Using the GEOS-5 Nature Run to Simulate 2053-nm Coherent Doppler Wind Lidar Observations
1. Introduction
Wind observations are a critical part of the current global observing system used by numerical weather prediction (NWP) centers around the world as wind observations provide improved initial dynamical conditions of 3D turbulence (e.g., Stoffelen et al. 2020), along with the transport of key variables in the atmosphere such as temperature and moisture. Additionally, Li et al. (2023) showed that wind observations proved more effective than temperature observations in constraining the temperature field in the tropics. In 2018, the Decadal Survey (NASEM 2018) highlighted the importance of missions focusing on 3D wind measurements including wind lidars, especially for observations in the planetary boundary layer (PBL). The first satellite-based wind lidar Aeolus ALADIN (Straume et al. 2020) has been used in operational forecasts at the ECMWF and has provided improved forecast skill (Rennie et al. 2021). A 2053-nm coherent detection lidar could be complementary to a direct detection 355-nm wind lidar such as ALADIN. A UV direct detection lidar provides measurements in regions of aerosol and cloud scatter, along with returns from molecular scattering. The 2053-nm radiation has weak molecular scattering which is unsuitable for a laser wind measurement. Instead, a 2053-nm lidar requires a scattering signal from an aerosol or cloud layer to retrieve 3D winds. Owing to a weaker molecular sensitivity, a 2053-nm lidar under cloud-free conditions could penetrate more deeply into an aerosol layer in the PBL, thus providing complementary information to a UV lidar.
An observing system simulation experiment (OSSE) can be used effectively to assess the potential impact of a new instrument such as a wind lidar (e.g., Atlas et al. 1985; Arnold and Dey 1986; Errico et al. 2013; Hoffman and Atlas 2016). Realistic performance can be assessed using an extensively validated framework such as the global OSSE framework developed at the NASA Global Modeling and Assimilation Office (GMAO) (Errico et al. 2013; Privé et al. 2021). An OSSE consists of using a free-running global model referred to as a nature run (NR) that replaces the real atmosphere, simulated observations generated from the NR that take the place of real observations, along with a data assimilation system (DAS) and associated NWP model to conduct experiments. The nature run used in this study is 2-yr-long free-running forecast of the Goddard Earth Observing System Model, version 5 (GEOS-5) (Rienecker et al. 2008), that is often referred to as the G5NR.
OSSEs highlighting the potential benefit from a wind lidar have been explored extensively (e.g., Atlas and Emmitt 2008; Atlas et al. 2015; Cardinali et al. 1998; Hoffman et al. 1990; Marseille et al. 2001; Stoffelen et al. 2006; Masutani et al. 2010; Okamoto et al. 2018). The treatment of aerosol scattering in these studies is not typically from an aerosol tracer in the nature run but rather a function of relative humidity (e.g., Atlas et al. 2015). In this study, dust, sea salt, black carbon, organic carbon, and sulfate aerosol tracers available in the G5NR are used to compute backscatter. Realistic aerosol sources are included in the G5NR. Sources include wind-dependent emissions of dust and sea salt, anthropogenic emissions, biogenic emissions, and volcanic sources. Sinks and other interactions are also considered including wet and dry deposition, scavenging from precipitation, along with hygroscopic particle growth for hydrophilic particles (Gelaro et al. 2015). While a full optical model of the lidar return is not used, a dataset from Bedka et al. (2021) is used to derive an empirical relationship based on backscatter to determine whether there is sufficient scattering to provide a wind measurement and thus sample the nature run’s wind field.
The goal of this paper is to describe the lidar observation generation system developed for an OSSE along with some preliminary analysis impacts highlighting the utility of a 2053-nm coherent wind lidar. The paper is organized into four additional sections. First, in section 2, the dataset and methods used to derive a relationship between aerosol backscatter and wind measurements are presented, along with strategies for generating simulated wind observations sampling aerosol and cloud fields from the G5NR. Next, in section 3, the distribution of observations generated and some initial analysis impact results are presented and discussed. Finally, in section 4, the results and implications for a future OSSE using the generated observations are discussed, along with concluding remarks. For the reader’s convenience, a list of acronyms and abbreviations has been provided in the appendix.
2. Comparison of aerosol fields in GEOS-FP to available aircraft observations to develop a wind lidar simulator
With the goal of defining a strategy to develop a simulator that uses backscatter computed from aerosol and cloud fields in the G5NR, aircraft measurements from the Doppler Aerosol Wind (DAWN; Kavaya et al. 2014) lidar and computed backscatter from the GEOS Forward Processing (GEOS-FP) system are compared. GEOS-FP system provides forecasts along with an analysis using the most current GEOS atmospheric DAS (GEOS-ADAS) that utilizes observations including existing conventional radiosondes, aircraft, and satellite observations. GEOS-FP products are primarily used for real-time support for NASA field campaigns, support for NASA science, and interaction and comparison with other organizations. In contrast to GEOS-FP, the G5NR itself cannot be used in direct comparison to the real world as it does not represent any particular realized atmospheric state in the past or future. The G5NR is used in an OSSE context to provide “truth” from which observations are simulated. Instead, GEOS-FP provides a near-real-time analysis including meteorological, cloud fraction, and aerosol fields similar to those in the G5NR. The goal is to use DAWN measurements in conjunction with the GEOS-FP analysis to characterize the cloud and aerosol scattering conditions while creating a simulator that determines the distribution of wind measurements from a DAWN-like instrument. This simulator can then be applied to the G5NR to simulate adding lidar observations to the global observing network within the GMAO OSSE framework.
The DAWN measurement flights aboard a DC-8 aircraft were largely conducted over the eastern Pacific with one flight over the southwestern United States. The DAWN can scan at programmable azimuth angles combined with the number of laser pulse averages per profile. The nominal operating mode employs five azimuth angles (45°, 22.5°, 0°, −22.5°, and −45°) at a zenith angle of 30°, with 0° oriented forward along the flight track. The flight track is not always parallel to the longitudinal axis due to crosswind. The nominal operating mode also uses 20 pulses per azimuth, providing wind profiles along the track every 4–5 km, assuming nominal DC-8 cruise speeds of 225–250 m s−1. Further details about the flight campaign and dataset are available in Bedka et al. (2021). A spaceborne version of this instrument would likely only have two azimuth views. In the simulations that follow, we consider one azimuth view.
Backscatter is computed based on aerosol types available in the GEOS and following Castellanos et al. (2019). An example of a day of aerosol distributions in the G5NR is shown in Fig. 1. A large orange region of dust off the west coast of Africa can be seen extending to the eastern United States, along with green regions of biomass burning toward the center of South America extending in the opposite direction. Several regions of light blue show windblown sea salt, and white regions represent anthropogenic sulfate emissions over China. Details of the backscatter calculation are thoroughly explained in Castellanos et al. (2019); however, a brief summary of the assumptions is of interest for this study. The size distributions for sulfate, black carbon (BC), and organic carbon (OC) are lognormal with mode radii of 0.0695, 0.0212, and 0.0118 μm, respectively. The geometric standard deviations for the distributions are 2.03 μm for sulfate and 2.0 μm for BC and OC. Dust and sea salt are tracked with five different size bins. The five size bins for dust are 0.1–1.0, 1.0–1.8, 1.8–3.0, 3.0–6.0, and 6.0–10.0 μm. The size bins for sea salt are 0.03–0.1, 0.1–0.5, 0.5–1.5, 1.5–5.0, and 5.0–10.0 μm. For dust and hydrophobic BC and OC, no growth factors are applied as a function of humidity. For sulfate, along with hydrophilic BC and OC, growth factors based on humidity are taken from Köepke et al. (1997). For sea salt, growth factors are taken from Gong et al. (1997). With the exception of dust, optical properties are computed assuming Mie theory using code following Wiscombe (1980) and taking refractive indices from Hess et al. (1998) at a wavelength of 2.0 μm. Dust optical properties are computed following Colarco et al. (2014), which uses nonspherical properties from Meng et al. (2010) with spectral refractive indices from Shettle and Fenn (1979) at 2.0 μm. All details regarding the optical properties used are available from netCDF-3 files posted at the NASA Center for Climate Simulation data share (https://portal.nccs.nasa.gov/datashare/iesa/aerosol/AerosolOptics/). For reference, the values for 0% relative humidity aerosol refractive index are provided in Table 1.
Global spatial distribution of aerosols in the G5NR second year 0000 UTC 1 Jul. Orange regions represent dust; green regions represent black and organic carbon, white regions represent sulfate aerosols, and light blue regions represent sea salt.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Summary of real and imaginary parts of aerosol refractive index at 0% relative humidity at 2.0-μm wavelength along with optical property filename.
For each flight line from Bedka et al. (2021), aerosol properties were interpolated from GEOS-FP to each DAWN observation, and aerosol backscatter was computed for each DAWN observation location. Figures 2a, 2b, 3a, and 3b show the measured signal-to-noise ratio of DAWN, along with the computed backscatter interpolated to each valid DAWN observation for flights starting on 17 and 29 April 2019. The blue regions represent areas where the signal-to-noise ratio was either too low for a wind retrieval, the lidar was not operating (represented by large gaps between columns), or above the aircraft as measurements were only taken below. The blue regions at approximately 2 km represent regions where aerosol loading was insufficient for a retrieval to take place. The blue regions near the surface represent where either a thick aerosol or cloud is present aloft. Comparing the SNR measured from DAWN in Figs. 2a and 3a to the backscatter computed from GEOS-FP in Figs. 2b and 3b, there is a reasonable agreement spatially with high SNR values associated with large values of backscatter considering the DAWN observations are meter scale and the analysis is a 12.5-km horizontal resolution. Regions of high backscatter generally align with regions of high SNR in a spatiotemporal sense, as do regions of low backscatter and low SNR.
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 1 × 10−6 and 1 × 10−2 km−1 sr−1 for the flight starting 17 Apr 2019. Red bars in (d) represent observation counts, while orange bars represent simulated observation counts.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 1 × 10−6 and 1 × 10−2 km−1 sr−1 for the flight starting 29 Apr 2019.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
While the simulated backscatter from GEOS-FP and SNR values measured by DAWN agrees fairly well, there are a number of uncertainties that can be addressed by identifying possible cases to simulate in an OSSE context within this dataset. It is beyond the scope of this work to simulate SNR based upon optical properties calculated from the G5NR. This could be done following the fundamental SNR calculations (Frehlich and Kavaya 1991; Baron et al. 2017) or using an approximation in Baron et al. (2017). Instead, calculated backscatter in regions where a DAWN retrieval takes place is used as a proxy for whether or not a wind observation will take place. As there is some uncertainty in this estimate, two cases are considered with different detection limits based upon aerosol scattering: the first with a larger range of backscatter (Case A) and the second with a more restrictive detection limit range (Case B). While the dataset available is large, it does not cover a wide range of cloud types; therefore, it is not used to constrain the simulation of clouds. Any cloud simulation assumption developed for the OSSE simulator will be uncertain; thus, a second case is developed as a bounding case. The cases developed in this study are summarized in Table 2.
Summary of cases developed for OSSE.
In Fig. 4a, 2D histogram shows the range of measured SNR versus calculated backscatter using GEOS-FP at locations where a DAWN wind retrieval takes place (and has a sufficient SNR for a successful retrieval), filtering out regions where the GEOS-FP analysis points have a nonzero cloud fraction (green regions in Figs. 2b and 3b). From Fig. 4, an empirical relationship between SNR and computed backscatter is not readily apparent aside from a minimum and maximum value from the distribution. The smaller cluster around 10−3 km−1 sr−1 in Fig. 4b, which is spread out widely in SNR is likely a mismatch between the analysis aerosol loading and what the actual conditions are, presuming the SNR is related to backscatter measured by the lidar. This relationship is more clearly shown in Fig. 5. The range of backscatter values generally lies between 1 × 10−6 and 1 × 10−2 km−1 sr−1, and while there are some points that extend below 1 × 10−6 km−1 sr−1, the bulk of the distribution lies in this range. For clarity, the distribution of backscatter is shown in a 1D histogram that more clearly shows why the range of 1 × 10−6 and 1 × 10−2 km−1 sr−1 was chosen as the nominal case (Case A). The values for backscatter are somewhat larger than those presented in Srivastava et al. (2001) and could be a result of different aerosol sources present at differing times or locations. A better comparison may be possible by transforming either the measured DAWN SNR to backscatter or the GEOS-FP backscatter to SNR using fundamental SNR equations or approximations (e.g., Frehlich and Kavaya 1991; Baron et al. 2017).
Histograms (2D) of calculated backscatter (binned in log space) against measured DAWN SNR for two flights. (a) 17–18 Apr 2019 and (b) 29–30 Apr 2019.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Histograms of backscatter (binned in log space) calculated from GEOS-FP in regions where a DAWN observation occurs against DAWN observation counts for two flights. (a) 17–18 Apr 2019 and (b) 29–30 Apr 2019.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
While the DAWN data are used as a guide to develop a sampling procedure that determines whether or not a wind retrieval takes place on a given level based on aerosol backscatter, developing criteria for cloudy regions requires some form of statistical sampling based upon cloud fraction in the G5NR as the G5NR is not a cloud-resolving model. Similar assumptions are necessary for the simulation of atmospheric motion vector wind retrievals that utilize cloud information (Errico et al. 2020). For each profile, every sampled level undergoes a series of criteria to determine if a retrieval takes place. This procedure is outlined in Fig. 6 and described as follows.
Flowchart describing wind lidar simulator/data sampler.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
First, for nominal Case A, if the cloud fraction is less than 5% and the aerosol backscatter coefficient is between 1 × 10−6 and 1 × 10−2 km−1 sr−1, a wind retrieval is considered clear and has sufficient aerosol for a retrieval (see Figs. 2 and 3) and continues to check the adjacent level below. As a bounding case (Case B), these thresholds are tightened to reduce observation counts to between 5 × 10−6 and 1 × 10−3 km−1 sr−1. This tighter limit was chosen to reduce the overestimate of observations simulated from the analysis compared with the number of measured observations shown in Figs. 2d and 3d.
Next, a check for thin cirrus clouds is performed. A random number generator with a possible range of values from 0 to 1 combined with the cloud fraction is used to sample if a cloud is detected. If the random number is less than the cloud fraction, then a cloud is detected. If the ice density of the cloud is less than 0.05 g m−3, it is considered a thin ice cloud, provides a wind retrieval, and continues to check lower levels.
In addition to the thin cirrus cloud test, a persistent opaque cloud test is performed. Again, the cloud fraction is combined with a random number generator to provide an element of probabilistic sampling, along with checking the liquid and ice density of the cloud. If the ice or liquid density is greater than 0.05 g m−3, an opaque cloud is detected using a maximum overlap assumption. For the purposes of this study, we consider two options for considering cloud returns. In the first option (Cases A and B), if an opaque cloud is detected, neighboring cells along the orbit track are checked for large cloud fractions (greater than 0.95); if this cloud fraction is exceeded in neighboring cells, a persistent cloud layer is detected, and no further retrieval is performed on lower levels in the profile. This is to simulate cases in which the lidar could retrieve under an opaque cloud in a fore/aft view while eliminating cases with a cloud deck that extends over a wide area. Alternatively, as a second option (Case C), once a cloud return is determined to be opaque, there is only a wind retrieval at the top of the cloud regardless of what the instrument may be able to retrieve from another fore or aft view below the cloud. A similar approach is used for persistently thick aerosols; if the backscatter coefficient consistently reaches the maximum threshold of 1 × 10−2 km−1 sr−1 among its neighbors, it is considered opaque, and no further retrievals are performed in the profile (adjusted to 10−3 km−1 sr−1 for Case B). No secondary option was developed for thick aerosols.
Finally, a check for cases where the aerosol loading is too thin is performed. If the aerosol backscatter coefficient is at the limit of 1 × 10−6 km−1 sr−1 (adjusted to 5 × 10−6 km−1 sr−1 for Case B) for a given level along with its neighbors, it is assumed there will be an insufficient aerosol loading for a wind retrieval, and no winds are retrieved at that level while continuing to check layers below. The checking of neighbors for thin aerosol loading was done to simulate the clusters of low backscatter which can result in no retrieval (see Figs. 2 and 3), while keeping cases where the instrument can compensate by adjusting the range gate.
The lidar simulator is applied to the two flights in Figs. 2 and 3 for the nominal Case A. Additionally, perturbation Case B is shown in Figs. 7 and 8, along with Case C shown in Figs. 9 and 10. Panels c and d in Figs. 2, 3, and 7–10 show the results of the lidar simulator’s observation classification and observation count distribution. Since there are relatively few clouds present for these flights, the probabilistic cloud assumption is ignored, and clouds are assumed present as otherwise determined by the procedure in Fig. 6. The observation distribution in Figs. 2d and 3d for each flight agree well above 2 km. Below 2 km, the simulator overestimates the observation count based on the GEOS-FP analysis used as input to the simulator. This could be due to either an overestimate of aerosol backscatter or a small-scale cloud where the analysis would have no indication of cloudy conditions resulting in fewer counts at lower altitudes. Applying the stricter aerosol criteria for Case B, the observation classifications are shown in panels c and d in Figs. 7 and 8. The observation distribution between the simulator and measured observation count agrees quite well on 29 April (Fig. 8d) but underestimates the observation count on 17 April (Fig. 7d). The stricter criterion results in fewer thin aerosol returns aloft (light purple in Figs. 7c and 8c) and fewer aerosol returns below areas of heavy aerosol loading (light orange in Figs. 7c and 8c). Applying the more restrictive cloud perturbation in Case C, fewer observations are selected from the analysis below 2 km in Figs. 9d and 10d than in Case A (Figs. 2d and 3d). This reduction in simulation counts appears to agree better below 2 km for Case C than for Case A.
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 5 × 10−6 and 1 × 10−3 km−1 sr−1 for the flight starting 17 Apr 2019.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 5 × 10−6 and 1 × 10−3 km−1 sr−1 for the flight starting 29 Apr 2019.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 1 × 10−6 and 1 × 10−2 km−1 sr−1 for the flight starting 17 Apr 2019. This assumes a stricter cloud return algorithm where no returns are possible below a thick cloud.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
(a) Plot of DAWN SNR at locations where wind retrieval product is available, (b) simulated backscatter at DAWN observation locations, (c) lidar return classification based upon GEOS-FP analysis, and (d) number of retrievals from the DAWN measurement and based upon GEOS-FP and classification algorithm. This assumes nominal backscatter thresholds of 1 × 10−6 and 1 × 10−2 km−1 sr−1 for the flight starting 29 Apr 2019. This assumes a stricter cloud return algorithm where no returns are possible below a thick cloud.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
The DAWN dataset was primarily used to guide the development of a lidar sampler based on aerosol backscatter; however, it is of interest to briefly discuss how DAWN wind observations compare with GEOS-FP analysis. In Figs. 11 and 12, scatterplots of DAWN wind observations are plotted against GEOS-FP analysis data points interpolated to the DAWN observation location. In Fig. 11, there is strong correlation for both wind components, with a Pearson correlation coefficient of 0.9592 for the zonal component and a value of 0.9655 for the meridional component. There is some scatter which is to be expected given the spatial scale of the DAWN observation (meters) versus the resolution of the GEOS-FP analysis which is C720 (12 km). In Fig. 12, the correlation for both wind components is less strong, with a Pearson correlation coefficient of 0.7973 for the zonal component and a value of 0.8040 for the meridional component. There is some scatter which is to be expected given the spatial scale of the DAWN observation (meters) versus the resolution of the GEOS-FP analysis, which is C720 (12 km). There is a small cluster below the main grouping of observations in the zonal component, but this is not surprising given small-scale features will be resolved by the lidar versus the relatively coarse resolution of the analysis.
Scatterplot of measured DAWN zonal (u) and meridional (υ) wind components vs GEOS-FP analysis interpolated to DAWN observation locations on 17–18 Apr flight. The black line represents the 1:1 correlation line, and R represents the Pearson correlation coefficient. (a) The zonal component and (b) the meridional component.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Scatterplot of measured DAWN zonal (u) and meridional (υ) wind components vs GEOS-FP analysis interpolated to DAWN observation locations on 29–30 Apr 2019 flight. The black line represents the 1:1 correlation line, and R represents the Pearson correlation coefficient. (a) The zonal component and (b) the meridional component.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
While the DAWN data provided some guidance to develop a method to sample winds using cloud parameters and aerosol properties from the GEOS model, there are a range of parameterizations that fit the data. Additionally, there is some uncertainty as to how a spaceborne lidar with a design similar to DAWN would perform on orbit. Thus, the bounding configurations for the simulator (Cases A–C) that have been developed based on the information available are judged to be reasonable for the simulation of data from a spaceborne lidar. In the next section, simulated data are used to perform some preliminary OSSEs.
3. Distribution of generated wind observations and initial analysis impacts
The spatiotemporal distribution of a new observation type in a numerical weather prediction context plays an important role in determining the observation impact. For some OSSE studies, a similar observation type such as an existing atmospheric motion vector, radiance, or GPS-radio occultation is used as a guide for a probabilistic generation of new observations (e.g., Errico et al. 2020; McCarty et al. 2021; Privé et al. 2022). Unfortunately, in the case of the 2053-nm lidar, there is no similar global observation to constrain the distribution of observations. While there is the Aeolus ALADIN UV lidar, the optical properties of aerosols and clouds at 2053 nm would be quite different and may not be realistically simulated from Aeolus. Wu et al. (2013) developed climatology at 2.1 μm using Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP) data at 550 nm and assessed the performance of a novel Doppler wind lidar but has built-in assumptions about aerosol sources and spectral aerosol properties. Instead, as outlined in section 2, the cloud fraction and aerosol distributions from the G5NR and a probabilistic assumption about cloud fraction, along with aircraft measurements to guide aerosol loading, are used as a way to constrain the distribution of wind observations. The assumption of a realistic cloud distribution is the most uncertain aspect as the G5NR is not a cloud-resolving model, and the performance of a similar instrument from space is somewhat uncertain and in part tested with Case C. Prior work simulating spaceborne lidar from a noncloud-resolving nature run (Marseille and Stoffelen 2003) was found to give realistic results. The G5NR has strongly bimodal cloud fractions, and in comparison with CERES, cloud observations have found that while cloud fractions are generally realistic, there are some regional discrepancies in the prevalence of both high and low cloud fractions (Gelaro et al. 2015). Hence, the main uncertainty related to clouds is that of adjusting the probabilistic methods to produce realistic observation distributions in the absence of a targeting dataset.
While there is some reasonable constraint applied to aerosol loading, using the DAWN observation counts as a guide, this is also somewhat uncertain and tested with Case B. In addition to the simulations against DAWN data for Cases A–C, we simulate observations using the G5NR and an assumed satellite orbit. Observations from the 2053-nm lidar are generated following the procedure outlined in section 2 and added to the GMAO NWP OSSE framework. The observations are sampled from the G5NR on its native grid (7 km and 72 levels). The observations are thinned using a 25-km box horizontally and 10 hPa in the vertical.
The GMAO OSSE version used here is described in Privé et al. (2023) and includes simulated observations to represent the global NWP observing system circa 2020 along with the forecast model and DAS used quasi-operationally at GMAO in 2022–2023. The GEOS-5 nature run is a 2-yr-long forecast of a version of the GEOS model (Rienecker et al. 2008) similar to that used for the MERRA-2 reanalysis at 7-km horizontal resolution (C1440) with 72 vertical levels and 30-min temporal output. The time period used for all OSSE results in this section is July of year 2 of the G5NR.
The simulated global observing network in the OSSE includes most of the data types that were used quasi-operationally in 2020, including microwave and infrared radiances, conventional observations, GPS radio occultation (GPS-RO), atmospheric motion vectors, and scatterometer winds. Compared with the previous version of the GMAO OSSE, this updated version includes the simulation of all-sky GMI, AMSR-2, and microwave humidity sounder (MHS) radiances, and the experiment forward model has more substantial differences in physics and dynamics compared to the G5NR, reducing the “twinning” effect. Simulated observation errors are added to the simulated data types (Privé et al. 2021) to account for instrument error, representativeness error, and operator error. These observation errors include both random uncorrelated errors and errors that may be correlated vertically, horizontally, or between channels depending on the observation characteristics. The simulated observations and their errors are carefully calibrated and validated to match as well as possible the statistical performance of real observations.
A control run for the OSSE is produced using the hybrid 4D-ensemble variational Gridpoint Statistical Interpolation analysis system (GSI) DAS with a 32-member ensemble. The control is spun up during the month of June in the second year of the G5NR with the June period used for calibration and as an adjustment period for the radiance bias correction. Only those data types used in 2020 were simulated and ingested in the control run. Additional OSSE tests including the lidar observations are run in stand-alone mode, where only the DAS is performed and not the forward model as shown in Fig. 13. The background state for each cycle time is taken from the OSSE control run, and the DAS ingests a different set of simulated observations, producing only an analysis field as output. This model is computationally inexpensive and also provides analyses that can be compared with the control analyses to isolate precisely the effects of new observing types without contributions from model feedback or general chaotic error growth. Because there is no accumulation of information between cycle times in stand-alone mode, the observation impacts tend to be of lower magnitude than in regular DAS cycling mode, with wind observations having roughly half the analysis impact in stand-alone. However, the spatial distribution of observation impacts in stand-alone is similar to that in the full cycling mode, just with a smaller magnitude.
Diagram illustrating the difference between the regular cycling mode of the data assimilation and model used in the (top) OSSE control and the (bottom) experiment stand-alone mode.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
The G5NR includes three-dimensional fields of five different aerosols (dust, sea salt, black carbon, organic carbon, and sulfate) that can be used for simulating the lidar wind observations in addition to clouds. The aerosols within the G5NR are well validated and realistic. Vertical and horizontal distributions of aerosols were well validated against reanalysis products and observational data from the CALIOP mission (Gelaro et al. 2015).
The G5NR is sampled using aerosol, cloud, and wind fields present as outlined in section 2. A single footprint at a 30° scan angle is used to sample the G5NR. The orbit is a sun-synchronous orbit with a local time of ascending node (LTAN) of 1330 taken from McCarty et al. (2021). A sample day of orbit points is shown in Fig. 14 color coded by their data assimilation cycle time. While a single point is used, it may be considered as a retrieval using two look angles from forward and aftward views. The fields are interpolated along the orbit track using the nearest profile in the 7-km G5NR for aerosol and cloud fraction, along with the zonal (u) and meridional (υ) wind components. The backscatter coefficient is calculated at 2053 nm using a lookup table used by the GEOS-ESM that includes an external mixture (each component is of a single composition/complex refractive index) of dust, sea salt, black carbon, organic carbon, and sulfates. The calculation of optical properties directly from aerosol fields present in the G5NR is an improvement from previous studies such as Atlas et al. (2015) which relied upon an assumed function of relative humidity. Interpolated fields for cloud include cloud fraction, mass fraction of liquid water, mass fraction of ice, and mass fraction of snow.
Observations sampled along orbit for four data assimilation cycles for 1 Jul.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
In an OSSE, observations are typically generated and ingested by the DAS in one of two ways. First, the observations might be used without adding simulated observation error–sometimes this is referred to as the “perfect” observation case and can be considered an evaluation of the ideal information content of the observations. The second method is to add a realistic simulated error to the simulated observations. For this initial work, no simulated error is added to the lidar observations, in order to avoid complications given the unknown error characteristics of real spaceborne 2-μm lidar. Observation impacts are strongly affected by the combination of the observation error assumed by the DAS and the actual observation error characteristics, with optimal impacts when the assumed error matches the actual error. In this work, the DAS-assumed lidar observation errors are tested over a range of magnitudes. If simulated observation errors were added to bring the total lidar observation error close to the DAS-assumed values, the observation impacts would be close to, but slightly weaker than, the impacts for the case tested here without added simulated lidar errors. Initially, for weighting purposes, the observation error assumed by the DAS is taken to be that of a rawinsonde which has a standard deviation slightly above 2 m s−1 aloft of 600 hPa and less than 2 m s−1 elsewhere. This assumed error is the black line shown in Fig. 15. This error is considered a combination of the small measurement error along with a more dominant representativeness error or the inability of the DAS to represent reality (interpolation, simulation errors, etc.). The DAS uses the assumed error to determine how heavily to weigh each observation type in comparison with the background field, where a higher assumed error results in a weaker weighting. The assumed error also plays a role in the quality control process used to reject observations with gross errors. This is considered a first step, and an OSSE with simulated lidar error will be included in a future study.
Assumed lidar DAS observation error standard deviation used by the GSI in this study.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Figure 16a shows the zonal distribution of assimilated lidar wind observations averaged per analysis cycle for Case A with Fig. 16b showing the same for Case B, the difference between the two in Fig. 16c, and the ratio of the difference between Cases A and B normalized by Case A in Fig. 16d. Overall, Case A has a larger number of observations distributed globally above 900 hPa at all latitudes relative to Case B, with greatest differences in polar regions and at high altitudes in the tropics. These differences are from the stricter minimum aerosol backscatter criteria imposed for Case B. Near the surface, there are also fewer observations owing to the stricter maximum aerosol threshold for Case B.
Zonal distribution of observations per cycle for Cases A and B. (a) The zonal distribution of observation counts for Case A, (b) the zonal distribution of observation counts for Case B, (c) the difference in counts between Cases A and B, and (d) the normalized difference between Cases A and B normalized by Case A.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Figure 17a shows the zonal mean distribution of assimilated lidar wind observations averaged per analysis cycle for Case A with Fig. 17b showing the same for Case C, the difference between the two cases in Fig. 17c, and the ratio of the difference between Cases A and C normalized by Case A in Fig. 17d. Overall, Case A has a slightly larger number of observations distributed globally below 750 hPa at all latitudes relative to Case C, with greatest differences between 30° and 80° in both the Northern and Southern Hemisphere. This is logical as the observation generation for Case A assumes that when an opaque cloud is not surrounded with areas of high cloud fraction, observations may be sampled deeper into the atmosphere below the detected cloud. In all cases, there tend to be more observations in midlatitudes in the Northern Hemisphere versus the Southern Hemisphere. This could be due to more aerosols being present in the Northern Hemisphere (e.g., dust, anthropogenic emissions) or due to an increase in cloud cover during July (e.g., convective clouds in the summer hemisphere).
Zonal distribution of observations per cycle for Cases A and C. (a) The zonal distribution of observation counts for Case A, (b) the zonal distribution of observation counts for Case C, (c) the difference in counts between Cases A and C, and (d) the normalized difference between Cases A and C normalized by Case A.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
When generating a new observation type for an OSSE, it is sometimes helpful to generate a single assimilation cycle of observations on a global grid to evaluate the spatial distribution of observations. This allows cloud and aerosol features to appear, giving spatial context to observation counts in lieu of a single line of observations from the orbiting lidar. Figure 18 shows a single data assimilation cycle for 0000 UTC 1 July. Figures 18a and 18b show the percent of the 42 vertical model levels sampled on a 2° grid for Cases A and B, with Fig. 18d showing the percent difference between Cases A and B. It should be noted that while there are 72 levels in the model fields, there are only 42 levels at a maximum which contain aerosol or cloud. Figure 18c shows the percent of the 42 levels sampled on a 2° grid for Case C, with Fig. 18e showing the percent difference between Cases A and C. Comparing Figs. 18a and 18c, the effect of clouds is clearly present in the Southern Hemisphere where heavy thick clouds are present during austral winter, reducing observation counts. Comparing Figs. 18a and 18b, the effect of stricter backscatter limits under Case B conditions can be seen globally although the percent of sampled observations in the Southern Hemisphere is substantially reduced. In Figs. 18c and 18e, the effect of using a stricter cloud assumption in Case C is seen (if a cloud is thick no further returns are allowed in the profile), with far fewer observation counts where cloud-like features appear.
Percent of sampled levels for all cases using a 2° grid in place of orbit for a single cycle at 0000 UTC 1 Jul. (a) The percent of levels sampled for Case A, (b) the percent of levels sampled for Case B, (c) the percent of levels sampled for Case C, (d) the normalized percent difference between Case A and B normalized by Case A, and (e) the normalized percent difference between Cases A and C normalized by Case A.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Positive fractional improvements indicate degradation or an increase in analysis error compared to the control, whereas negative values indicate improvement. Analysis states are used for the calculations at 0000, 0600, 1200, and 1800 UTC. The DAS is run in stand-alone mode for Cases A, B, and C for the month of July in the second year of the G5NR.
Figure 19a shows the zonal mean ϵ for Case A zonal wind, and Fig. 19b shows the same for Case B, with Fig. 19c showing the difference in ϵ between Cases A and B. Figures 19d–f mirror Figs. 19a–c replacing meridional wind for zonal wind. The areas of weaker analysis impact in Case B are largely collocated with regions of fewer observations in Case B owing to the change in the backscatter threshold for a wind observation to be simulated. Again, these changes are typically at higher altitudes, and an analysis improvement in the tropics above 200 hPa has been eliminated in Case B. This is due to imposing a higher detection limit on backscatter in the simulator for Case B. Overall, both cases show an improvement mostly distributed in the tropics and polar regions. This stems from heavier aerosol loading in the tropics, along with an orbit configuration that has a relatively large number of observations concentrated in polar regions. In fact, there are so many observations at the South Pole that a degradation appears relative to the control for both cases. The degradation from having too many observations at the poles could stem from representativeness error, along with possibly conflicting information in terms of interpolation of direction near the poles, along with the DAS and sampled G5NR having different spatial resolutions.
Zonal RMS analysis error ϵ for Cases A and B for the zonal (u) and meridional (υ) wind components. (a) The ϵ for Case A zonal wind, (b) ϵ for Case B zonal wind, (c) the difference in ϵ for zonal wind between Cases A and B, (d) ϵ for Case A meridional wind, (e) ϵ for Case B meridional wind, and (f) the difference in ϵ for meridional wind between Cases A and B. Note that the difference in (c) and (f) has the same limits as other panels.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Figure 20a shows the zonal mean ϵ for Case A zonal wind, and Fig. 20b shows the same for Case C, with Fig. 20c showing the difference in ϵ between Cases A and C. Figures 20d–f mirror Figs. 20a–c replacing meridional wind for zonal wind. Given the relatively small change in observation counts between the two cases, similar analysis impacts should be expected. While there is a small degradation of analysis impact between 50° and 80° latitude in each hemisphere, the analysis impact between the two cases is fairly similar. The differences between the two cases are extremely small.
Zonal RMS analysis error (ϵ) for Cases A and C for the zonal (u) and meridional (υ) wind components. (a) The ϵ for Case A zonal wind, (b) ϵ for Case C zonal wind, (c) the difference in ϵ for zonal wind between Cases A and C, (d) ϵ for Case A meridional wind, (e) ϵ for Case C meridional wind, and (f) the difference in ϵ for meridional wind between Cases A and C. Note that the limits on the color scale of (c) and (f) are 10 times less than other panels.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Another aspect of an OSSE that requires careful examination is the observation error used to weigh the observations in the DAS. The observation error used in experiments up to this point has been that of a rawinsonde. To test the effect of observation weighting, cases were run under the same assumptions in Case A, but multiplying the assumed DAS observation error by a factor of 0.75, 1.5, 2.0, 4.0, and 8.0 (see Fig. 15). The global areal means of fractional analysis error for an upper (∼192 hPa, in blue), middle (∼500 hPa, in orange), and lower (∼992.5 hPa, in green) level are plotted in Fig. 21. While there is variation between levels and meridional and zonal winds, there is a somewhat counterintuitive trend where a decreased weighting (increased error) results in an improvement (lower fractional analysis error). For factors greater than 2.0, there is a more intuitive response of weaker beneficial lidar impact as observation error is increased. The reason for the reduced fractional error with a decrease in weighting using a factor in the range of 0.75–2.0 is that the true representativeness error of the observation lies somewhere between 2 and 4 times the observation error used for Cases A–C. This representativeness error may include errors from temporal and spatial interpolation and also errors from the difference in resolution between the 7-km G5NR and the approximately 50-km DAS.
Mean fractional analysis error for different assumed observation errors within the GEOS-ADAS (using a factor 0.75, 1.0, 1.5, 2.0, 4.0, and 8.0 of the nominal assumed error).
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
Another aspect of the representativeness error that requires consideration is the error correlation between observations. While the observations here have been thinned to reduce error correlation as in McCarty et al. (2021), there may be residual spatial correlation between observations. Figure 22 shows the horizontal correlation of observation innovations for the Case A lidar dataset calculated over the month of July. The correlations seen in Fig. 22 are a combination of correlations from the representativeness error of the simulated lidar and from the correlation of the background field. The length scale at which correlations of O-F drop to 0.5 is approximately 85 km. The optimal use of these observations in a real-world NWP system would require an analysis of spatial correlation to either optimize thinning and inflation of observation error or combine the observations to form a so-called super observation. Doing so would help the DAS best capture the 4D state providing better initial conditions for the model.
Horizontal correlations of O-F for Case A lidar observations, calculated over the month of July. Solid black line with triangles, 1000 hPa; dash–dotted red line with circles, 500 hPa; and dotted blue line with stars, 200 hPa.
Citation: Journal of Atmospheric and Oceanic Technology 41, 7; 10.1175/JTECH-D-23-0117.1
4. Summary and conclusions
Wind observations are a critical part of the current global observing system used by NWP centers around the world as wind observations provide improved initial dynamical conditions of 3D turbulence (e.g., Stoffelen et al. 2020) along with the transport of key variables in the atmosphere such as temperature and moisture. Wind observations are particularly important in the tropics, where temperature observations cannot be used effectively to inform the wind field (Gordon et al. 1972; Žagar et al. 2004). Li et al. (2023) found that wind observations in the tropics are not only important for wind analysis but also more useful than temperature observations for informing the atmospheric humidity. The Aeolus ALADIN lidar has been found to improve weather forecasts by providing direct wind observations (Rennie et al. 2021; Garrett et al. 2022; Laroche and St-James 2022; Pouret et al. 2022). Unlike a 355-nm lidar such as Aeolus ALADIN, a 2053-nm lidar has no sensitivity to molecular scattering and would require an aerosol or cloud layer to retrieve 3D winds. In this work, lidar observations are simulated using optical properties derived from aerosols that are interactive with the G5NR meteorology. This is an advancement compared to previous OSSEs that have simulated wind lidars.
Two assumptions regarding how clouds could be simulated in an OSSE are tested, with the results showing a relatively small effect on the distribution of observations and an even smaller difference in analysis error impact. In all cases presented, the inclusion of 2053-nm lidar resulted in an overall improvement in terms of analysis error for the wind field. Two assumptions were tested regarding aerosol backscatter limits required for a wind retrieval to take place: one with a nominal assumption based on available DAWN flight campaign data and the other with stricter criteria also within the range of the flight campaign data. The stricter criteria resulted in fewer observations in the mid- to upper troposphere in the tropics and polar regions, and as a result, the lidar showed less impact in those regions compared with the less restrictive case. All cases showed an improvement in the midtroposphere wind error in the tropics and polar regions, with the less restrictive case having a greater impact in the upper stratosphere owing to a lower backscatter requirement.
Additionally, the impact of observation weighting in the GEOS-ADAS was tested by running an additional six cases scaling the DAS-assumed observation error by factors ranging from 0.75 to 8.0. While it would take more cases to determine the optimal assumed observation error, it appears that the optimal weighting lies somewhere between 2 and 4 times that of the assumed rawinsonde error. It should be noted that this is without adding correlated or random error to the observations themselves and is considered the intrinsic representativeness error of the simulated observations when used in the GEOS-ADAS. Additional work would be needed to determine the optimal weighting with errors added to the observations.
The lidar simulations described here reflect the distribution of clouds and aerosols in boreal summer; however, there are large seasonal differences in the global distribution of aerosols. The G5NR aerosol climatology has been evaluated (Gelaro et al. 2015) over the full 2-yr period, and additional seasonal aerosol calculations have recently been performed. Sampling a range of dates in January and July, there are clear seasonal differences in total aerosol concentration, especially in the middle and upper troposphere. There is a particularly large difference in concentrations over the central Atlantic, with total aerosol concentrations that are one and two orders of magnitude larger in July than in January in the middle and upper levels, likely due to Saharan dust. Smaller seasonal differences are seen in order regions, with smallest differences in the tropics and larger differences in the extratropics. However, in the midlatitudes, upper-level and lower-level trends are reversed, with larger concentrations in the middle and upper troposphere in the local summer and larger concentrations in lower levels in the winter. These differences are due to the different types of aerosols and the differing transport properties of the species. Some seasonal variation in the distribution of lidar observations is anticipated due to cycles in the regional concentration of both aerosols and cloud climatology.
This work was an initial step in developing a full OSSE, and there are a number of additional steps to reach a comprehensive conclusion regarding how the instrument would perform. Further work includes adding correlated or random errors to the observations, along with running forecasts and associated metrics. However, this initial work indicates that the addition of a 2053-nm wind lidar would improve the wind analysis within the GEOS-ADAS and provides guidance for future work.
Acknowledgments.
We thank NASA’s Earth Science Technology Office for funds to pursue this work. Computing resources supporting this work were provided by the NASA High-End Computing (HEC) Program through the NASA Center for Climate Simulation (NCCS) at the Goddard Space Flight Center. We also thank Patricia Castellanos for providing code to calculate aerosol backscatter based on the G5NR and GEOS-FP. 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) of the EUMETSAT, with the assistance of Sean Healy at the ECMWF.
Data availability statement.
The dataset on which this paper is based is too large to be retained or publicly archived with available resources. The G5NR dataset is available via online portal at https://gmao.gsfc.nasa.gov/global_mesoscale/7km-G5NR/. Documentation and methods used to support this study are available from Bryan M. Karpowicz at the NASA GMAO.
APPENDIX
Acronyms and Abbreviations
| 4D |
Four-dimensional |
| ALADIN |
Atmospheric laser Doppler instrument |
| AMSR-2 |
Advanced Microwave Scanning Radiometer 2 |
| CALIOP |
Cloud–Aerosol Lidar with Orthogonal Polarization |
| CERES |
Clouds and the Earth’s Radiant Energy System |
| DAS |
Data assimilation system |
| DAWN |
Doppler Aerosol Wind |
| ECMWF |
European Centre for Medium-Range Weather Forecasts |
| EUMETSAT |
European Organisation for the Exploitation of Meteorological Satellites |
| GEOS |
Goddard Earth Observing System |
| G5NR |
GEOS-5 nature run |
| GEOS-ESM |
GEOS Earth System Model |
| GEOS-FP |
Goddard Earth Observing System Forward Processing |
| GMAO |
Global Modeling and Assimilation Office |
| GMI |
Global Precipitation Measurement (GPM) Microwave Imager |
| GPS |
Global positioning system |
| GPS-RO |
Global positioning system radio occultation |
| GSI |
Gridpoint Statistical Interpolation analysis system |
| LTAN |
Local time of ascending node |
| MERRA-2 |
Modern-Era Retrospective Analysis for Research and Applications 2 |
| MHS |
Microwave humidity sounder |
| NASA |
National Aeronautics and Space Administration |
| NWP |
Numerical weather prediction |
| OSSE |
Observation system simulation experiment |
| PBL |
Planetary boundary layer |
| ROPP |
Radio Occultation Processing Package |
| ROM |
Radio Occultation Meteorology |
| SAF |
Satellite Application Facility |
| SNR |
Signal-to-noise ratio |
| UV |
Ultraviolet |
| UTC |
Coordinated universal time |
REFERENCES
Arnold, C. P., Jr., 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., and G. Emmitt, 2008: Review of observing system simulation experiments to evaluate the potential impact of lidar winds on numerical weather prediction. 24th Int. Laser Radar Conf. 2008 ILRC24, Boulder, CO, International Coordination Group on Laser Atmospheric Studies, 725–729, https://iclas-ilrc.org/.
Atlas, R., E. Kalnay, and M. Halem, 1985: Impact of satellite temperature sounding and wind data on numerical weather prediction. Opt. Eng., 24, 341–346, https://doi.org/10.1117/12.7973481.
Atlas, R., and Coauthors, 2015: Observing system simulation experiments (OSSEs) to evaluate the potential impact of an optical autocovariance wind lidar (OAWL) on numerical weather prediction. J. Atmos. Oceanic Technol., 32, 1593–1613, https://doi.org/10.1175/JTECH-D-15-0038.1.
Baron, P., and Coauthors, 2017: Feasibility study for future spaceborne coherent Doppler wind lidar, Part 2: Measurement simulation algorithms and retrieval error characterization. J. Meteor. Soc. Japan, 95, 319–342, https://doi.org/10.2151/jmsj.2017-018.
Bedka, K. M., and Coauthors, 2021: Airborne lidar observations of wind, water vapor, and aerosol profiles during the NASA Aeolus calibration and validation (Cal/Val) test flight campaign. Atmos. Meas. Tech., 14, 4305–4334, https://doi.org/10.5194/amt-14-4305-2021.
Cardinali, C., J. Pailleux, and J.-N. Thépaut, 1998: Use of simulated Doppler wind lidar data in NWP: An impact study. Le Centre national de la recherche scientifique Note de travail du Groupe de Modelisation pour L’assimilation et le Prevision 6, 25 pp.
Castellanos, P., A. M. Da Silva, A. S. Darmenov, V. Buchard, R. C. Govindaraju, P. Ciren, and S. Kondragunta, 2019: A geostationary instrument simulator for aerosol observing system simulation experiments. Atmosphere, 10, 2, https://doi.org/10.3390/atmos10010002.
Colarco, P. R., E. P. Nowottnick, C. A. Randles, B. Yi, P. Yang, K.-M. Kim, J. A. Smith, and C. G. Bardeen, 2014: Impact of radiatively interactive dust aerosols in the NASA GEOS-5 climate model: Sensitivity to dust particle shape and refractive index. J. Geophys. Res. Atmos., 119, 753–786, https://doi.org/10.1002/2013JD020046.
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., 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.
Frehlich, R. G., and M. J. Kavaya, 1991: Coherent laser radar performance for general atmospheric refractive turbulence. Appl. Opt., 30, 5325–5352, https://doi.org/10.1364/AO.30.005325.
Garrett, K., H. Liu, K. Ide, R. N. Hoffman, and K. E. Lukens, 2022: Optimization and impact assessment of Aeolus HLOS wind assimilation in NOAA’s global forecast system. Quart. J. Roy. Meteor. Soc., 148, 2703–2716, https://doi.org/10.1002/qj.4331.
Gelaro, R., and Coauthors, 2015: Evaluation of the 7-km GEOS-5 nature run. NASA Tech. Rep. NASA/TM-2014-104606, 305 pp., https://gmao.gsfc.nasa.gov/pubs/docs/Gelaro736.pdf.
Gong, S. L., L. A. Barrie, and J.-P. Blanchet, 1997: Modeling sea-salt aerosols in the atmosphere: 1. Model development. J. Geophys. Res., 102, 3805–3818, https://doi.org/10.1029/96JD02953.
Gordon, C. T., L. Umscheid Jr., and K. Miyakoda, 1972: Simulation experiments for determining wind data requirements in the tropics. J. Atmos. Sci., 29, 1064–1075, https://doi.org/10.1175/1520-0469(1972)029<1064:SEFDWD>2.0.CO;2.
Hess, M., P. Koepke, and I. Schult, 1998: Optical properties of aerosols and clouds: The software package OPAC. Bull. Amer. Meteor. Soc., 79, 831–844, https://doi.org/10.1175/1520-0477(1998)079<0831:OPOAAC>2.0.CO;2.
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.
Hoffman, R. N., C. Grassotti, R. G. Isaacs, J.-F. Louis, T. Nehrkorn, and D. C. Norquist, 1990: Assessment of the impact of simulated satellite lidar wind and retrieved 183 GHz water vapor observations on a global data assimilation system. Mon. Wea. Rev., 118, 2513–2542, https://doi.org/10.1175/1520-0493(1990)118<2513:AOTIOS>2.0.CO;2.
Kavaya, M. J., J. Y. Beyon, G. J. Koch, M. Petros, P. J. Petzar, U. N. Singh, B. C. Trieu, and J. Yu, 2014: The Doppler Aerosol Wind (DAWN) airborne, wind-profiling coherent-detection lidar system: Overview and preliminary flight results. J. Atmos. Oceanic Technol., 31, 826–842, https://doi.org/10.1175/JTECH-D-12-00274.1.
Köepke, P., M. Hess, I. Schult, and E. Shettle, 1997: Global aerosol data set. Max-Planck-Institut für Meteorologie Rep. 243, 46 pp., https://hdl.handle.net/21.11116/0000-0009-EB9B-0.
Laroche, S., and J. St-James, 2022: Impact of the Aeolus Level-2B horizontal line-of-sight winds in the environment and climate change Canada global forecast system. Quart. J. Roy. Meteor. Soc., 148, 2047–2062, https://doi.org/10.1002/qj.4300.
Li, L., N. Žagar, K. Raeder, and J. L. Anderson, 2023: Comparison of temperature and wind observations in the tropics in a perfect-model, global EnKF data assimilation system. Quart. J. Roy. Meteor. Soc., 149, 2367–2385, https://doi.org/10.1002/qj.4511.
Marseille, G. J., and A. Stoffelen, 2003: Simulation of wind profiles from a space-borne Doppler wind lidar. Quart. J. Roy. Meteor. Soc., 129, 3079–3098, https://doi.org/10.1256/qj.02.96.
Marseille, G. J., A. Stoffelen, F. Bouttier, C. Cardinali, D. de Haan, and D. Vasiljevic, 2001: Impact assessment of a Doppler wind lidar in space on atmospheric analyses and numerical weather prediction. KNMI Rep. WR-2001-03, 8 pp.
Masutani, M., and Coauthors, 2010: Observing system simulation experiments at the national centers for environmental prediction. J. Geophys. Res., 115, D07101, https://doi.org/10.1029/2009JD012528.
McCarty, W., D. Carvalho, I. Moradi, and N. C. Privé, 2021: Observing system simulation experiments investigating atmospheric motion vectors and radiances from a constellation of 4–5-μm infrared sounders. J. Atmos. Oceanic Technol., 38, 331–347, https://doi.org/10.1175/JTECH-D-20-0109.1.
Meng, Z., P. Yang, G. W. Kattawar, L. Bi, K. N. Liou, and I. Laszlo, 2010: Single-scattering properties of tri-axial ellipsoidal mineral dust aerosols: A database for application to radiative transfer calculations. J. Aerosol Sci., 41, 501–512, https://doi.org/10.1016/j.jaerosci.2010.02.008.
NASEM, 2018: Thriving on Our Changing Planet: A Decadal Strategy for Earth Observation from Space. The National Academies Press, 28 pp., https://doi.org/10.17226/25437.
Okamoto, K., T. Ishibashi, S. Ishii, P. Baron, K. Gamo, T. Y. Tanaka, K. Yamashita, and T. Kubota, 2018: Feasibility study for future space-borne coherent Doppler wind lidar, Part 3: Impact assessment using sensitivity observing system simulation experiments. J. Meteor. Soc. Japan, 96, 179–199, https://doi.org/10.2151/jmsj.2018-024.
Pouret, V., M. Savli, J.-F. Mahfouf, D. Raspaud, A. Doerenbecher, H. Bénichou, and C. Payan, 2022: Operational assimilation of Aeolus winds in the Météo-France global NWP model ARPEGE. Quart. J. Roy. Meteor. Soc., 148, 2652–2671, https://doi.org/10.1002/qj.4329.
Privé, N. C., R. M. Errico, and W. McCarty, 2021: The importance of simulated errors in observing system simulation experiments. Tellus, 73A, 1886795, https://doi.org/10.1080/16000870.2021.1886795.
Privé, N. C., R. M. Errico, and A. E. Akkraoui, 2022: Investigation of the potential saturation of information from global navigation satellite system radio occultation observations with an observing system simulation experiment. Mon. Wea. Rev., 150, 1293–1316, https://doi.org/10.1175/MWR-D-21-0230.1.
Privé, N. C., E. L. McGrath-Spangler, D. Carvalho, B. M. Karpowicz, and I. Moradi, 2023: Robustness of observing system simulation experiments. Tellus, 75A, 309–333, https://doi.org/10.16993/tellusa.3254.
Rennie, M. P., L. Isaksen, F. Weiler, J. de Kloe, T. Kanitz, and O. Reitebuch, 2021: The impact of Aeolus wind retrievals on ECMWF global weather forecasts. Quart. J. Roy. Meteor. Soc., 147, 3555–3586, https://doi.org/10.1002/qj.4142.
Rienecker, M., and Coauthors, 2008: The GEOS-5 Data Assimilation System—Documentation of versions 5.0.1, 5.1.0, and 5.2.0. Tech. Memo. NASA/TM-2008-104606, Vol. 27, 97 pp., http://gmao.gsfc.nasa.gov/pubs/docs/Rienecker369.pdf.
Shettle, E. P., and R. W. Fenn, 1979: Models for the aerosols of the lower atmosphere and the effects of humidity variations on their optical properties, Vol. 79. Air Force Geophysics Laboratory, Air Force Systems Command Environmental Research Paper 66, 25 pp., https://web.gps.caltech.edu/∼vijay/Papers/Aerosol/SF79-Aerosol-Models-part1of4.PDF.
Srivastava, V., and Coauthors, 2001: Wavelength dependence of backscatter by use of aerosol microphysics and lidar data sets: Application to 2.1-μm wavelength for space-based and airborne lidars. Appl. Opt., 40, 4759–4769, https://doi.org/10.1364/AO.40.004759.
Stoffelen, A., G. J. Marseille, F. Bouttier, D. Vasiljevic, S. de Haan, and C. Cardinali, 2006: ADM-Aeolus Doppler wind lidar observing system simulation experiment. Quart. J. Roy. Meteor. Soc., 132, 1927–1947, https://doi.org/10.1256/qj.05.83.
Stoffelen, A., and Coauthors, 2020: Wind profile satellite observation requirements and capabilities. Bull. Amer. Meteor. Soc., 101, E2005–E2021, https://doi.org/10.1175/BAMS-D-18-0202.1.
Straume, A., and Coauthors, 2020: ESA’s space-based Doppler wind lidar mission Aeolus – First wind and aerosol product assessment results. EPJ Web Conf., 237, 01007, https://doi.org/10.1051/epjconf/202023701007.
Wiscombe, W. J., 1980: Improved Mie scattering algorithms. Appl. Opt., 19, 1505–1509, https://doi.org/10.1364/AO.19.001505.
Wu, D., J. Tang, Z. Liu, and Y. Hu, 2013: Simulation of coherent Doppler wind lidar measurement from space based on CALIPSO lidar global aerosol observations. J. Quant. Spectrosc. Radiat. Transfer, 122, 79–86, https://doi.org/10.1016/j.jqsrt.2012.11.017.
Žagar, N., N. Gustafsson, and E. Källén, 2004: Variational data assimilation in the tropics: The impact of a background-error constraint. Quart. J. Roy. Meteor. Soc., 130, 103–125, https://doi.org/10.1256/qj.03.13.
