AMSU-A data from either NOAA-18 or MetOp-A over both the NH and SH. When the UPP LAS data are added to the control run with all GFS operational datasets except for microwave sensor data, the UPP data with the new noise reduction algorithm produces neutral (in the NH) or slightly positive (in the SH) impacts on forecast skill at both 500 and 1000 mb. When the control run uses all the GFS operational data, the F-16 UPP LAS data, with the new BC and snow and sea ice emissivity algorithms applied, result in a neutral impact in the NH or weakly positive impact in the SH. In general, the use of SSMIS UPP data from the F-16 satellite has neutral or positive impacts on forecast skill.

This study has developed the new BC, noise reduction, and snow and sea ice emissivity regression algorithms using F-16 data; the same methodology is applicable for other SSMIS observations. In particular, the radiance noise reduction and snow and sea ice emissivity algorithms can be easily used for the assimilation of F-17 (UPP) and F-18 SSMIS data. This study represents a first step toward assimilating all SSMIS data in GFS. The methodology can also be applied to other instruments that exhibit similar problems related to calibration anomaly, geographically dependent biases, and large noise.

Acknowledgments

This research is jointly supported by Chinese Ministry of Science and Technology Project 2010CB951600 and the Joint Center for Satellite Data Assimilation Program. The authors thank Drs. Russ Treadon and John Derber for their help in running the NCEP GFS and GSI experiments. Thanks also go to Gregory S. Krasowski for his help in preparing for the UPP data. The views expressed in this publication are those of the authors and do not necessarily represent those of NOAA.

APPENDIX

Derivation of Discriminator Indices for SSMIS LAS Data

For the SSMIS LAS data, three DIs at 19.35, 37, and 91.655 GHz are defined in the polarization-weighted emissivity:
ea1
where the subscript k is an index from 1 to 3 corresponding to frequencies at 19.35, 37, and 91.655 GHz, respectively; θzen is the satellite zenith angle of SSMIS (53.1°); and ɛυ and ɛh are the simulated emissivities at the vertical and horizontal polarizations at the above three frequencies, respectively, which are computed using one of the following fitting equations depending on frequency. For the channels from 12 [19.35 GHz on a horizontal polarization (H-Pol) to 16 (37 GHz at vertical polarization or V-Pol), the emissivity is simulated using brightness temperatures at the following five channels:
ea2
where the subscript ich is the channel index from 12 to 16 and T19H, T19V, T22V, T37H, and T37V are brightness temperatures at the corresponding channels. For the channels from 17 (91.655 GHz at V-Pol) to 18 (91.655 GHz at H-Pol), the emissivity is simulated using brightness temperatures at the following seven channels from 12 to 18 and also channel 8 (150 GHz at H-Pol):
ea3
where the subscript jch is the channel index from 17 to 18 and T91V, T91H, and T150H are the brightness temperatures at the 17th, 18th, and 8th channels, respectively (see Table 1). The coefficients from a0 to a9 for snow and sea ice emissivities in Eqs. (A2) and (A3) are given in Tables A1 and A2, and are derived using the training dataset of emissivity and SSMIS brightness temperatures under clear-sky conditions through the CRTM and NCEP GDAS atmospheric profiles and surface information, which is similar to the methodology used in Yan and Weng (2011).
Table A1.

Fitting coefficients used in Eqs. (A2) and (A3) for microwave snow emissivity calculations.

Table A1.
Table A2.

Fitting coefficients used in Eqs. (A2) and (A3) for microwave sea ice emissivity calculations.

Table A2.

REFERENCES

  • Andreadis, K., Liang D. , Tsang L. , Lettenmaier D. P. , and Josberger E. G. , 2008: Characterization of errors in a coupled snow hydrology–microwave emission model. J. Hydrometeor., 9, 149164.

    • Search Google Scholar
    • Export Citation
  • Bell, W., English S. , and Swadley S. , 2005: SSMIS calibration issues. Proc. First SSMIS Working Group Meeting, Monterey, CA, Naval Research Laboratory.

  • Bell, W., and Coauthors, 2008: The assimilation of SSMIS radiances in numerical weather prediction models. IEEE Trans. Geosci. Remote Sens., 4, 884900.

    • Search Google Scholar
    • Export Citation
  • Bell, W., Michele S. , Bauer P. , Mcnally T. , English S. , Atkinson N. , Hilton F. , and Charlton J. , 2010: The radiometric sensitivity requirements for satellite microwave temperature sounding instruments for numerical weather prediction. J. Atmos. Oceanic Technol., 27, 443456.

    • Search Google Scholar
    • Export Citation
  • Derber, J. C., and Wu W. S. , 1998: The use of TOVS cloud-cleared radiances in the NCEP SSI analysis system. Mon. Wea. Rev., 126, 22872299.

    • Search Google Scholar
    • Export Citation
  • Han, Y., van Delst P. , Liu Q. , Weng F. , Yan B. , Treadon R. , and Derber J. , 2006: Community Radiative Transfer Model (CRTM)—Version 1. NOAA Tech. Rep. 122, 33 pp.

  • Han, Y., Van Delst P. , Weng F. , Liu Q. , Groff D. , Yan B. , Chen Y. , and Vogel R. L. , 2009: An overview on the JCSDA Community Radiative Transfer Model (CRTM) version 2. Preprints, Sixth Annual Symp. on Future National Operational Environmental Satellite Systems-NPOESS and GOES-R, Atlanta, GA, Amer. Meteor. Soc., 7.1. [Available online at http://ams.confex.com/ams/pdfpapers/163619.pdf.]

  • Hewison, T. J., and English S. J. , 1999: Airborne retrievals of snow and ice surface emissivity at millimeter wavelengths. IEEE Trans. Geosci. Remote Sens., 37, 18711879.

    • Search Google Scholar
    • Export Citation
  • Ide, K., Courtier P. , Ghil M. , and Lorenc A. C. , 1997: Unified notation for data assimilation operational, sequential and variational. J. Meteor. Soc. Japan, 75, 181189.

    • Search Google Scholar
    • Export Citation
  • Kazumori, M., 2007: Impact study of DMSP F-16 SSMIS radiances in NCEP global data assimilation system. Fifth Workshop on Satellite Data Assimilation, College Park, MD, Joint Center for Satellite Data Assimilation. [Available online at http://www.jcsda.noaa.gov/documents/meetings/wkshp7/Session2_Instruments/JCSDA.Kazumori.may2007.pdf.]

  • Kunkee, D. B., Poe G. A. , Boucher D. J. , Swadley S. , Hong Y. , Wessel J. , and Uliana E. , 2008: Design and evaluation of the first Special Sensor Microwave Imager/Sounder (SSMIS). IEEE Trans. Geosci. Remote Sens., 46, 863883.

    • Search Google Scholar
    • Export Citation
  • Lahoz, W. A., 1999: Predictive skill of the UKMO Unified Model in the lower stratosphere. Quart. J. Roy. Meteor. Soc., 125, 22052238.

    • Search Google Scholar
    • Export Citation
  • Mätzler, C., 1994: Passive microwave signatures of landscapes in winter. Meteor. Atmos. Phys., 54, 241260.

  • Parrish, D. F., and Derber J. C. , 1992: The National Meteorological Center’s spectral statistical interpolation analysis system. Mon. Wea. Rev., 120, 17471763.

    • Search Google Scholar
    • Export Citation
  • Poe, G., Germain K. , Bobak J. , Swadley S. , Wessel J. , Thomas B. , Wang J. , and Burns B. , 2001: DMSP calibration/validation plan for the Special Sensor Microwave Imager Sounder (SSMIS). Naval Research Laboratory Rep., 32 pp.

  • Rosenkranz, P. W., Hutchison K. D. , Hardy K. R. , and Davis M. S. , 1997: An assessment of the impact of satellite microwave sounder incidence angle and scan geometry on the accuracy of atmospheric temperature profile retrievals. J. Atmos. Oceanic Technol., 14, 488494.

    • Search Google Scholar
    • Export Citation
  • Swadley, S., Poe G. , Uliana A. , and Kunkee D. , 2005: SSMIS Cal/Val calibration anomaly analysis. NOAA–JCSDA Seminar, Washington, DC, JCSDA. [Available online at http://www.jcsda.noaa.gov/documents/seminardocs/ssmis_calibration_anomaly_jcsda_july_27_2005.pdf.]

  • Swadley, S., Poe G. , Kunkee D. , Bell W. , Brown S. , Prata I. , Long E. , and Boucher D. , 2008: SSMIS calibration anomalies: Observed F-16 and F-17 anomalies, detailed analysis of the root causes, and the path forward. Proc. 16th Int. TOVS Study Conf., Angra dos Reis, Brazil, Int. TOVS Working Group, 1.8.

  • Swadley, S., Poe G. , Ruston B. , Bell W. , Kunkee D. , and Boucher D. , 2010: SSMIS radiance assimilation, calibration anomaly mitigation and early results from F-18. 11th Specialist Meeting on Microwave Radiometry and Remote Sensing of the Environment, Washington, DC, IEEE/Geoscience and Remote Sensing Society.

  • Weng, F., and Grody N. C. , 1994: Retrieval of cloud liquid water using the Special Sensor Microwave Imager (SSM/I). J. Geophys. Res., 99, 25 53525 551.

    • Search Google Scholar
    • Export Citation
  • Weng, F., Yan B. , and Grody N. , 2001: A microwave land emissivity model. J. Geophys. Res., 106, 20 11520 123.

  • Weng, F., Han Y. , van Delst P. , Liu Q. , and Yan B. , 2005a: JCSDA Community Radiative Transfer Model (CRTM). Tech. Proc. 14th Int. ATOVS Study Conf., Beijing, China, Int. TOVS Working Group. [Available online at http://cimss.ssec.wisc.edu/itwg/itsc/itsc14/proceedings/6_8_Weng.pdf.]

  • Weng, F., Yan B. , and Sun N. , 2005b: Correction of SSMIS radiance anomalies. Proc. First SSMIS Working Group Meeting, Monterey, CA, Naval Research Laboratory.

  • Wiesmann, A., and Mätzler C. , 1999: Microwave emission model of layered snowpacks. Remote Sens. Environ., 70, 307316.

  • Wójcik, R., Andreadis K. , Tedesco M. , Wood E. , Troy T. , and Lettenmeier D. , 2008: Multimodel estimation of snow microwave emission during CLPX 2003 using operational parameterization of microphysical snow characteristics. J. Hydrometeor., 9, 14911505.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2008: Intercalibration between Special Sensor Microwave Imager and Sounder (SSMIS) and Special Sensor Microwave Imager (SSM/I). IEEE Trans. Geosci. Remote Sens., 46, 984995.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2009: Assessments of F16 Special Sensor Microwave Imager and Sounder antenna temperatures at lower atmospheric sounding channels. Adv. Meteor., 2009, 420985, doi:10.1155/2009/420985.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2011: Effects of microwave desert surface emissivity on AMSU-A data assimilation. IEEE Trans. Geosci. Remote Sens., 49, 12631276.

    • Search Google Scholar
    • Export Citation
  • Yan, B., Weng F. , and Okamoto K. , 2004: A microwave snow emissivity model. Eighth Specialist Meeting on Microwave Radiometry and Remote Sensing Applications, Rome, Italy, IEEE/Geoscience and Remote Sensing Society.

  • Yan, B., Weng F. , and Derber J. , 2008a: Microwave emissivity model update. Sixth Workshop on Satellite Data Assimilation, Linthicum, MD, NOAA/JCSDA. [Available online at http://www.jcsda.noaa.gov/documents/meetings/wkshp2008/1/01_02_yan_Snowem_Impact.pdf.]

  • Yan, B., Weng F. , and Meng H. , 2008b: Retrieval of snow surface microwave emissivity from Advanced Microwave Sounding Unit (AMSU). J. Geophys. Res., 113, D19206, doi:10.1029/2007JD009559.

    • Search Google Scholar
    • Export Citation
  • Yan, B., Weng F. , and Derber J. , 2010: Assimilation of satellite microwave water vapor sounding channel data in NCEP Global Forecast System (GFS). 17th Int. TOVS Study Conf., Monterey, CA, Int. TOVS Working Group, 7.11.

  • Zapotocny, T. H., Jung J. A. , Marchall J. F. , and Treadon R. E. , 2007: A two-season impact study of satellite and in situ data in the NCEP Global Data Assimilation System. Wea. Forecasting, 22, 887909.

    • Search Google Scholar
    • Export Citation
Save
  • Andreadis, K., Liang D. , Tsang L. , Lettenmaier D. P. , and Josberger E. G. , 2008: Characterization of errors in a coupled snow hydrology–microwave emission model. J. Hydrometeor., 9, 149164.

    • Search Google Scholar
    • Export Citation
  • Bell, W., English S. , and Swadley S. , 2005: SSMIS calibration issues. Proc. First SSMIS Working Group Meeting, Monterey, CA, Naval Research Laboratory.

  • Bell, W., and Coauthors, 2008: The assimilation of SSMIS radiances in numerical weather prediction models. IEEE Trans. Geosci. Remote Sens., 4, 884900.

    • Search Google Scholar
    • Export Citation
  • Bell, W., Michele S. , Bauer P. , Mcnally T. , English S. , Atkinson N. , Hilton F. , and Charlton J. , 2010: The radiometric sensitivity requirements for satellite microwave temperature sounding instruments for numerical weather prediction. J. Atmos. Oceanic Technol., 27, 443456.

    • Search Google Scholar
    • Export Citation
  • Derber, J. C., and Wu W. S. , 1998: The use of TOVS cloud-cleared radiances in the NCEP SSI analysis system. Mon. Wea. Rev., 126, 22872299.

    • Search Google Scholar
    • Export Citation
  • Han, Y., van Delst P. , Liu Q. , Weng F. , Yan B. , Treadon R. , and Derber J. , 2006: Community Radiative Transfer Model (CRTM)—Version 1. NOAA Tech. Rep. 122, 33 pp.

  • Han, Y., Van Delst P. , Weng F. , Liu Q. , Groff D. , Yan B. , Chen Y. , and Vogel R. L. , 2009: An overview on the JCSDA Community Radiative Transfer Model (CRTM) version 2. Preprints, Sixth Annual Symp. on Future National Operational Environmental Satellite Systems-NPOESS and GOES-R, Atlanta, GA, Amer. Meteor. Soc., 7.1. [Available online at http://ams.confex.com/ams/pdfpapers/163619.pdf.]

  • Hewison, T. J., and English S. J. , 1999: Airborne retrievals of snow and ice surface emissivity at millimeter wavelengths. IEEE Trans. Geosci. Remote Sens., 37, 18711879.

    • Search Google Scholar
    • Export Citation
  • Ide, K., Courtier P. , Ghil M. , and Lorenc A. C. , 1997: Unified notation for data assimilation operational, sequential and variational. J. Meteor. Soc. Japan, 75, 181189.

    • Search Google Scholar
    • Export Citation
  • Kazumori, M., 2007: Impact study of DMSP F-16 SSMIS radiances in NCEP global data assimilation system. Fifth Workshop on Satellite Data Assimilation, College Park, MD, Joint Center for Satellite Data Assimilation. [Available online at http://www.jcsda.noaa.gov/documents/meetings/wkshp7/Session2_Instruments/JCSDA.Kazumori.may2007.pdf.]

  • Kunkee, D. B., Poe G. A. , Boucher D. J. , Swadley S. , Hong Y. , Wessel J. , and Uliana E. , 2008: Design and evaluation of the first Special Sensor Microwave Imager/Sounder (SSMIS). IEEE Trans. Geosci. Remote Sens., 46, 863883.

    • Search Google Scholar
    • Export Citation
  • Lahoz, W. A., 1999: Predictive skill of the UKMO Unified Model in the lower stratosphere. Quart. J. Roy. Meteor. Soc., 125, 22052238.

    • Search Google Scholar
    • Export Citation
  • Mätzler, C., 1994: Passive microwave signatures of landscapes in winter. Meteor. Atmos. Phys., 54, 241260.

  • Parrish, D. F., and Derber J. C. , 1992: The National Meteorological Center’s spectral statistical interpolation analysis system. Mon. Wea. Rev., 120, 17471763.

    • Search Google Scholar
    • Export Citation
  • Poe, G., Germain K. , Bobak J. , Swadley S. , Wessel J. , Thomas B. , Wang J. , and Burns B. , 2001: DMSP calibration/validation plan for the Special Sensor Microwave Imager Sounder (SSMIS). Naval Research Laboratory Rep., 32 pp.

  • Rosenkranz, P. W., Hutchison K. D. , Hardy K. R. , and Davis M. S. , 1997: An assessment of the impact of satellite microwave sounder incidence angle and scan geometry on the accuracy of atmospheric temperature profile retrievals. J. Atmos. Oceanic Technol., 14, 488494.

    • Search Google Scholar
    • Export Citation
  • Swadley, S., Poe G. , Uliana A. , and Kunkee D. , 2005: SSMIS Cal/Val calibration anomaly analysis. NOAA–JCSDA Seminar, Washington, DC, JCSDA. [Available online at http://www.jcsda.noaa.gov/documents/seminardocs/ssmis_calibration_anomaly_jcsda_july_27_2005.pdf.]

  • Swadley, S., Poe G. , Kunkee D. , Bell W. , Brown S. , Prata I. , Long E. , and Boucher D. , 2008: SSMIS calibration anomalies: Observed F-16 and F-17 anomalies, detailed analysis of the root causes, and the path forward. Proc. 16th Int. TOVS Study Conf., Angra dos Reis, Brazil, Int. TOVS Working Group, 1.8.

  • Swadley, S., Poe G. , Ruston B. , Bell W. , Kunkee D. , and Boucher D. , 2010: SSMIS radiance assimilation, calibration anomaly mitigation and early results from F-18. 11th Specialist Meeting on Microwave Radiometry and Remote Sensing of the Environment, Washington, DC, IEEE/Geoscience and Remote Sensing Society.

  • Weng, F., and Grody N. C. , 1994: Retrieval of cloud liquid water using the Special Sensor Microwave Imager (SSM/I). J. Geophys. Res., 99, 25 53525 551.

    • Search Google Scholar
    • Export Citation
  • Weng, F., Yan B. , and Grody N. , 2001: A microwave land emissivity model. J. Geophys. Res., 106, 20 11520 123.

  • Weng, F., Han Y. , van Delst P. , Liu Q. , and Yan B. , 2005a: JCSDA Community Radiative Transfer Model (CRTM). Tech. Proc. 14th Int. ATOVS Study Conf., Beijing, China, Int. TOVS Working Group. [Available online at http://cimss.ssec.wisc.edu/itwg/itsc/itsc14/proceedings/6_8_Weng.pdf.]

  • Weng, F., Yan B. , and Sun N. , 2005b: Correction of SSMIS radiance anomalies. Proc. First SSMIS Working Group Meeting, Monterey, CA, Naval Research Laboratory.

  • Wiesmann, A., and Mätzler C. , 1999: Microwave emission model of layered snowpacks. Remote Sens. Environ., 70, 307316.

  • Wójcik, R., Andreadis K. , Tedesco M. , Wood E. , Troy T. , and Lettenmeier D. , 2008: Multimodel estimation of snow microwave emission during CLPX 2003 using operational parameterization of microphysical snow characteristics. J. Hydrometeor., 9, 14911505.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2008: Intercalibration between Special Sensor Microwave Imager and Sounder (SSMIS) and Special Sensor Microwave Imager (SSM/I). IEEE Trans. Geosci. Remote Sens., 46, 984995.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2009: Assessments of F16 Special Sensor Microwave Imager and Sounder antenna temperatures at lower atmospheric sounding channels. Adv. Meteor., 2009, 420985, doi:10.1155/2009/420985.

    • Search Google Scholar
    • Export Citation
  • Yan, B., and Weng F. , 2011: Effects of microwave desert surface emissivity on AMSU-A data assimilation. IEEE Trans. Geosci. Remote Sens., 49, 12631276.

    • Search Google Scholar
    • Export Citation
  • Yan, B., Weng F. , and Okamoto K. , 2004: A microwave snow emissivity model. Eighth Specialist Meeting on Microwave Radiometry and Remote Sensing Applications, Rome, Italy, IEEE/Geoscience and Remote Sensing Society.

  • Yan, B., Weng F. , and Derber J. , 2008a: Microwave emissivity model update. Sixth Workshop on Satellite Data Assimilation, Linthicum, MD, NOAA/JCSDA. [Available online at http://www.jcsda.noaa.gov/documents/meetings/wkshp2008/1/01_02_yan_Snowem_Impact.pdf.]

  • Yan, B., Weng F. , and Meng H. , 2008b: Retrieval of snow surface microwave emissivity from Advanced Microwave Sounding Unit (AMSU). J. Geophys. Res., 113, D19206, doi:10.1029/2007JD009559.

    • Search Google Scholar
    • Export Citation
  • Yan, B., Weng F. , and Derber J. , 2010: Assimilation of satellite microwave water vapor sounding channel data in NCEP Global Forecast System (GFS). 17th Int. TOVS Study Conf., Monterey, CA, Int. TOVS Working Group, 7.11.

  • Zapotocny, T. H., Jung J. A. , Marchall J. F. , and Treadon R. E. , 2007: A two-season impact study of satellite and in situ data in the NCEP Global Data Assimilation System. Wea. Forecasting, 22, 887909.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Distribution of brightness temperature differences on 5 Aug 2008 for F-16 UPP at (a) 54.4 and (b) 55.5 GHz, and for MetOp-A AMSU-A at (c) 54.4 and (d) 55.5 GHz, which are calculated using the data passing the current GFS quality control test after the GFS bias correction (Derber and Wu 1998) is applied. Here, std. represents the standard deviation of the brightness temperature differences.

  • Fig. 2.

    F-16 UPP brightness temperature difference maps on 28 Aug 2008 at (a) 54.4 and (b) 55.5 GHz, where no noise reduction algorithm is applied. As in Fig. 1, std. represents the standard deviation of the brightness temperature differences.

  • Fig. 3.

    Time series of the daily or weekly averages of the longitudinal-mean brightness temperature (TB) differences at 54.4 and 55.5 GHz. (a) Daily average of longitudinal-mean TB differences at 54.4 GHz for 7 days from 5 to 11 Aug 2011. (b) As in (a), but for 55.5 GHz. (c) Weekly average of longitudinal-mean TB differences at 54.4 GHz for 8 weeks from 5 Aug to 30 Sep, which include the results with (dark gray) and without (light gray) the newly developed BC algorithm. (d) As in (c), but at 55.5 GHz. Here, all TB differences have been plotted by adding the integer number of each overlapping dashed line for clarity.

  • Fig. 4.

    As in Fig. 2, but for application of the noise reduction algorithm.

  • Fig. 5.

    (a) Microwave snow emissivity as a function of frequency across the range between 4.9 and 150 GHz, where the five new spectra are labeled with RS, which implies a distinct emissivity spectrum but cannot be directly associated with a physical snow type while the other snow types followed the classification of snow types in Mätzler (1994). (b) Microwave sea ice emissivity as a function of frequency across the range between 6.7 and 157 GHz, where the six new spectra are labeled with RS and have similar meanings as in (a) but for sea ice while the other sea ice types followed the classification of Hewison and English (1999).

  • Fig. 6.

    Utilization rates of the UPP LAS data at three LAS channels from 50.3 to 53.6 GHz over (a) snow, where the utilization rates are calculated by using the MLEM and the newly derived (SNOWEM) emissivity approaches, and (b) sea ice, where the utilization rates are calculated by using a constant of 0.9 and the newly derived (ICEEM) emissivity approaches.

  • Fig. 7.

    Anomaly correlations at 500 mb over the (a) NH and (b) SH for one control and four experimental simulations (see solid lines vs left vertical axis), which cover the period from 1 Aug to 30 Sep 2008. Here, Cntrl Exp. is a satellite data denied experiment where only operational conventional in situ data are included. AMSUA Exp1. and AMSU-A Exp2. are two experimental runs with the addition of the AMSU-A data from NOAA-18 and MetOp-A, respectively, to the control dataset. SSMIS Exp1 and SSMIS Exp2 are two experimental runs that add the UPP LAS data without and with, respectively, the new BC. Note that the standard deviations of the anomaly correlations (see dash lines vs right vertical axis) are also plotted.

  • Fig. 8.

    Anomaly correlations over the SH at (a) 500 and (b) 1000 mb for one control and two experimental runs (see solid lines vs left vertical axis), which cover the period from 1 Aug to 15 Sep 2008. Here, Cntrl Exp uses all operational data, including conventional in situ data, and all microwave sensors except for SSMIS. The two experiments further add the SSMIS UPP LAS data from channels 2–7 to the Cntrl Exp dataset, where one is for the original UPP LAS data (SSMIS Exp1.) and the other one is for the UPP LAS data with the new BC (SSMIS Exp2). The new snow and sea ice emissivity algorithms in section 3 are applied to the both experimental runs for the assimilation of SSMIS LAS data. Note that the standard deviations of the anomaly correlations (see the dashed lines vs the right vertical axis) are also plotted.

  • Fig. 9.

    Anomaly correlations over the SH at (a) 500 and (b) 1000 mb for one control and two experimental runs. Here, Cntrl Exp. is the control run having operational conventional in situ data and all satellite sensors except for microwave sounding data; two experimental runs add the SSMIS LAS data from 52.8 to 59.4 GHz to the control run dataset but one uses the original UPP data (no noise reduction processing) (SSMIS Exp1.) and the other one uses the noise reduction processing (SSMIS Exp2.). Note that the standard deviations of anomaly correlations (see dashed lines vs right vertical axis) are also plotted.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 161 64 13
PDF Downloads 89 21 7