Observing System Simulation Experiments Investigating Atmospheric Motion Vectors and Radiances from a Constellation of 4–5-μm Infrared Sounders
1. Introduction
The understanding of the Earth system is linked to the ability to observe its processes. Conventional measurements are the easiest to link to the physical equations that govern the Earth system but are irregular in space and time. To compensate for these spatiotemporal gaps, remotely sensed observations are used to fill the void to achieve global coverage. To characterize the atmosphere’s multidimensional mass state, a data assimilation procedure is used to translate the radiation measured from space to information of the emitting molecules (e.g., Derber and Wu 1998; Susskind et al. 1998; McNally et al. 2000). The improved use of these radiance observations and advances in the observing system have resulted in equally skillful forecasts in the Northern Hemisphere, which is well-observed by conventional observations, and the Southern Hemisphere, which is sparsely observed by conventional observations (Bauer et al. 2015; Diniz and Todling 2020).
There remains a fundamental gap in the observing system for global winds. Well-established methods exist to infer the winds from spaceborne radiance measurements. First, a feature must be discernable in radiance space—i.e., cloud edges or contrasts in water vapor fields. Second, the features must be able to be tracked by sequential measurements. If these two requirements are met, the displacement of the feature and time-spacing of the measurements can then be used to infer a wind measurement, known as an atmospheric motion vector (AMV; Nieman et al. 1997; Velden et al. 1997; Key et al. 2003). This method has been widely applied to imagers on geostationary platforms. It is noted that other measurements also are beginning to address these wind data gaps. For example, ESA has launched the Aeolus mission. Via its Doppler wind lidar, Aeolus provides vertically profiling wind measurements from space but lacks spatial sampling available of AMVs.
AMVs make up a significant component of the global observing system of wind (McCarty et al. 2016), but they have fundamental limitations compared to the mass observations. Spatial and temporal resolution requirements limit these methods to the visible and infrared imagery. Also, there is no profiling of the wind below the cloud top. The climatological distribution of cloud top height (Wylie and Menzel 1999; Wylie et al. 2007) limits the bulk of the cloud-derived AMVs to the upper and lower troposphere. Water vapor AMVs can fill gaps between cloud tracers spatially, but they are limited by the spectral information content of the imagers—three spectral bands on the present-day geostationary imagers and no water vapor bands on the current operational polar orbiting imagers. By these constraints, the cloud- and water vapor–derived AMVs complement each other spatially, but their vertical distributions are largely noncomplementary, resulting in a data void in the midtroposphere (Santek et al. 2019b). Also, the error characteristics of the clear-sky water vapor AMVs are complicated because of the difficulty in assigning a height to winds derived from radiance measurements with deep kernels (Velden and Bedka 2009).
In contrast to imagers, infrared sounders have prioritized spectral information over spatial resolution. The Cross-Track Infrared Sounder (CrIS) on Suomi National Polar-Orbiting Partnership (SNPP) and on NOAA-20 measures 2211 spectral bands with a nominal footprint size of 14 km. Conversely, the Visible Infrared Imaging Radiometer Suite (VIIRS) on the same satellites measures 22 bands, six of which spectrally overlap with CrIS, with a nominal footprint of 750 m in the thermal infrared.
This study focused on a proposed observing strategy to better characterize the global, three-dimensional winds by extending the existing AMV methods to advanced instrumentation. The proposed observing strategy is for multiple infrared instruments with sounder-like spectral resolution and imager-like spatial resolution to fly in constellation, measuring the same ground-relative points at close time intervals. This strategy provides the temporal information that has previously only been obtainable from geostationary orbit. Since the spectral coverage and resolution now provide more information content in the vertical, the tracking strategy emphasizes resolution of shallow water vapor features in the midtroposphere, which can substantially increase the number of wind observations in this region. This strategy is a global extension of the methods demonstrated in Santek et al. (2019a), and this study is performed in the construct of the Midwave Infrared Sounding of Temperature and humidity in a Constellation for Winds (MISTiC Winds; Maschhoff et al. 2019) strategy proposed by BAE Systems.
Although not unique to the MISTiC concept, the AMV methods investigated in this effort are based on the retrieval of AMVs in retrieval space instead of radiance space. While this approach has been investigated in previous studies, three key issues have limited its effectiveness for providing global, three-dimensional distributions of horizontal wind:
The spectral information needed to effectively invert and resolve the vertical water vapor on pressure surfaces has previously only been measured from sun synchronous, polar-crossing, low-Earth orbits (LEO).
The temporal information necessary for AMV inversion is limited to overlap near the poles, with a refresh rate akin to the orbital period, (Jedlovec and Atkinson 1998; Santek et al. 2019a).
The spatial resolution has largely limited the number of resolvable and trackable features (Jedlovec and Atkinson 1998; Santek et al. 2019a; Posselt et al. 2019).
While these three limitations are important, previous studies have illustrated the effectiveness of determining AMVs in polar regions in retrieval space using overlaps of consecutive orbits of the Atmospheric Infrared Sounder (AIRS) measurements (Santek et al. 2019a).
To characterize the potential impact of this observing strategy, a suite of observing system simulation experiments (OSSEs; Atlas et al. 1985; Arnold and Dey 1986; Errico et al. 2013; Hoffman and Atlas 2016) was performed. The experiments were constructed in a manner to aid in the quantification and understanding of the benefit this strategy can provide to numerical weather prediction (NWP). The OSSEs were designed to provide not only an assessment of impact of this observing strategy, but also to provide an understanding of the elements of its uncertainty. Specifically, the following components of the observing strategy are considered. First, both the mass information directly measured by the sounder radiances and the wind information derived from the constellation observing strategy were considered together to fully capture the information content of the constellation approach. Second, “perfect” simulated observations were compared to “imperfect” observations—constructed by simulating the observations and observation errors—to provide an understanding of idealized and realistic information content within the construct of the experimentation systems as they presently exist.
The structure of the document is as follows. First, the MISTiC Winds observing concept is presented. Second, the baseline Global Modeling and Assimilation Office (GMAO) OSSE system is introduced, and this system serves as the control for experimentation. Third, the MISTiC radiances and AMVs are introduced, including details on the method used to simulate these proposed data. Fourth, the experiment design is presented. Fifth, results are presented in three steps—analysis validation, forecast verification, and the assessment via the forecast sensitivity to observation impact (FSOI) metric. Sixth, the results are summarized, and conclusions are presented.
2. MISTiC winds concept
As technology advances, the traditional signal-to-noise trade-offs continue to exist in terms of resolution—spatial, spectral, temporal, etc. Furthermore, many advances in technology allow for reductions in both the size and cost of proposed instrumentation. These tradeoffs have resulted in the advent of small satellites, or SmallSats. Present SmallSat concepts exploit their inexpensive, small design to base missions around constellations (e.g., Ruf et al. 2013; Masters et al. 2020). A primary advantage of this approach is that their orbits can be configured to overcome traditional spatiotemporal sampling limitations of remote sensing from polar orbits.
Building on these advancements, BAE Systems developed the MISTiC Winds concept (Maschhoff et al. 2019). Core to the concept is the MISTiC instrument, which has been developed as part of the NASA Earth Science Technology Office (ESTO) Instrument Incubator Program (IIP).
The MISTiC instrument investigated in this study is designed with the goal of measuring AMVs via the constellation approach while maintaining a configuration that that meets the size and weight standards of a SmallSat. It measured 590 bands over a spectral range of 1750 to 2450 cm−1—or 5.7 to 4.1 μm—with a spectral resolution comparable to apodized CrIS radiances. This spectral range covers the shortwave side of the 6.7-μm water vapor and both sides of the 4.3 CO2 absorption bands. Specifically of importance to the determination of AMVs on retrieved surfaces, Maschhoff et al. (2019) illustrated the error dependence of temperature and water vapor retrievals as a function of spectral coverage of this water vapor band.
Per specifications provided by BAE Systems for this study, radiances are measured at a spatial resolution of 3 km at nadir, and the simulated constellation flies in sun-synchronous orbits at 705 km. The local time of the ascending node (LTAN) is treated as variable in this study as a function of constellation coverage and is described in the experimental configuration. Under this observing strategy, global coverage is achieved is by increasing the number of orbits covered by the constellation. In global terms, four orbital planes, with each plane consisting of three MISTiC instruments flying in a configuration to overlap the ground locations, would provide a 3-h refresh rate. No actual feature tracking is performed in this study, as described in the MISTiC observation generation section later. As such the nominal spacing between the three instruments in an orbital plane is not considered. Rather, the center instrument is considered for its geolocation, and full overlap is considered along the entire orbit.
For brevity hereinafter, MISTiC is used to describe the entire observing system, including the radiances measured by the instrument and the AMVs derived from the temporal sequence of instrument overpasses.
3. Baseline OSSE system configuration
The GMAO OSSE is a highly sophisticated, well-validated framework for the evaluation of potential new observing systems. The GMAO OSSE has several main components: a nature run (hereinafter NR), which replaces the true atmosphere for the study; a comprehensive set of global observations simulated from the NR fields that mimics the actual observational data used in operational numerical weather prediction; and a data assimilation system and forecasting model used to conduct the various experiments for the new observing system.
The Goddard Earth Observing System, version 5 (GEOS-5; Rienecker et al. 2008), Nature Run (G5NR; Gelaro et al. 2015) is a well-validated, 2-yr, free-running integration of the GEOS atmospheric model, starting from 1 May 2005, at a horizontal resolution of approximately 7 km and 72 vertical levels ranging from the surface to 0.01 hPa. The G5NR includes 16 aerosols and has output every 30 min. Although the boundary conditions (sea surface temperatures, ice fractions, etc.) are for 2005–06, the atmospheric state in the free-running model does not represent the actual atmosphere for that period. Instead, the intention of the G5NR is to provide a physically and dynamically consistent simulation of a possible atmospheric state that mimics in a statistical way the physical nature of the real atmosphere.
The simulated observations were generated by sampling the G5NR fields at specified spatiotemporal observation locations and using forward operators to project the G5NR into observation space. For this study, real observations from June to September of 2015 were used as a basis for generating simulated observations. The baseline observing system is listed in Table 1, and the observations described in this section refer specifically to the control (CTL). The complexity of this forward transformation differs by observation type and is described in Errico et al. (2017). The extension of these methods to the MISTiC observations are addressed later in section 4.
List of observations in the control (CTL), the perfect MISTiC observations experiment (4PERF), and the error-modeled MISTiC observations experiment (4ERR). All observations in CTL contain modeled errors.
In the real world, there are many sources of error that can impact observations and their use, including instrument errors, representativeness errors, as well as errors introduced by the data assimilation process in handling the observations. Simulated observations tend to have less intrinsic errors than are estimated for real observations. Because the data assimilation procedure relies on estimates of observation errors when determining how to ingest the observational data, it is important to add realistic errors for the simulated observations, in order to accurately portray the new data types in the OSSE. To this end, two types of errors were generated and added to the simulated observations: uncorrelated random errors and correlated random errors.
Using the methods of Errico et al. (2017), uncorrelated random errors were added to all observational types, and correlated errors are added to select data types. Horizontally correlated errors were added to microwave radiances and atmospheric wind vectors, vertically correlated errors were added to rawinsondes, GPS-RO, and atmospheric wind vectors, and spectrally correlated errors were added to the Infrared Atmospheric Sounding Interferometer (IASI), AIRS, and CrIS data. The magnitude and types of simulated errors were adjusted as a method of calibrating the simulated observations to match particular statistical metrics of the data assimilation process. The length scales of correlated errors were chosen to match observed measurement correlations of real data for those data types that receive correlated errors. For all simulated observations, the variance of the background departure, defined as the difference between the observation and that simulated from the short-term forecasted background state, was matched as closely as possible in the OSSE system to the same metrics for real observations. Further details of the methods used to generate the simulated observations and their errors can be found in Errico et al. (2017, 2020).
The atmospheric model used for the OSSE experiments is GEOS atmospheric data assimilation system (ADAS; Rienecker et al. 2008), version 5.14.2p2. The GEOS model (Molod et al. 2015) was run on the cubed-sphere dynamical core (Putman and Lin 2007) with a spatial resolution of 25 km on 72 hybrid-eta levels to 0.01 hPa. The analysis component of the ADAS is the Gridpoint Statistical Interpolation analysis system (GSI; Wu et al. 2002; Kleist et al. 2009) data assimilation procedure, and the analysis is generated at 0.5° × 0.625° using a 6-h assimilation window and a 3D-Var assimilation solution. Within the GSI, the CRTM (Han et al. 2006; Chen et al. 2008) was used for the assimilation of satellite radiances.
To reduce the similarity of the models used in the G5NR and the ADAS, changes were made to the physics package in the experiment forward model. The most significant change was a switch of the microphysical convection scheme from the default single moment (Bacmeister et al. 2006), which was also used in the G5NR, to a more sophisticated two-moment microphysics scheme (Barahona et al. 2014). This change was most impactful in regions that experience frequent moist convection, such as the tropics and the summer hemisphere midlatitudes. Additionally, a relative humidity threshold that also impacts moist convection was adjusted, and a slightly older version of the boundary layer physics scheme was used. These changes were observed to have a modest but positive effect on the realism of the observation impacts seen in the OSSE. These configuration changes help represent a source of model error between the G5NR and the ADAS model, which is representative of the model errors that exist between the ADAS and nature when assimilating real data. Without these configuration adjustments, the system would be considered as suboptimal “identical twin” experiments (Privé and Errico 2013).
This baseline OSSE system has been under continual development and advancement at the GMAO for over a decade. The methods for validation have been presented in Errico et al. (2013), Privé et al. (2013a,b), and Privé and Errico (2013). The ongoing application of these methods have been shown in Errico et al. (2017), Privé and Errico (2019), and Privé et al. (2021).
4. Assimilated MISTiC observations
a. MISTiC radiances
MISTiC radiance measurements were simulated and assimilated using the CRTM. Core to this is that forward calculations are performed using a fast transmittance model (Chen et al. 2012), which was trained as a function of the instrument’s characteristics. For MISTiC, new coefficients were needed to represent the instrument since no predecessor instrument exists. These coefficients were generated using instrument specifications such as the center frequency and the response function of each channel. Transmittance coefficients were regressed against transmittance profiles generated from the ECMWF 83 standard profile (Chevallier et al. 2006) dataset and the line-by-line radiative transfer model (LBLRTM; Clough et al. 1992, 2005).
The MISTiC spectral response functions (SRFs) were considered to be triangular with 2:1 oversampling ratio, as illustrated in Fig. 1, and a CRTM-calculated MISTiC spectrum instrument is shown relative to simulations for IASI and AIRS instruments in Fig. 2. Although there was overall a good agreement between the MISTiC and IASI/AIRS simulations, the MISTiC simulations for the channels operating above 2500 cm−1 were notably different from IASI and AIRS simulations. Further investigation after this study found that the discrepancy was due to a misspecification of the absorption continuum within LBLRTM in the coefficient training procedure at these high wavenumbers, but this discrepancy did not affect the results of this study.
Sample spectral response functions for the MISTiC radiometer, as specified by BAE.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
Example brightness temperature spectra for AIRS (purple), IASI (black), and MISTiC (green), all calculated for the same temperature profile. The red dots indicate the subset of MISTiC radiance bands that was selected for assimilation in the experiments.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
In all simulated infrared radiance measurements in the OSSE, the G5NR clouds were used to simulate simplistic clouds in the observations via a graybody assumption. Fundamentally, clouds are signal not error, even if they are actively avoided by most assimilation schemes via quality control procedures. The methods used to prescribe the cloudiness in infrared observations are described in Errico et al. (2017). The cloud tuning parameters were consistent with those of other infrared sounders; however, due to the fact that MISTiC has a finer spatial resolution, the effective probability of clear-sky measurements based on this scheme was higher than heritage sounders with larger footprints.
For the case of simulated MISTiC radiances that include added simulated error, both uncorrelated and correlated errors were applied. The methods of Desroziers et al. (2005) were used to estimate the error components via the use of IASI radiances convolved onto the MISTiC spectrum. The magnitude of the errors ranged from 0.6 to 2.1 K, accounting for both instrument and representativeness errors. The correlated errors were constructed using the methods of Errico et al. (2017) and applied using a horizontal decorrelation length of 120 km and using the spectral error correlation matrix presented in Fig. 3. Of note, the use of real, convolved IASI data in error estimation procedure will project any shortcomings in the assimilation procedure (i.e., shortcomings in radiative transfer modeling) into simulated error space. These correlations are important as they represent a real source of representativeness error that is routinely assessed in the assimilation of real data. They correspond to vertical and spatial misrepresentations between the underlying assumptions of the assimilation system, misrepresentations of the scales of the model resolution and reality, and how they project into observation space.
Correlated error matrix used to add spectrally correlated errors to the MISTiC radiances.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
Although the observing strategy is built upon three instruments flying in formation, the radiances were only simulated for the middle platform on each orbital plane represented in this study. The rationale for this approach was twofold. First, this study was built around a 3D-Var system, so it was not capable of exploiting any temporal information content among the three platforms in a single orbit. Second, even the most current four dimensional data assimilation systems at the GMAO assume an hourly time binning of observations, which is below the expected separation of the three instruments per orbital plane—estimated to be around 15 min based on geostationary AMV applications.
b. MISTiC winds atmospheric motion vectors
As the observing system was designed around the concept of deriving AMVs on retrieved pressure surfaces, it was core to this study to consider the information content of these observations. As such, the AMVs were decomposed into two types—those derived from tracking clouds (Schmetz et al. 1993; Le Marshall et al. 1994) and from tracking localized water vapor anomalies (Velden et al. 1997).
However, these observations were difficult to simulate within an OSSE. AMVs are dependent on the underlying atmospheric state (Errico et al. 2020). It was not adequate to simply use the observed locations of AMVs from the real-world observations used in the calibration phase of the OSSE. Furthermore, while the G5NR has a grid spacing of 7 km, its effective resolution is less than that because of various diffusion and smoothing methods implemented to improve numerical stability. Because of this inherent coarseness, featuring tracking the NR was not a reasonable option for the proposed MISTiC spatial resolution. Therefore, it was essential for this study to leverage a method for simulating the MISTiC AMVs that was a function of the underlying meteorology of the G5NR but also reflective of the retrieval-based approach.
In an attempt to determine AMV locations that were a function of the meteorology of the G5NR, a probabilistic approach was developed and presented in Carvalho and McCarty (2020). While the method was simplistic, it was tuned to give results similar to those from the modern era of geostationary imagers while also including more information content vertically—particularly in the midtroposphere where the retrieval-based tracking approach is expected to give more information. For cloud-tracked AMVs, the likelihood of an observation was a function of the columnar cloud fractions as determined by the three-dimensional G5NR fields and a maximum random overlap assumption. For water vapor-derived AMVs, the likelihood of an observation was a function of the horizontal water vapor gradient and the cloud fraction profile above a given level. Since the MISTiC approach proposes feature tracking on retrieved pressure surfaces as opposed to radiance space, these levels were predefined as 150, 200, 250, 300, 400, 600, and 700 hPa. These levels were chosen because they approximately reflect the vertical pieces of water vapor information content seen in an AIRS spectrum (Irion et al. 2018). It is noted that the water vapor AMVs do not guarantee vertical profiling; each isobaric surface is tested individually.
The errors applied to these measurements were based on those applied to Himawari AMVs in the baseline observing system, again using the methods of Errico et al. (2017). For cloud-derived AMVs, this corresponded to an added error magnitude ranging from 0.9 to 3.2 m s−1 as a function of height and a horizontal decorrelation length of 200 km. For water vapor AMVs, the error magnitude ranged from 0.8 to 3.9 m s−1 as a function of height and the decorrelation length was 300 km. Both AMV types had a vertical decorrelation length of 350 m. While perhaps not ideal, the lack of real data against which the error models could be approximated led to the use of Himawari. Himawari was chosen as it was the most advanced platform from which AMVs were derived during the baseline period. Also, a 25-km spatial thinning was performed, which was akin to the resolution of the model background fields.
5. Experiment design
The baseline OSSE experiment, described in section 3, served as the CTL for experimentation, and its observations are shown in Table 1. Two experiments, also shown in Table 1, were performed on this control. First, perfect radiance and AMV observations from a MISTiC Winds constellation consisting of four orbital planes were added to the control. This experiment is hereinafter referred to as 4PERF. Second, the same observations but with added errors—or imperfect observations—were considered. This experiment is hereinafter referred to as 4ERR. For all experimentation, a 15-day spinup was performed and discarded. After spinup, experimentation focused on the period of 1 July–31 August 2006 OSSE time.
In the assimilation of the hyperspectral infrared radiances, only a subset of channels was considered, consistent with other efforts (e.g., McNally et al. 2006; McCarty et al. 2009). The reasoning for this was multifaceted. First, this was done to mitigate the redundancy and correlations that exist in spectral space. While these correlations were added to the simulated observations, they were not accounted for in the assimilation procedure. Second, this also avoided sensitivities to constituents that were not well represented in the forward calculations.
The selected channels are shown in Fig. 2, and the thermal contrast in brightness temperature space is akin to spectral information content. A principal component analysis of an ensemble of calculated MISTiC radiances determined that there were approximately 49 pieces of information within its spectrum, and a channel selection exercise resulted in 46 channels selected. Two key points are noted in terms of the channel selection. First, uncertainty in the simulation of the shortwave side of the 4-μm CO2 absorption band resulted in the use of the less-optimal longwave side, which has sensitivity to carbon monoxide and nitrous oxide. In practice, these trace constituents are not considered in a real data assimilation system. However, for this OSSE, both are considered to follow the same climatology in both simulation and assimilation of these bands. However, this side of the absorption band is also less affected by solar contamination issues. Second, channels with wavenumbers greater than 2275 cm−1 were excluded as nonlocal thermodynamic equilibrium effects were not properly represented in the radiative transfer (DeSouza-Machado et al. 2007).
6. Results
The results of the experimentation are presented in the three categories. First, the performance of the data assimilation procedure in the context of MISTiC Winds is presented. Second, the impact of the measurements on the atmospheric forecasts is shown. Third, the observations are assessed in the context of the FSOI (Langland and Baker 2004; Zhu and Gelaro 2008; Gelaro and Zhu 2009) metric.
a. Analysis results
In the assimilation of real observations, the analysis procedure aims to minimize the analysis error by considering the error statistics of both the background fields and the observations. However, it is not possible to directly quantify that error globally as the truth is unknown. In an OSSE, the NR is the “truth” of the atmosphere. Thus, a strength of an OSSE is that the analysis error can be quantified by comparing the analysis solution directly to the NR.
The zonally averaged analysis error variance of the zonal wind for CTL is shown in Fig. 4 (top). The largest tropospheric errors were seen in the upper tropical troposphere, with a mean analysis error variance of 8.99 m2 s−2 over a layer ranging from 20°S to 20°N and from 100 to 300 hPa. This uncertainty extended toward the surface. In this region, the system lacked columnar wind observations, particularly in the middle troposphere, relative to mass information derived from passively measured microwave and infrared radiance observations. Additionally, the uncertainty affects the convective processes in the tropics, which are driven heavily by the model parameterizations. In the midlatitudes, atmospheric balance constraints imposed in the data assimilation solution infer wind information from the mass observations. However, these constraints are less applicable toward the equator, and thus minimal wind information was inferred from the radiances.
(top) Zonal wind error variance for the CTL vs the nature run, and the change in error variance relative to the CTL in (middle) 4PERF and (bottom) 4ERR. In the bottom two panels, blue or red indicates an improvement or degradation, respectively, by the addition of the MISTiC observations.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
The change in error by the addition of the MISTiC observations is shown for 4PERF (Fig. 4, middle) and 4ERR (Fig. 4, bottom). In the tropical upper troposphere, the error variance was reduced by 18.9% when considering the full suite of perfect MISTiC observations in 4PERF. In the 4ERR experiment, the addition of modeled observation errors was shown to reduce the magnitude to 15.6%. The lesser reduction in 4ERR was expected due to added MISTiC observation errors. The tropical improvements extended into the middle and lower troposphere for both experiments.
While the aforementioned balance constraints resulted in the winds being better analyzed in the midlatitudes, secondary maxima of wind error were still seen in both hemispheres in the upper troposphere. To quantify for the layer of 200–400 hPa, the mean analysis error variance of the CTL was 3.78 and 5.27 m2 s−2 in the Northern and Southern Hemisphere, respectively.
For the Southern Hemisphere upper-tropospheric layer, the 4PERF experiment reduced the error variance by 18.8% relative to the CTL. The magnitude of error reduction decreased to 1.48% for the 4ERR experiment. For the Northern Hemisphere, the perfect observations of 4PERF resulted in a 9.8% error reduction. This was less than the magnitudes seen in the tropics and Southern Hemisphere due to the relative abundance of conventional observations in the Northern Hemisphere. The error-added observations increased the error variance by 1.3% in the 4ERR experimentation. The sensitivity of analysis error to the added MISTiC observation errors was larger in both extratropical hemispheres than in the tropics.
There were additional maxima in terms of wind error variance in the lower polar troposphere in both hemispheres. Based on regions defined as poleward of 60°S or 60°N and from 1000 to 800 hPa, the zonal error wind variance was 3.8 and 2.3 m2 s−2 in the Antarctic and Arctic, respectively. In the Antarctic, this error variance was reduced by 33.9% and 11.7% for 4PERF and 4ERR, respectively. In the Arctic, this error variance was reduced by 39.7% and 12.2% for 4PERF and 4ERR, respectively. While these impacts were large, there was also some concern that the AMV estimator may have generated too many AMVs in this region.
The largest improvements in the zonal-mean temperature error variance were seen near the surface, as shown in Fig. 5. The largest error variances in the CTL were seen in the upper troposphere/lower stratosphere in the tropics, the near surface in both hemispheres near the poles, and above the surface in the midlatitudes. Specifically, the largest error variances were 1.8 and 1.7 K2 in the Antarctic and Arctic lower troposphere, respectively.
As in Fig. 4, but for temperature error variance.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
In the tropics, the zonal mean temperature error variance was reduced by the addition of MISTiC observations across both experiments and all levels, though the absolute magnitudes were small in this sense. The upper-, middle-, and lower-tropospheric temperature error variances were reduced by 8.5%, 8.3%, and 8.8%, respectively, in the 4PERF experiment, and these values decreased to 3.3%, 2.4%, and 3.0% in 4ERR.
In the extratropical upper troposphere, the 4PERF experiment indicated a signal of degradation, particularly near 80°S and 30°N. Relative to CTL, the mean error variance reduction for the Southern Hemisphere upper troposphere was 0.6% and the observations acted to increase the error variance by 6.7% in the Northern Hemisphere. The added errors in 4ERR resulted in the error variance increasing by 8.6% and 13.3% in the southern and northern extratropical upper troposphere, respectively.
In the polar regions, the zonally averaged temperature error variance was reduced through the column in 4PERF, with a larger impact seen toward the northern pole (Fig. 5, middle, bottom). In 4ERR, the polar middle-troposphere improvement seen in 4PERF was absent, but some of the near-surface polar improvement was retained. To quantify, the near-surface error reduction decreased from 16.6% to 4.6% from 4PERF to 4ERR.
Last, the impact on the extrapolar near-surface was considered—specifically for the regions bounded by 60°–10°S and 10°–60°N and from 900 to 700 hPa. In the Southern Hemisphere, this region encapsulated the third largest of the error variance maxima, with a magnitude of 1.2 K2. The addition of perfect observations reduced this error by 12.9% in 4PERF. The addition of errors to the MISTiC observations lessened the magnitude of the reduction to 8.0%. The Northern Hemisphere region had an error variance of 0.6 K2, which was smaller than its Southern Hemisphere counterpart. The error variances were reduced by 11.2% and 5.4% in 4PERF and 4ERR, respectively, in this region. Though not shown, the largest magnitudes in this layer were seen to be over the oceans, particularly in regions of persistent marine stratocumulus clouds.
The zonally averaged specific humidity analysis error is shown in Fig. 6 (top). The largest errors reflect the global climatological distribution of water vapor. Therefore, the changes in error variance by the addition of MISTiC observations are shown in terms of relative difference in Fig. 6 (middle and bottom), where the changes were normalized by the error variance of the CTL.
As in Fig. 4, but for specific humidity error variance. The middle and bottom panels indicate relative change.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
Broadly, there was a general reduction in the tropospheric specific humidity error variance at all latitudes in both experiments. However, there was a degradation that followed the tropopause, and the degradation extended toward 300 hPa between 30° and 50°N. In the tropics, the MISTiC observations consistently reduced the analysis error in 4PERF throughout the troposphere (Fig. 6, middle). When the simulated errors are added to the MISTiC observations, this signal was consistently retained at a slightly lower magnitude in 4ERR (Fig. 6, bottom).
In the Southern Hemisphere extratropical middle and upper troposphere, there again was consistent improvement below the tropopause by the addition of both perfect and erroneous MISTiC observations. More than half of the error reduction in 4PERF was retained in 4ERR. In the Northern Hemisphere, the upper-tropospheric improvement was less apparent, largely due to the degradation in the middle and upper troposphere between 30° and 50°N. In the near-surface polar regions, 4PERF showed an improvement through the polar troposphere in both hemispheres. In 4ERR, the magnitude of the reduction was again lessened.
b. Forecast results
The effect of the MISTiC AMVs and radiances on forecast scores for the two experiments was considered. All results shown were for 5-day forecasts initialized at 0000 UTC from G5NR dates 1 July to 31 August 2006. For all of these results, the forecasts were validated against the G5NR.
Figure 7 shows the difference in forecasted height anomaly correlation for 4PERF and 4ERR relative to the control for the Northern and Southern Hemispheres. The height anomalies were calculated by removing the 30-yr climatological mean height patterns, which were derived from MERRA-2 (Gelaro et al. 2017). A positive shading indicated that the forecasted height anomaly at a given height and forecast hour was more correlated to the G5NR in the experiment relative to CTL, thus indicating an improved forecast. Contrarily, a negative shading indicated a forecast degradation in the experiment compared to the control. Stippling on the figures indicated a statistical confidence to 95%.
The change in height anomaly correlation as a function of pressure height and forecast time for (left) 4PERF and (right) 4ERR in the (top) Northern and (bottom) Southern Hemispheres. Stippling indicates 0.95 significance.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
When considering the Northern Hemisphere, 4PERF (Fig. 7, top left) showed forecast improvement maintaining statistical significance throughout the column to day 5; 4ERR (Fig. 7, top right) showed statistically significant improvement through day 5 at 700 hPa and below. Above 700 hPa, however, the improvement did not maintain statistical significance from day 2.5 to day 4 but was significantly positive otherwise. Consistent with the results of the previous section, the magnitude of the 4ERR improvements relative to the control was reduced in magnitude compared to 4PERF. In both cases, the largest improvements were near the surface in this hemisphere.
In the Southern Hemisphere, the 4PERF (Fig. 7, bottom left) forecast improvement was positive and statistically significant improvement over CTL through the column and over the entire forecast period. The improvement in height anomaly correlation in the middle troposphere and significance was maintained to day 5 between 200 and 500 hPa. In 4ERR, (Fig. 7, bottom right) forecast anomaly correlation differences were positive though statistically significant improvement was not maintained past day 3.5.
In terms of magnitude of the difference relative to CTL, larger improvement by this metric was seen in the Northern Hemisphere than in the Southern Hemisphere. In one sense, this was contrary to common thought that improvement due to satellite data should be largest in the Southern Hemisphere because there are more conventional observations in the Northern Hemisphere. However, forecast gaps are generally not as readily apparent between the hemispheres in modern times (Bauer et al. 2015; Diniz and Todling 2020). This difference may be more indicative of a seasonal difference, where the extra wind and radiance information content could be more useful in the Northern Hemisphere summer than in the Southern Hemisphere winter. Testing in the complementary season, which was not done in this study, would be necessary to pinpoint the sensitivity.
To consider the impact of the MISTiC observations on the tropics, the two experiments are compared to the CTL in terms of the RMS difference relative to the G5NR in Fig. 8 for zonal wind, temperature, and specific humidity. For RMS difference, a lower value indicates a more accurate solution than that of CTL.
The change in the tropical RMS as a function of pressure height and forecast time for (left) 4PERF and (right) 4ERR for the (top) zonal wind, (middle) temperature, and (bottom) specific humidity. Stippling indicates 0.95 significance.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
The winds at all levels in both experiments (Fig. 8, top) were improved consistently at early forecast periods. To account for increasing wind speeds with height, the relative impact on the zonal wind is shown. There were two maxima of improvement relative to the CTL, seen in the lower-troposphere peaking at 850 hPa and in the upper-troposphere peaking near 200 hPa. For the 4PERF and 4ERR experiments, improvement throughout the column was seen to day 2 and day 2.5, respectively. The upper-tropospheric improvement continued to be maintained to day 4 and day 5 for 4PERF and 4ERR, while the lower-tropospheric improvement maintained significance to day 5 in both.
For temperature, the 4PERF experiment (Fig. 8, middle left) showed the most improvement compared to the CTL, though there was a degradation to day one at 150 hPa. Improvement was maintained through the entire forecast period for 400–200 hPa. The early period forecast degradation at 150 hPa changed sign at days 4.5 and 5, but the improvement was not significant. The 4ERR (Fig. 8, middle right) experiment showed similar results compared to the CTL as those of the 4PERF experiment, but of a lesser magnitude. Of note, however, is that the degradation at 150 hPa early in the forecast period was still present, but the improvement at this level at days 4 and 5 were statistically significant.
For both experiments, the water vapor forecast improvement showed a bimodality in the tropics, peaking near 500 and 200 hPa. In both 4PERF (Fig. 8, bottom left) and 4ERR (Fig. 8, bottom right), the lower-tropospheric improvement extended to day 5. There were significant improvement in the upper troposphere seen to extend to near the end of the forecast—specifically, improvements at 300 and 200 hPa are seen to be significant to day 4.5 in 4PERF, and improvement at 300 hPa is seen to be significant at day 5 in 4ERR.
The improvement in water vapor is likely related to three key facts. First, MISTiC radiances fully sampled the shortwave side of the 6.7-μm water vapor continuum. Thus, as a sounding instrument, it was capable of vertically resolving water vapor. These radiances were assimilated, and part of this improvement fundamentally resulted in an improved characterization of water vapor in the initial conditions. Second, the AMVs derived from the constellation approach resulted in improved short-term forecasts due to improved water vapor transport. Third, the improvement in the medium-range forecasts—days 3–5—was additionally driven by improved synoptic forecasts due to the MISTiC observations. By these longer ranges, the water vapor fields were largely driven by model physics and not initial conditions. This was apparent as the large improvements seen early in the forecasts are not fully retained throughout the forecast.
c. FSOI results
The MISTiC observations were also characterized in the context of the analysis using the FSOI (Langland and Baker 2004; Zhu and Gelaro 2008; Gelaro and Zhu 2009) metric. FSOI is a quantified assessment of how each observation acts to change 24 h forecast error, where the forecast error was integrated across variables and quantified using a moist energy norm (Ehrendorfer et al. 1999; Holdaway et al. 2014).
To provide an overall context to the role of the MISTiC AMVs and radiances with the respect to the overall global observing system, the global FSOI, or “impact” for brevity, per analysis and as a function of each component of the observing system is shown in Fig. 9. In 4PERF, the MISTiC AMVs ranked highest and the radiances ranked third. In 4ERR, MISTiC AMVs ranked second, dropping one position. The radiances were ranked eleventh, which was a drop of eight positions relative to 4PERF. However, for total impact, it is caveated that these data are known to carry larger FSOI total impact magnitudes than seen in reality, as illustrated in Privé and Errico (2019). So, although the AMVs show a large total FSOI, it may be overstated.
FSOI per analysis as a function of observing system component for (left) 4PERF and (right) 4ERR.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
For the MISTiC AMVs, the assimilated counts, FSOI per observation, and FSOI per analysis cycle are shown globally as a function of height in Fig. 10 for both experiments. A bimodality was seen in the observation counts, with maxima in the upper and lower troposphere, which is corroborated by observations and the G5NR (Wylie et al. 2007; Gelaro et al. 2015). However, if these were compared with the standard AMV observing systems, the wind count minimum in the midtroposphere would be far more pronounced. This is by construction of the estimated MISTiC AMV distributions, as the increased spectral resolution was expected to increase midtropospheric wind measurements.
MISTiC AMV observation (left) count per analysis, (center) FSOI per observation, and (right) FSOI per analysis as a function of pressure (i.e., height).
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
While the counts had a maximum at 250 hPa, the impacts per analysis (Fig. 10, right) are seen to be relatively low. This was due to an existing abundance of AMV observations, particularly from geostationary imagers. The same can be said for low-level winds corresponding to the maximum at 850 hPa. This indicates that the MISTiC AMVs were not adding new information to the baseline observing system in the CTL, and this is further emphasized by considering the FSOI per observation. However, at 150 hPa, the per observation metric shows a peak, though counts and total FSOI are low. In simulation, observations are constrained to be below the tropopause, but the AMV distributions for MISTiC may not be fully accounting for uncertainties due to the lack of thermal contrast near the tropopause. Further investigation would need to be performed to determine if these AMVs could be reasonably measured using the proposed MISTiC method.
A fundamental argument for the AMV from sounding method is improved and increased wind measurements in the midtroposphere. This was emphasized by the impact metric. For total FSOI (Fig. 10, right), the largest impact was in the region that corresponds to the International Satellite Cloud Climatology Project definition (Rossow and Schiffer 1991) of the midtroposphere—from 700 to 440 hPa. Similarly, the per observation FSOI was maximized in this region. This was because the observations were filling a data void and thus providing unique information to the global observing system. The addition of observation errors on the AMVs were shown to have a limited impact in that they only reduced the FSOI for these data.
The total FSOI per analysis for the MISTiC radiances is shown in the context of other sounding instruments in Fig. 11. In both experiments, the MISTiC radiances were separated by orbital plane, with MISTiC 0130, MISTiC 0730, MISTiC 1330, and MISTiC 1930 corresponding to the center instruments per orbit with an LTAN of 0130, 0730, 1330, and 1930 local time, respectively. In 4PERF, each MISTiC sounder ranked among the other infrared and microwave sounders, particularly IASI, AIRS, Advanced Technology Microwave Sounder (ATMS), and the healthy Advanced Microwave Sounding Unit-A (AMSU-A) instruments.
FSOI per analysis as a function of satellite radiance instrument for (left) 4PERF and (right) 4ERR.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
In 4ERR the impact of the radiances decreased dramatically. This was already shown in Fig. 9 but is reillustrated from a per-platform perspective. In these experiments, the ranking is similar to CrIS, which was known to be used in a suboptimal manner in this version of the ADAS, the Special Sensor Microwave Imager/Sounder (SSMIS) instrument, on which key bands known to have a large effect on the FSOI metric had failed, and the High Resolution Infrared Radiation Sounder (HIRS/4) on MetOp-A. While healthy, HIRS/4 was a heritage infrared instrument and measured two-orders-of-magnitude fewer observations with broader spectral response functions over the coincident spectral regions.
To further understand the degradation of the FSOI metric with the addition of simulated observation errors, the impact per analysis is presented as a function of channel index in Fig. 12. In 4PERF, there were few bands seen to cause a degradation in the analysis, as denoted by a positive FSOI, but these degradations are very small. The largest benefits, as denoted by a negative FSOI, were seen in the temperature-sensitive channels around 2250 cm−1, with a secondary benefit seen in the midtropospheric water vapor channels around 1860 cm−1.
MISTiC radiance FSOI per analysis as a function of channel index for (left) 4PERF and (right) 4ERR.
Citation: Journal of Atmospheric and Oceanic Technology 38, 2; 10.1175/JTECH-D-20-0109.1
When the simulated observation errors were applied, the benefit corresponding to the water vapor channels was largely retained between 4PERF and 4ERR. However, the benefit seen in the temperature channels was lost and changed sign indicating a net degradation. Since real data—IASI measurements convolved to the MISTiC spectral resolution—were used in the determination of the correlated error structures in spectral space, we have some confidence that these degradations were legitimate and expected within our assimilation system.
These degradations were likely explicitly due to shortcomings in handling shortwave channels within the data assimilation procedure. The CRTM, which was used both in simulation and assimilation of the MISTiC radiances, had shortcomings in these regions—particularly in the characterization of nonlocal thermodynamic equilibrium effects, solar absorption and reflectance, and surface reflectance. In the simulation and assimilation of perfect observations, these shortcomings are handled consistently. Therefore, the information content in the perfect observation experiments is considered reasonable, though overstated. However, improved radiative transfer and data assimilation methods can be developed to retain more benefit from these bands (Jones et al. 2020).
7. Summary and conclusions
The results of this study illustrate the potential impact of a constellation capable of providing radiance and three-dimensional wind measurements on a global forecast system. However, from a high level, these results should be taken with some amount of caveat. An OSSE should be viewed as a tool in a toolkit, and this paper attempts to not only present the results of the experimentation but to also lay out the assumptions that must be made to simulate, integrate, and assimilate observations from a proposed future observing system.
Radiance observations were simulated using specifications provided by BAE. While some effort was made to simulate errors that represented the underlying uncertainty in this portion of the spectral domain by using corresponding convolved IASI data, that method in and of itself was subject to potential limitations. Furthermore, any potential error sources such as calibration and geolocation uncertainties were entirely unaddressed in this study, though they may be notable in the realm of miniaturized instrumentation and small satellites. It is very difficult to fully model both the instrument and representativeness errors for a proposed instrument as it must depend on the use of assumptions and proxy datasets.
Also, even as correlated errors were added to the MISTiC radiance observations, the channels were assumed to be uncorrelated in the assimilation solution. The current GMAO ADAS now accounts for these spectrally correlated errors for AIRS, CrIS, and IASI, but this capability has not been advanced and tested within the GMAO OSSE framework. Also, temperature sounding channels from the 15-μm CO2 absorption band are typically assimilated in lieu of those from 4.3 μm. Although not shown, it was verified that the ADAS was able to assimilate those observations in a stable manner by confirming that cloud screening was effective in this region and that bias correction coefficient remained stable. However, both the lack of accounting for correlated errors and the general lack of maturity in assimilating 4.3-μm bands likely resulted in the degraded performance of the 4ERR case, as illustrated by the FSOI results. Nonetheless, the 4PERF experimentation still illustrated the information content of this spectral region, and that is in line with NASA heritage retrieval methods focusing on temperature retrieval at 4.3 μm (Susskind et al. 1998).
Similarly, the simulation of AMVs in this study attempted to ensure that wind measurements were assimilated in regions of the G5NR that were meteorologically consistent with this observing strategy. Entire research projects can and should continue to advance this capability, as global nature runs continue to lack the effective spatial resolution needed for full end-to-end AMV simulation. As other studies are performed, there remains a need for a robust, flexible, and portable AMV simulation method within the OSSE community. Nonetheless, should the proposed AMV method in fact fill the midtropospheric data void of wind measurements, this study illustrated that those observations would provide the largest contributions in short-term forecast error, as illustrated by vertical profiles of FSOI.
Another limitation is that this study was performed in the auspice of a three-dimensional data assimilation system. Current experimentation at the GMAO has since evolved to a hybrid four-dimensional ensemble-variational assimilation method for experimentation in both real data and OSSE frameworks. This study predates the advancement of the OSSE to the new algorithm. While the 4D solution would more effectively get the large-scale flow information out of the global observing system, it is not expected to have a significant impact on the overall results of the OSSE, particularly as it relates to solving for the atmospheric motion based on the observed evolution of the radiance fields, also known as the “tracer effect” (Peubey and McNally 2009). The current 4D-EnVar system at the GMAO assumes hourly observation binning, which is coarser than the spacing between individual radiometers in each individual orbital plane, though some tracer-derived flow could be captures from the assimilation of the instrumentation from multiple orbital planes.
Also, these results are somewhat comparable, and potentially offset, by future geostationary hyperspectral infrared sounders. The current hyperspectral geostationary sounder, the Geostationary Interferometric Infrared Sounder (GIIRS) instrument on board the China Meteorological Administration Feng-Yun-4A satellite, is limited in terms of the spatial resolution and extent. The upcoming Infrared Sounder (IRS) on the Meteosat Third Generation Sounder satellite will provide broader spatial coverage and finer spatial resolution, but its scanning schedule will provide full-disk coverage every three hours, which is coarser than the constellation approach and potentially limiting for AMV retrievals. In both cases, the differences are due to resolution compensations in dealing with the higher satellite orbit height for geostationary coverage. Furthermore, geostationary platforms will still lead to global coverage gaps, both in terms of the need for a full “ring” of geostationary satellites and in polar coverage, including the LEO–GEO gap.
While this study aims to be forthright with the assumptions, limitations, and critical analysis of the results, it does successfully characterize the potential contributions of the constellation approach to measuring radiances and retrieving AMVs from space in the context of global modeling and data assimilation system. Should the proposed method be successful, this study illustrates the current system’s potential benefit with an increase in midtropospheric wind observations and shows the potential information content of assimilating the 4.3-μm CO2 channels for temperature. However, to fully utilize these data, particularly the radiances, some effort is needed to optimize the assimilation method for these data. This is achievable by integrating existing methods used in retrievals today into data assimilation procedures—namely, accounting for nonlinearity in the observation errors, improved handling of the solar spectrum within the observations, and ensuring that fast radiative transfer methods are fully capable of handling these observations.
Acknowledgments
This work was cofunded by the NASA Modeling, Analysis and Prediction (MAP) Program and by the NASA Earth Science Technology Office (ESTO). Computing resources supporting this work were provided by the NASA High-End Computing (HEC) Program through the NASA Center for Climate Simulation (NCCS) at Goddard Space Flight Center. The software for simulating GPS-RO observations was provided by the Radio Occultation Processing Package (ROPP) of the Radio Occultation Meteorology (ROM) Satellite Applications Facility (SAF) or EUMETSAT, with the assistance of Sean Healy at ECMWF. Specific thanks are given to Ronald M. Errico, whose baseline system made this experimentation possible. The authors also thank John Blaisdell, Mohar Chattopadhyay, Marangelly Cordero-Fuentes, Ronald Gelaro, Meta Sienkiewicz, and Steven Pawson who all provided additional support to this work.
REFERENCES
Arnold, C. P., and C. H. Dey, 1986: Observing-systems simulation experiments: Past, present, and future. Bull. Amer. Meteor. Soc., 67, 687–695, https://doi.org/10.1175/1520-0477(1986)067<0687:OSSEPP>2.0.CO;2.
Atlas, R., E. Kalnay, and M. Halem, 1985: Impact of satellite temperature sounding and wind data on numerical weather prediction. Opt. Eng., 24, 242341, https://doi.org/10.1117/12.7973481.
Bacmeister, J. T., M. J. Suarez, and F. R. Robertson, 2006: Rain reevaporation, boundary layer–convection interactions, and Pacific rainfall patterns in an AGCM. J. Atmos. Sci., 63, 3383–3403, https://doi.org/10.1175/JAS3791.1.
Barahona, D., A. Molod, J. Bacmeister, A. Nenes, A. Gettelman, H. Morrison, V. Phillips, and A. Eichmann, 2014: Development of two-moment cloud microphysics for liquid and ice within the NASA Goddard Earth Observing System Model (GEOS-5). Geosci. Model Dev., 7, 1733–1766, https://doi.org/10.5194/gmd-7-1733-2014.
Bauer, P., A. Thorpe, and G. Brunet, 2015: The quiet revolution of numerical weather prediction. Nature, 525, 47–55, https://doi.org/10.1038/nature14956.
Carvalho, D., and W. McCarty, 2020: Position estimation of atmospheric motion vectors for observation system simulation experiments. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, Vol. 54, 46 pp.
Chen, Y., F. Weng, Y. Han, and Q. Liu, 2008: Validation of the Community Radiative Transfer Model by using CloudSat data. J. Geophys. Res., 113, D00A03, https://doi.org/10.1029/2007JD009561.
Chen, Y., Y. Han, and F. Weng, 2012: Comparison of two transmittance algorithms in the Community Radiative Transfer Model: Application to AVHRR. J. Geophys. Res., 117, D06206, https://doi.org/10.1029/2011JD016656.
Chevallier, F., S. D. Michele, and A. McNally, 2006: Diverse profile datasets from the ECMWF 91-level short-range forecasts. NWP SAF Tech. Rep. 10, 14 pp., https://www.ecmwf.int/node/8685.
Clough, S. A., M. J. Iacono, and J.-L. Moncet, 1992: Line-by-line calculations of atmospheric fluxes and cooling rates: Application to water vapor. J. Geophys. Res., 97, 15 761–15 785, https://doi.org/10.1029/92JD01419.
Clough, S. A., M. W. Shephard, E. J. Mlawer, J. S. Delamere, M. J. Iacono, K. Cady-Pereira, S. Boukabara, and P. D. Brown, 2005: Atmospheric radiative transfer modeling: A summary of the AER codes. J. Quant. Spectrosc. Radiat. Transfer, 91, 233–244, https://doi.org/10.1016/j.jqsrt.2004.05.058.
Derber, J. C., and W.-S. Wu, 1998: The use of TOVS cloud-cleared radiances in the NCEP SSI analysis system. Mon. Wea. Rev., 126, 2287–2299, https://doi.org/10.1175/1520-0493(1998)126<2287:TUOTCC>2.0.CO;2.
DeSouza-Machado, S. G., L. L. Strow, S. E. Hannon, H. E. Motteler, M. Lopez-Puertas, B. Funke, and D. P. Edwards, 2007: Fast forward radiative transfer modeling of 4.3 μm nonlocal thermodynamic equilibrium effects for infrared temperature sounders. Geophys. Res. Lett., 34, L01802, https://doi.org/10.1029/2006GL026684.
Desroziers, G., L. Berre, B. Chapnik, and P. Poli, 2005: Diagnosis of observation, background and analysis-error statistics in observation space. Quart. J. Roy. Meteor. Soc., 131, 3385–3396, https://doi.org/10.1256/qj.05.108.
Diniz, F. L. R., and R. Todling, 2020: Assessing the impact of observations in a multi-year reanalysis. Quart. J. Roy. Meteor. Soc., 146, 724–747, https://doi.org/10.1002/qj.3705.
Ehrendorfer, M., R. M. Errico, and K. D. Raeder, 1999: Singular-vector perturbation growth in a primitive equation model with moist physics. J. Atmos. Sci., 56, 1627–1648, https://doi.org/10.1175/1520-0469(1999)056<1627:SVPGIA>2.0.CO;2.
Errico, R. M., R. Yang, N. C. Privé, K.-S. Tai, R. Todling, M. E. Sienkiewicz, and J. Guo, 2013: Development and validation of observing-system simulation experiments at NASA’s Global Modeling and Assimilation Office. Quart. J. Roy. Meteor. Soc., 139, 1162–1178, https://doi.org/10.1002/qj.2027.
Errico, R. M., and Coauthors, 2017: Description of the GMAO OSSE for weather analysis software package: Version 3. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, Vol. 48, 156 pp.
Errico, R. M., D. Carvalho, N. C. Privé, and M. Sienkiewicz, 2020: Simulation of atmospheric motion vectors for an observing system simulation experiment. J. Atmos. Oceanic Technol., 37, 489–505, https://doi.org/10.1175/JTECH-D-19-0079.1.
Gelaro, R., and Y. Zhu, 2009: Examination of observation impacts derived from observing system experiments (OSEs) and adjoint models. Tellus, 61A, 179–193, https://doi.org/10.1111/j.1600-0870.2008.00388.x.
Gelaro, R., and Coauthors, 2015: Evaluation of the 7-km GEOS-5 nature run. NASA Tech. Rep. Series on Global Modeling and Data Assimilation NASA/TM-2014-104606v36, Vol. 36, 305 pp.
Gelaro, R., and Coauthors, 2017: The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2). J. Climate, 30, 5419–5454, https://doi.org/10.1175/JCLI-D-16-0758.1.
Han, Y., P. van Delst, Q. Liu, F. Weng, B. Yan, R. Treadon, and J. Derber, 2006: Community Radiative Transfer Model (CRTM): Version 1. NOAA Tech. Rep. NESDIS 122, 122 pp.
Hoffman, R. N., and R. Atlas, 2016: Future observing system simulation experiments. Bull. Amer. Meteor. Soc., 97, 1601–1616, https://doi.org/10.1175/BAMS-D-15-00200.1.
Holdaway, D., R. Errico, R. Gelaro, and J. G. Kim, 2014: Inclusion of linearized moist physics in NASA’s Goddard Earth Observing System data assimilation tools. Mon. Wea. Rev., 142, 414–433, https://doi.org/10.1175/MWR-D-13-00193.1.
Irion, F. W., and Coauthors, 2018: Single-footprint retrievals of temperature, water vapor and cloud properties from AIRS. Atmos. Meas. Tech., 11, 971–995, https://doi.org/10.5194/amt-11-971-2018.
Jedlovec, G. J., and R. J. Atkinson, 1998: The Marshall automated wind algorithm: Error analysis, quality control and climate applications. Fourth Int. Winds Workshop, Saanenmöser, Switzerland, EUMETSAT.
Jones, E., C. D. Barnet, Y. Ma, K. Garrett, K. Ide, and S.-A. Boukabara, 2020: Efforts to evaluate shortwave observations from the CrIS hyperspectral infrared instrument in the NOAA Global Data Assimilation System. 24th Conf. on Integrated Observing and Assimilation Systems for the Atmosphere, Oceans, and Land Surface, Boston, MA, Amer. Meteor. Soc., 5A.6, https://ams.confex.com/ams/2020Annual/webprogram/Paper368688.html.
Key, J. R., D. Santek, C. S. Velden, N. Bormann, J.-N. Thepaut, L. P. Riishojgaard, Y. Zhu, and W. P. Menzel, 2003: Cloud-drift and water vapor winds in the polar regions from MODIS. IEEE Trans. Geosci. Remote Sens., 41, 482–492, https://doi.org/10.1109/TGRS.2002.808238.
Kleist, D. T., D. F. Parrish, J. C. Derber, R. Treadon, W.-S. Wu, and S. Lord, 2009: Introduction of the GSI into the NCEP Global Data Assimilation System. Wea. Forecasting, 24, 1691–1705, https://doi.org/10.1175/2009WAF2222201.1.
Langland, R. H., and N. L. Baker, 2004: Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus, 56A, 189–201, https://doi.org/10.3402/tellusa.v56i3.14413.
Le Marshall, J., N. Pescod, B. Seaman, G. Mills, and P. Stewart, 1994: An operational system for generating cloud drift winds in the Australian region and their impact on numerical weather prediction. Wea. Forecasting, 9, 361–370, https://doi.org/10.1175/1520-0434(1994)009<0361:AOSFGC>2.0.CO;2.
Maschhoff, K., J. Polizotti, H. Aumann, J. Susskind, D. Bowler, C. Gittins, M. Janelle, and S. Fingerman, 2019: Concept development and risk reduction for MISTiC winds, A micro-satellite constellation approach for vertically resolved wind and IR sounding observations in the troposphere. Remote Sens., 11, 2169, https://doi.org/10.3390/rs11182169.
Masters, D., and Coauthors, 2020: Status and accomplishments of the Spire Earth observing nanosatellite constellation. Proc. SPIE, 11530, 115300V, https://doi.org/10.1117/12.2574110.
McCarty, W., G. Jedlovec, and T. L. Miller, 2009: Impact of the assimilation of Atmospheric Infrared Sounder radiance measurements on shortterm weather forecasts. J. Geophys. Res., 114, D18122, https://doi.org/10.1029/2008JD011626.
McCarty, W., and Coauthors, 2016: MERRA-2 input observations: Summary and assessment. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, NASA TM-2016-104606, Vol. 46, 51 pp.
McNally, A. P., J. C. Derber, W. Wu, and B. B. Katz, 2000: The use of TOVS level-1b radiances in the NCEP SSI analysis system. Quart. J. Roy. Meteor. Soc., 126, 689–724, https://doi.org/10.1002/qj.49712656315.
McNally, A. P., P. D. Watts, J. A. Smith, R. Engelen, G. A. Kelly, J. N. Thépaut, and M. Matricardi, 2006: The assimilation of AIRS radiance data at ECMWF. Quart. J. Roy. Meteor. Soc., 132, 935–957, https://doi.org/10.1256/qj.04.171.
Molod, A., L. Takacs, M. Suarez, and J. Bacmeister, 2015: Development of the GEOS-5 atmospheric general circulation model: Evolution from MERRA to MERRA2. Geosci. Model Dev., 8, 1339–1356, https://doi.org/10.5194/gmd-8-1339-2015.
Nieman, S. J., W. P. Menzel, C. M. Hayden, D. Gray, S. T. Wanzong, C. S. Velden, and J. Daniels, 1997: Fully automated cloud-drift winds in NESDIS operations. Bull. Amer. Meteor. Soc., 78, 1121–1133, https://doi.org/10.1175/1520-0477(1997)078<1121:FACDWI>2.0.CO;2.
Peubey, C., and A. P. McNally, 2009: Characterization of the impact of geostationary clear-sky radiances on wind analyses in a 4D-Var context. Quart. J. Roy. Meteor. Soc., 135, 1863–1876, https://doi.org/10.1002/qj.500.
Posselt, D. J., and Coauthors, 2019: Quantitative assessment of state-dependent atmospheric motion vector uncertainties. J. Appl. Meteor. Climatol., 58, 2479–2495, https://doi.org/10.1175/JAMC-D-19-0166.1.
Privé, N. C., and R. M. Errico, 2013: The role of model and initial condition error in numerical weather forecasting investigated with an observing system simulation experiment. Tellus, 65A, 21740, https://doi.org/10.3402/tellusa.v65i0.21740.
Privé, N. C., and R. M. Errico, 2019: Uncertainty of observation impact estimation in an adjoint model investigated with an observing system simulation experiment. Mon. Wea. Rev., 147, 3191–3204, https://doi.org/10.1175/MWR-D-19-0097.1.
Privé, N. C., R. M. Errico, and K.-S. Tai, 2013a: Validation of the forecast skill of the Global Modeling and Assimilation Office observing system simulation experiment. Quart. J. Roy. Meteor. Soc., 139, 1354–1363, https://doi.org/10.1002/qj.2029.
Privé, N. C., R. M. Errico, and K.-S. Tai, 2013b: The influence of observation errors on analysis error and forecast skill investigated with an observing system simulation experiment. J. Geophys. Res. Atmos., 118, 5332–5346, https://doi.org/10.1002/jgrd.50452.
Privé, N. C., R. M. Errico, R. Todling, and A. El Akkraoui, 2021: Evaluation of adjoint-based observation impacts as a function of forecast length using an OSSE. Quart. J. Roy. Meteor. Soc., 147, 121–138, https://doi.org/10.1002/qj.3909.
Putman, W. M., and S.-J. Lin, 2007: Finite-volume transport on various cubed-sphere grids. J. Comput. Phys., 227, 55–78, https://doi.org/10.1016/j.jcp.2007.07.022.
Rienecker, M. M., and Coauthors, 2008: The GEOS-5 data assimilation system—Documentation of versions 5.0.1, 5.1.0, and 5.2.0. NASA Tech. Rep. Series on Global Modeling and Data Assimilation, Vol. 27, 109 pp.
Rossow, W. B., and R. A. Schiffer, 1991: ISCCP cloud data products. Bull. Amer. Meteor. Soc., 72, 2–20, https://doi.org/10.1175/1520-0477(1991)072<0002:ICDP>2.0.CO;2.
Ruf, C., and Coauthors, 2013: The NASA EV-2 Cyclone Global Navigation Satellite System (CYGNSS) mission. 2013 IEEE Aerospace Conf., Big Sky, MT, IEEE, https://doi.org/10.1109/AERO.2013.6497202.
Santek, D., S. Nebuda, and D. Stettner, 2019a: Demonstration and evaluation of 3D winds generated by tracking features in moisture and ozone fields derived from AIRS sounding retrievals. Remote Sens., 11, 2597, https://doi.org/10.3390/rs11222597.
Santek, D., and Coauthors, 2019b: 2018 atmospheric motion vector (AMV) intercomparison study. Remote Sens., 11, 2240, https://doi.org/10.3390/rs11192240.
Schmetz, J., K. Holmlund, J. Hoffman, B. Strauss, B. Mason, V. Gaertner, A. Koch, and L. Van De Berg, 1993: Operational cloud-motion winds from Meteosat infrared images. J. Appl. Meteor., 32, 1206–1225, https://doi.org/10.1175/1520-0450(1993)032<1206:OCMWFM>2.0.CO;2.
Susskind, J., C. Barnet, and J. Blaisdell, 1998: Determination of atmospheric and surface parameters from simulated AIRS/AMSU/HSB sounding data: Retrieval and cloud clearing methodology. Adv. Space Res., 21, 369–384, https://doi.org/10.1016/S0273-1177(97)00916-2.
Velden, C. S., and K. M. Bedka, 2009: Identifying the uncertainty in determining satellite-derived atmospheric motion vector height attribution. J. Appl. Meteor. Climatol., 48, 450–463, https://doi.org/10.1175/2008JAMC1957.1.
Velden, C. S., C. M. Hayden, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc., 78, 173–195, https://doi.org/10.1175/1520-0477(1997)078<0173:UTWDFG>2.0.CO;2.
Wu, W.-S., R. J. Purser, and D. F. Parrish, 2002: Three-dimensional variational analysis with spatially inhomogeneous covariances. Mon. Wea. Rev., 130, 2905–2916, https://doi.org/10.1175/1520-0493(2002)130<2905:TDVAWS>2.0.CO;2.
Wylie, D. P., and W. P. Menzel, 1999: Eight years of high cloud statistics using HIRS. J. Climate, 12, 170–184, https://doi.org/10.1175/1520-0442(1999)012<0170:EYOHCS>2.0.CO;2.
Wylie, D. P., E. Eloranta, J. D. Spinhirne, and S. P. Palm, 2007: A comparison of cloud cover statistics from the GLAS lidar with HIRS. J. Climate, 20, 4968–4981, https://doi.org/10.1175/JCLI4269.1.
Zhu, Y., and R. Gelaro, 2008: Observation sensitivity calculations using the adjoint of the Gridpoint Statistical Interpolation (GSI) analysis system. Mon. Wea. Rev., 136, 335–351, https://doi.org/10.1175/MWR3525.1.