1. Introduction
Satellite meteorology is entering its sixth decade. The concept of geostationary (GEO) imaging, realized in the 1970s, marks a turning point in the development of the space segment of the Earth Observing System (EOS). Within the spatially fixed field of view of a GEO satellite, every point can be observed essentially at all times, that is, with a refresh rate of 15 min or less (~1 min on demand in limited sectors). The nominal (nadir) spatial resolution of GEO imagers is on the order of 1 km for visible imagery and 3 km for infrared imagery. Low-Earth-orbiting (LEO) satellites, which circle around the globe on high-inclination orbits, represent another key component of the EOS. LEO satellites are essential to the observation of high-latitude regions since the latitudinal limit for quantitative utilization of GEO satellites is about 60°. However, even a sizeable constellation of LEO satellites cannot provide high temporal imagery over polar regions, especially near the GEO limit. Specifically, it was estimated that as many as 23 LEO satellites would be required to obtain the equivalent of 15-min imagery at 60° of latitude (Trishchenko and Garand 2012). The compositing of imagery from many satellites is certainly feasible, and is in fact done routinely to the extent possible to facilitate the work of operational meteorologists. Still, there are significant issues for scientific/quantitative utilization of composite imagery, such as time and viewing-angle discontinuities and the incomplete view of the polar domain. In addition, the time needed to produce composite images and derived products grows with the number of satellites and receiving stations.
A satellite constellation defined by the smallest number of satellites is desirable to emulate GEO capabilities at high latitudes. From that viewpoint, imaging from a highly elliptical orbit (HEO) is a noteworthy option. Both Canada and the Russian Federation are considering HEO missions for meteorological observation of the circumpolar Arctic by the end of this decade. It was demonstrated that with only two satellites operating at heights varying between about 25 000 and 45 000 km, which is not that different from the constant GEO heights (~36 000 km), it is possible to image the entire circumpolar region (60°–90°N) at all times (Kidder and Vonder Haar 1990; Trishchenko et al. 2011). Another pair of satellites would be required to cover the Antarctic area (60°–90°S). The ultimate goal of “seeing” each portion of the globe at all times would then be reached. For that reason, the concept of an HEO constellation is endorsed by the World Meteorological Organization (WMO 2009).
The HEO mission being proposed by Canada is called the Polar Communications and Weather (PCW) mission. It is defined by a two-satellite HEO constellation providing meteorological observation, space weather, and communication services in the Arctic (Canadian Space Agency; http://www.asc-csa.gc.ca). The instrument selected for meteorological imaging would be an advanced imager similar to that planned for the Geostationary Operational Environmental Satellite-R (GOES-R; Schmit et al. 2005) or Meteosat Third Generation (EUMETSAT 2007). If a mission like PCW is realized, the high temporal applications based on GEO imagers would then be extended to high latitudes, along with applications more specific to these latitudes, such as sea ice monitoring.
The particular focus of this study is to assess the potential impact on forecasts of HEO-derived atmospheric motion wind vectors (AMVs). AMVs are a very important satellite product for numerical weather prediction (NWP) (Velden et al. 2005). An international working group under the auspices of WMO, the International Winds Working Group (IWWG; http://cimss.ssec.wisc.edu/iwwg/iwwg.html) focuses on AMV product generation, error characterization, and impact in NWP [see Payan and Cotton (2012) for an intercomparison of AMV impact]. The high-latitude observation gap (relative to GEO imaging) is particularly striking for that product. Figure 1 shows the typical AMV coverage for a recent 6-h period. GEO coverage is generally good between 55°N and 55°S. There is also good coverage in regions 70°–90°N/S from LEO satellites [Moderate Resolution Imaging Spectrometer (MODIS) sensors pictured in Fig. 1]. Advanced Very High Resolution Radiometer (AVHRR)-derived AMVs from Meteorological Operational (MetOp) and National Oceanic and Atmospheric Administration (NOAA) satellites can also be used. The gap is severe between latitudes 55° and 70°, a region often characterized by rapidly evolving weather systems, notably polar lows. The gap is due to the fact that the production of AMVs requires temporal sequences involving three consecutive scenes to identify suitable targets (i.e., clouds) for tracking, to evaluate the speed and direction at which these targets are moving, and to estimate the height of the targets. It is possible to work with only image pairs, but the overall resulting quality is then degraded. AMVs from combining MetOp-A and MetOp-B AVHRR imagery are currently under evaluation, with the potential to eliminate part of the gap seen in Fig. 1. Polar AMVs as shown in Fig. 1 are obtained from the same MODIS satellite, and therefore the time difference between scenes is about 90 min. Ideally, the time difference should not exceed 20 min (Shimoji and Hayashi 2012). In an attempt to remedy the gap situation, LEO–LEO and even LEO–GEO AMVs are now being produced, combining imagery from multiple satellites (Hoover et al. 2012). The time difference between image pairs or triplets is then highly variable. There are often large viewing-angle differences between the scenes used for tracking, which is not the case for GEO or HEO temporal sequences. As well, combining information from different satellites raises issues linked to differences in channel characteristics, calibration, and bias correction. For these reasons, it is anticipated that HEO AMVs would be a preferred option to LEO–LEO or LEO–GEO AMVs in terms of overall quality, spatiotemporal coverage, and timeliness of delivery to NWP centers. In addition, MODIS is approaching its end of life, and neither AVHRR nor Visible Infrared Imaging Radiometer Suite (VIIRS; on board the Suomi National Polar-Orbiting Partnership satellite) instruments have a water vapor channel for AMV production (and improved height assignment).
Example of GEO (five satellites) and LEO (Terra and Aqua) AMV availability after quality control and thinning for the 6-h period centered at 0000 UTC 1 Nov 2012. Satellites are identified by name and associated color at the top of the figure.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
To complete the description of possible sources of atmospheric wind data from space, lidar wind profiles from the Atmospheric Dynamic Mission Aeolus satellite (ADM-Aeolus; Stoffelen et al. 2006) should be available by 2016 from active retrievals on narrow (50 km) swaths. Daytime wind observations from the Multiangle Imaging Spectroradiometer (MISR on the Terra satellite; Davies et al. 2007) constitute another dataset for potential use in numerical weather prediction, but a follow-up mission would be needed in the future.
A specific characteristic of HEO viewing is that the height of the satellite varies continually along the orbit, although not by much for several hours near the apogee point. This is shown in Fig. 2 for a 16-h three-apogee (hence labeled TAP) orbit with apogees separated by 120° in longitude [see Trishchenko et al. (2011) for comparisons of TAP with other orbits]. This orbit was proposed as an optimal choice that combines good viewing conditions and a benign ionizing radiation environment. The satellite would start imaging when, on ascent, it crosses 35°N (height of about 28 000 km). The satellite would reach its apogee position at 66°N (height of ~43 000 km) about 4–5 h later depending on the imaging requirements (continuous imaging sequence of 8–10 h per orbit). We show in Fig. 2 that the satellite spends a lot of time between 30° and 66°N. As the two satellites, operating simultaneously in the same orbital plane, would be offset by 8 h, they would provide stereo views between 3 and 5 h from apogee. AMVs would be derived from scenes remapped onto a common fixed grid, facilitating the utilization of time sequences for AMV production. While GEO viewing angles are fixed for a given location, viewing angles from HEO satellites vary in time. Thus, for views closer to nadir, HEO products could supplement advantageously GEO products in the overlap region equatorward 60° of latitude (i.e., where GEO viewing angles are high). Figure 3 shows the spatiotemporal coverage for 16- and 24-h HEO constellations of two satellites designed for Northern Hemisphere high-latitude imaging. The figure was derived from the model presented by Trishchenko and Garand (2012). The coverage is 100% north of 60°N, and it remains as high as 80% at 45°N. Figure 3 also shows that image triplets for AMV production are separated by less than 25 min down to 30°N, with the spatial coverage dropping to 66% at that latitude. The 24-h orbit called Tundra has some capabilities for AMV production extending to the Antarctic region, with 45-min image triplets available about 8 h day−1.
Ground track (latitude–longitude, blue) and height positions (km, blue) for a 16-h highly elliptical orbit (red). Dots represent hourly intervals. The three-apogee ground track is repeated every 48 h. The imaging period per orbit is 10 h and starts at about 35°N (height ~ 28 000 km), reaching apogee at 66°N (height ~ 43 000 km) 5 h later. The second satellite would be in the same orbital plane with an 8-h difference.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
(top) Number of observations per day for any point at a given latitude for the 16-h TAP orbit (contiguous squares) and the 24-h Tundra orbit (eccentricity, 0.3; inclination, 90°; solid lines) assuming two-satellite constellations and 15-min repeat cycle. (bottom) Corresponding intervals (min) between scenes and average times between single (black), two (pair, red), and three (triplet, green) consecutive scenes.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
For the impact evaluation in NWP, this study relies on observing system simulation experiments (OSSEs). OSSEs are increasingly used in support of future satellite missions [see Masutani et al. (2010a,b) for a review, and specific studies such as Lahoz et al. (2005), Hartung et al. (2011), and Errico et al. (2007)]. All observation data types currently used operationally at the Canadian Meteorological Center are simulated in addition to the new data type of interest: HEO AMVs filling the high-latitude gap. The OSSEs use the same forecast–assimilation system as the real system. The added value of HEO AMVs is assessed in forecasts evaluated against a reference atmosphere serving as “truth.”
Section 2 provides details on the OSSE setup and the definition of the experiments. Results are presented in section 3. Several questions are being investigated. First, there is a need to assess the realism of the OSSE. This is done by comparing the impact of real and simulated AMVs on forecasts, with the same observation distribution in both. We then proceed with evaluating the added value of HEO AMVs in the Northern Hemisphere, on top of what is currently available. An evaluation is also made of the relative impacts of HEO AMVs in both northern (Arctic) and southern (Antarctic) polar regions. Section 4 concludes the study.
2. OSSE definition
a. Nature run
The idea of an OSSE is to define a four-dimensional truth atmosphere, referred to as the nature run (NR), to evaluate various aspects of a data assimilation system and the added value of future observations. Simulated observations representing known observation types are extracted from the NR and assimilated using procedures as close as possible to those used with real observations. This defines the control experiment or cycle. The data from the future observation source are simulated according to defined coverage and error characteristics. The impact of a future data type on forecasts can then be evaluated in comparison to that seen in the control experiment.
The nature run used in this study was created by the European Centre for Medium-Range Weather Forecasts (ECMWF) using the forecast model version of the CY31r1cycle from the ECMWF Integrated Forecasting System (for model details see http://www.ecmwf.int/research/ifsdocs/CY31r1). The intent was that the NR could be used by various centers as part of the Joint OSSE program (Reale et al. 2007; Masutani et al. 2010a,b; www.emc.ncep.noaa.gov/research/JointOSSEs). It is a free forecast run at about 40-km horizontal resolution with 91 vertical levels from the surface to 0.02 hPa. The prescribed sea surface temperatures (SSTs) and ice cover were those provided by the National Centers for Environmental Prediction (NCEP) covering May 2005–May 2006. The NR was interpolated onto the global Canadian Meteorological Centre (CMC) model grid used in 2008, which has a similar resolution (i.e., 33 km at 49° of latitude and 80 levels with the model top at 0.1 hPa). The interpolated NR can then be used for validation of OSSEs conducted with the CMC data assimilation system (see Gauthier et al. 1999).
The initial background field used to start assimilation cycles was a 5-day forecast initiated with the CMC version of the NR, using the operational Global Environmental Multiscale (GEM) model (see Charron et al. 2012 and references therein). The resulting initial conditions are therefore not identical but rather synoptically similar to the NR. The global CMC model is used for all forecast and assimilation cycles. It is a different model from that used to create the NR, as recommended by Stoffelen et al. (2006) and other OSSE investigators. The SST and albedo fields used in the assimilation cycles are the ones available at CMC, which also differ from the ones present in the original NR.
b. Simulation of existing observations
All observation types available for operational assimilation at CMC during 2008 and 2009 were simulated. Table 1 (nonradiances) and Table 2 (radiances) list the various data types and the corresponding numbers of observations per analysis or per day. For global models, radiances typically account for more than 85% of the total number of assimilated observations (e.g., Moll et al. 2012). In this study, assimilation cycles were run for the 2-month period from 23 December 2005 to 23 February 2006. In 2008, new data types, such as radiances from the Infrared Atmospheric Sounding Interferometer (IASI, on board MetOp) and refractivities from Ground Positioning System Radio Occultation (GPSRO; several satellites listed in Table 1), were introduced for use with the operational suite. Since these data types were not present at the time of the NR, all observations were simulated at the locations and times of the data assimilated in 2008 and 2009 for exactly the same period, but 3 yr apart. The simulation process involves two steps.
Observation types assimilated operationally in 2008 excluding radiances. Details on observed variables, applied resolution, and data density are provided. The variable Ps is surface pressure. COSMIC is the Constellation Observing System for Meteorology, Ionosphere, and Climate; GRACE is the Gravity Recovery and Climate Experiment; CHAMP is the Challenging Minisatellite Payload.
Satellite radiances assimilated operationally in 2008 with details provided on instrument, platform, typical number of observations assimilated per day, and contributing channels. DMSP is the Defense Meteorological Satellite Program.
1) Extraction of pseudo-observations from the NR
Observations representing model variables such as temperature, wind, humidity, or surface pressure are directly extracted from the NR following interpolation in space and time to the locations of the true observations (including currently available AMVs). For observation types that are indirect (i.e., not part of the model state), like radiances and GPSRO data, a forward model is required. Heilliette et al. (2013) provides details on the simulations of all-sky (cloudy) infrared radiances, followed by quality control procedures that are identical to those used operationally to identify non-cloud-affected radiances. The forward model for GPSRO refractivity is defined by Aparicio et al. (2009). HEO AMV extraction is described in section 2c.
2) Perturbation of the pseudo-observations
Perturbations are defined in terms of a factor β, typically in the range 0.2–1.0, that multiplies the assigned observation error standard deviation. To account for effects such as spatiotemporal or interchannel error correlations (for radiances), the assigned observation error standard deviations are typically larger than those attributed to measurements or retrieval error standard deviations. This is why β factors are not unity. The optimum β factor is found by minimizing the difference between real and simulated standard deviations of observed minus analyzed values (O − A), pertaining to each data type. The general procedure establishing perturbation levels is described by Rochon et al. (2012), and in the specific case of hyperspectral infrared radiances by Heilliette et al. (2013). The perturbation factor pertaining to AMVs is defined below in section 2c(4). No attempt was made to simulate horizontal error correlations, which may contribute to too optimistic results in the OSSE. Errico et al. (2013) modeled these correlations for AMVs and radiances, demonstrating that they remain significant up to about 500 km.
c. Simulation of HEO AMVs
Ideally, AMVs should be generated with dedicated software used by data producers. Geostationary AMVs are extracted from image triplets at 15–30-min intervals using imagery at a resolution on the order of 4 km. AMVs can be derived from visible and infrared windows, as well as infrared water vapor channels. The study of AMV error characteristics from simulated imagery is a promising area of research (e.g., Hernandez-Carrascal et al. 2012). However, this is not possible within the current context because the nature run is at too coarse a resolution, both temporally and horizontally. A practical approach was devised that takes NR cloudiness into account. There is no incentive to exactly simulate the HEO spatiotemporal coverage since it is 100% above 60° latitude and nearly complete (>85%) above 50° latitude (Fig. 3). HEO AMV observations at the NWP model resolution are generated at each grid point, except where the sky is clear. Currently, clear-air AMVs based on water vapor imagery are not assimilated in the polar regions at CMC. There are four steps for generating HEO AMVs:
Global imagery was simulated from the NR (every 3 h) by representing channel 31 of MODIS (10.7–11.3 μm). This is done using the Radiative Transfer for the Television and Infrared Observation Satellite (TIROS) Operational Vertical Sounder (RTTOV), version 8.7 (Matricardi and Saunders 1999). Cloud input from the NR is layer cloud fraction and cloud water or ice content. Cloud properties are defined exactly as stated in Garand et al. (2011). That study demonstrated the realism of simulated all-sky imagery based on RTTOV and model output.
The cloud top is derived from the cloud transmittance tc, an RTTOV output defined from each level to the top of the atmosphere (TOA). Following Garand et al. (2011), the effective cloud top is defined at the model level L, where tc (L, TOA) reaches 0.9. Thus, very thin clouds are ignored (i.e., the scene is defined as clear if the cloud transmittance exceeds 0.9). It was demonstrated that this definition provides cloud-top estimates that agree well with lidar observations from the Cloud–Aerosol Lidar with Orthogonal Polarization (CALIOP; see, e.g., Winker et al. 2009). No attempt was made to generate AMVs from simulated visible or water vapor imagery. Only infrared-derived AMVs (from the 11-μm window and 6.7-μm water vapor channels) are assimilated operationally at CMC in polar regions.
Wind speed and direction are extracted from the NR at the estimated cloud top. No estimate is available in clear air. Values in the range 100–925 hPa are retained, and high-latitude AMVs are thinned at the resolution of 180 km, as is done with real polar AMVs. The HEO AMVs are defined between 50° and 90°N/S, a reasonable demarcation between GEO and HEO best coverage areas. Figure 4 shows an example of HEO AMV locations after thinning. The color scale indicates the pressure height. Since the analysis cycle is defined by 6-h intervals, HEO AMVs are assimilated every 6 h as well, although they are available every 3 h. Table 3 compares the density of GEO, LEO, and HEO AMV observations used in this study. HEO data clearly dominate at latitudes 50°–70°N/S. At latitudes 70°–90°N/S, the density of LEO data is higher than that of HEO data due to their availability throughout the 6-h assimilation window. In actual implementation, HEO AMVs would be available hourly over the circumpolar domain. The cloud distribution resulted in more HEO AMVs in the Arctic polar region in comparison with the Antarctic region.
The simulated AMV observations are perturbed as a function of the assigned observation errors. Assigned observation errors are listed in Table 4 for radiosonde winds, aircraft winds, and AMVs as a function of pressure level. Typical standard deviations of observed minus background (O − B, where background B is a short-term forecast, here 6 h) for AMVs in the Arctic and Northern Hemisphere midlatitudes are also presented. It is seen that assigned AMV errors are “inflated,” that is, larger than (O − B) standard deviations; this is mostly to compensate for horizontal–vertical error correlations not taken into account in the assimilation (diagonal error matrix). The perturbation factors β for radiosonde, aircraft, and AMV winds are 0.92, 0.56, and 0.28, respectively. In the case of AMVs, this means that the perturbation standard deviation varies between 0.84 and 1.68 m s−1 depending on level [i.e., about 40% of the (O − B) values listed in Table 4 for the Arctic region]. A Gaussian random generator with zero mean and standard deviation equal to β times the assigned observation error standard deviation is used to define the perturbations. The zonal and meridional wind components are perturbed independently. No attempt was made to consider perturbation factors that vary with pressure level.
Example of simulated HEO AMV positions extracted from the NR for the 50°–90°N region. Pressure height (Pa) is indicated on the left [blue (red) color for highest (lowest) clouds], where e+XX indicates multiplication by 10 raised to the +XX power. Regions devoid of data represent either clear skies or clouds outside the range 100–925 hPa. Thinning is done for a resolution of 180 km.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
Average number of AMV observations assimilated per 6-h period as a function of latitudinal region. Five satellites contribute to the GEO dataset (GOES-11 and -12, Meteosat-7–9, and MTSAT-1R), and two satellites contribute to the LEO dataset (Terra and Aqua).
Assigned observation error std (m s−1) as a function of pressure level pertaining to radiosondes (raobs), Aircraft Meteorological Data Relay (AMDAR), and AMV observations. Typical values of standard deviations for satellite AMV mean vector differences (MVD, m s−1) between observed and 6-h background values (O − B) for two latitudinal regions are shown in the right columns (source: CMC monitoring).
d. Assimilation cycles
Assimilation cycles cover the 2-month period between 15 December 2005 and 28 February 2006. Validation of forecasts up to 5 days covered the period 27 December 2005–23 February 2006. Experiments done with real observations correspond to the same dates but for the year 2008. Assimilation cycles were run in 3D-FGAT mode (FGAT: with the first guess interpolated at appropriate time, i.e., near the time of each observation). Using the three-dimensional variational data assimilation (3DVAR) method instead of the available 4DVAR method was motivated by the much faster turnaround of the assimilation cycles. With the assimilation done in 3DVAR mode, using HEO AMVs available every 6 h right on the analysis time is considered sufficient (as opposed to using all data available at 3-h intervals). The validation of forecasts can be done both versus their “own analyses” generated within a given cycle and versus the nature run.
This study is based on two observing system experiment (OSE) cycles, making use of real observations, and four OSSE cycles, making use of simulated observations. The following experiments are defined in Table 5: REAL_CNTL, REAL_NOAMV, SIM_CNTL, and SIM_NOAMV. These experiments are designed to reassess the impact of real AMV data, and to compare it with the impact seen in the simulated system. Experiment SIM_PCW1 differs from SIM_CNTL by adding HEO AMV in the Arctic. The SIM_PCW2 cycle is designed to evaluate the relative impact of HEO AMVs in the Arctic and Antarctic regions in comparison with SIM_NOAMV.
List of assimilation experiments. The first two use real observations (OSEs) while the following four are OSSEs.
3. Results
For OSSE experiments, forecasts can be validated against the NR and against their “own” analyses generated by the cycle. The “experiment” forecasts can also be evaluated against the “control” analysis (i.e., from SIM_CNTL). Validation against the NR is favored as it represents the truth. However, since observations are generated from the NR, it is expected that, at least in the short term, the impact of these observations on forecasts will appear stronger from evaluation against NR analyses than against their own analyses. Over longer forecasts (and for analysis differences that are not too large), convergence of results from both cases should increase the confidence on the inferred impact. With a real data assimilation system, validation can be performed against one's own analysis, which is the norm, or against an independent analysis obtained from another center, in addition to validation against observations such as radiosonde profiles. Validation against analyses can be examined in multiple ways: by level, region, variable, time series, and using various metrics such as mean and standard deviation differences or anomaly correlation. For conciseness, a selection of key results is presented.
a. Real and simulated AMV impact based on operational coverage
An OSSE system should be validated by comparing the impact of most important data types with that seen in the real system. Privé et al. (2013) did such a validation using an adjoint technique. As part of this OSSE system, the validation was done for infrared radiances (Heilliette et al. 2013), and here for AMVs using a data-denial approach. Figure 5 compares the impact of operational AMVs for 72-h forecasts obtained from both the real and simulated data assimilation systems. Top panels in Fig. 5 refer to the entire world while the bottom panels refer to the 50°–90°N region. The impact is shown in terms of mean and standard deviation (std) differences against their own analyses (for consistency) for profiles of four variables: zonal wind component, relative humidity, geopotential, and temperature (labeled UU, HR, GZ, and TT, respectively).
Validation of 72-h forecasts against their own cycle analyses for the meridional wind component (UU, m s−1), relative humidity (HR, %), geopotential (GZ, dam), and temperature (TT, °C): (top two rows) global and (bottom two rows) 50°–90°N regions: (left two columns) real and (right two columns) simulated assimilation systems. Experiment periods are indicated at the top of each four-panel set. Blue lines denote the control (CNTL) and the red lines are for the no-AMV (NOAMV) experiment. Full lines represent std and dashed lines represent bias with differences (NOAMV − CNTL) emphasized with horizontal color bars [blue (red) bars denote negative (positive) impacts of the NOAMV experiment]. The vertical axis is pressure (hPa).
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
The std profiles (shape and absolute values) pertaining to the four variables depicted in Fig. 5 are very similar for the real and simulated systems. Standard deviation differences between NOAMV (red) and CNTL (blue) are modest (1%–5%), but systematically positive from 100 hPa to the surface. Those differences are pictured along vertical axes (std to the right, bias to the left), with scaling based on the maximum difference seen in the profile. The positive impact of AMV is clear from the dominant positive (blue) std differences. Similar results were obtained for other specific regions as well as for validation against the NR in the case of OSSE cycles. The positive AMV impact seen in the OSSE tends to be higher than in the real system, which is most evident for UU and GZ global results.
Biases are generally near zero, except for a −1-K TT bias near the surface in the 50°–90°N region. This bias is maintained for the 24–120-h forecasts (not shown) and may be due in part to the differing CMC and NR SST analyses (Heilliette et al. 2013). An important issue for this work was that the different NWP models used for the NR (ECMWF model) and the OSSE (CMC model) could result in significant biases in the forecasts. This does not seem to be the case.
Further examination of AMV impact differences by region between the real and simulated systems is presented in Fig. 6. The figure shows the differences between standard deviations of 500-hPa temperature errors: REAL_NOAMV minus REAL_CNTL (left panels) and SIM_NOAMV minus SIM_CNTL (right panels) for five latitudinal regions (a positive difference implies a positive impact from AMV). Each cycle is evaluated against its own analysis for forecasts up to 5 days. For real data cycles, evaluation against independent analyses from ECMWF (not shown) gave very similar results to those presented in the left column of Fig. 6. The following conclusions can be drawn:
The impact of real AMVs is positive in all regions, and remains significant up to day 3 or day 4. At day 5, the impact is either neutral or positive with significance level < 95%. As seen by examining actual std values (not shown), AMV assimilation contributes to a reduction of the forecast error std at 72 h by 1%–2% for temperature and 2%–3% for wind.
The impact from both real and simulated AMVs is higher in the Southern Hemisphere than in the Northern Hemisphere.
The impact of simulated AMVs is larger than that of real AMVs, with increasing differences for most regions after 72 h. The impact from simulated AMVs is about twice that of real AMVs in the 20°–50°S region and is a factor of 3 higher in the tropics (20°S–20°N).
Differences in standard deviation of 500-hPa temperature (K) between NOAMV and CNTL experiments pertaining to (left) real and (right) simulated assimilation systems as a function of time into the forecast: (top to bottom) the five regions indicated (50°–90°N/S, 20°–50°N/S, 20°N–20°S). Cycles are validated against their own analyses. Gray shading indicates significance at the 95% level.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
b. Added value of Arctic HEO AMVs
Polar AMVs from LEO satellites will be produced in the coming years even if MODIS is not replaced following the end of the Aqua and Terra satellite missions. Polar AMVs are currently generated from AVHRR imagery (Dworak and Key 2009) and will soon be available from VIIRS imagery. Thus, a relevant question remains: what is the added value of HEO AMVs relative to what is currently available or anticipated in the near future? To study this question, the impact of HEO AMVs covering the circumpolar domain 50°–90°N is evaluated vis-à-vis a control experiment that includes all operational data types, including MODIS AMVs (i.e., SIM_PCW1 vs SIM_CNTL). As seen in Fig. 1, the main AMV gap area for the SIM_CNTL cycle is in the region 55°–70°N. A substantial portion of the latter domain is over land, benefiting from conventional observations, although sparser in coverage than at lower latitudes.
Figure 7 presents 24- and 72-h results in the form of profiles for cycles against their own and NR analyses for the 50°–90°N region. The significances of the standard deviation differences are again indicated by markers on the right vertical axis. At 24 h, the positive impact is quite strong for the validation against the NR but nearly neutral, and even locally negative, when validating against their own analysis. As alluded to at the beginning of this section, closer proximity to the NR is expected at short lead times. On the other hand, the values of the std's themselves are higher, especially at low levels when validating against the NR. This is expected, as validating a short-term forecast against the truth should yield larger (and, at best, similar) differences than validating it against its own analysis. Similarly with real observation systems, validation against an independent analysis normally yields larger differences than against its own cycle analysis.
OSSE results against (left two columns) the experiments' own analyses and (right two columns) NR at (top two rows) 24 and (bottom two rows) 72 h, comparing the PCW1 (red) and control (blue) experiments for the 50°–90°N region. Variables are defined as in Fig. 5. Red (blue) horizontal bars denote positive (negative) impacts from the PCW1 data.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
Examining now 72-h results, it is seen that the initially positive impact of PCW1 AMVs remains positive, especially for UU and GZ but also for HR and TT. At longer lead times, results validated against own analyses and the NR become very similar. Actual values of the std profiles are close (generally within 5%), with the exception of the near-surface TT and GZ std's, which remain higher for the validation against NR. This convergence of results provides confidence on the impact evaluation. The positive impact is found to be significant at the 95% level up to 72 h for the four variables from the surface to 300 hPa. Results become close to neutral at day 5 for that experiment (not shown). A slight positive impact was also noted up to 72 h at 300–500-hPa levels in the 20°–50°N region (not shown).
c. Comparing impacts from Arctic and Antarctic HEO AMVs
Zonal mean differences between the PCW2 and NOAMV std's scored against (left) their own analyses and (right) the NR for (top) 24- and (bottom) 120-h forecasts, i.e., PCW2 minus NOAMV. Vertical axis is pressure (hPa). Blue (red) color is indicative of positive (negative) impacts from PCW2 data.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
The relative impact of HEO AMVs in both hemispheres is further examined in Fig. 9 in terms of error std and bias profiles at 72 h against the NR for the four regions: 20°–50°N/S and 50°–90°N/S. The impact is generally positive at 100–1000-hPa levels for the four meteorological variables. The impact is clearly stronger in the Southern Hemisphere, including the 20°–50°S region.
Comparison of std and bias profiles for NOAMV (blue) and PCW2 (red) OSSE cycles validated against the NR for 72-h forecasts in regions (top left four panels) 20°–50°N, (top right four panels) 50°–90°N, (bottom left four panels) 20°–50°S, and (bottom right four panels) 50°–90°S. Variables are as in Fig. 5.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
Validation referring to the 500-hPa level is commonly used to estimate a predictability gain in terms of hours. Figure 10 presents 500-hPa TT and GZ std differences versus the nature run pertaining to NOAMV and PCW2 cycles as a function of lead time up to 120 h. Results refer again to the 50°–90°N/S regions. The positive impact is significant at all lead times for both variables and regions. The predictability gain is estimated to be of the order of 3 h for the 50°–90°N region and 6 h for the 50°–90°S region beyond day 3. These estimates can be reduced by about one-third (respectively 2 and 4 h) to account for the noted tendency of the OSSE to exaggerate the impact of simulated observations (Fig. 6). On the other hand, it is expected that higher impacts would have resulted from 4DVAR assimilation.
The 500-hPa (left) geopotential (dam) and (right) temperature (K) std (full lines) and biases (dashed lines) for NOAMV (blue) and PCW2 (red) forecasts up to 120 h evaluated against the NR. Differences between the two results are shown at the bottom of each panel with the gray shading denoting 95% significance in the differences. Results are shown for the (top) 50°–90°N and (bottom) 50°–90°S regions.
Citation: Journal of Applied Meteorology and Climatology 52, 8; 10.1175/JAMC-D-12-0333.1
4. Conclusions
The focus of this study was the evaluation of the added value for weather prediction of assimilating HEO AMVs filling the high-latitude observation gap. This research effort was conducted with the OSSE capability developed at Environment Canada. Here, the ECMWF nature run produced for the international Joint OSSE program defined the truth atmosphere. In this particular study, we also benefited from our experience with real AMV assimilation. It was then possible to conduct assimilation cycles based on both real and simulated data. Our main findings can be summarized below:
The study reaffirms that real AMVs have a significant positive impact in all regions at least up to 72 h in the Canadian global deterministic data assimilation and forecast system. The impact of polar MODIS AMVs is higher in the Antarctic region than in the Arctic region.
The impact of simulated AMVs corresponding to operational data is also positive in all regions, but significantly higher than that seen with real data, notably in the tropics, where the std error reduction is 2–3 times higher.
From simulations, the added impact of HEO AMVs in the 50°–90°N region relative to the impact of currently available data (including MODIS AMVs) is significant (1%–3%) from the surface to 300 hPa up to day 3 for the four atmospheric variables examined. The positive impact is mostly limited to the region of HEO data collection.
The impact of HEO AMVs in the 50°–90°S region is stronger by a factor of about 2 in comparison with the corresponding impact in the 50°–90°N region, and corresponds to a gain in predictability at days 3–5 of about 4 h (2 h in the Arctic region). This result is perhaps not surprising, given the paucity of conventional observations in the Southern Hemisphere. Furthermore, the positive impact of 50°–90°S AMVs extends up to 30°S in medium-range forecasts.
Validations against the experiments' own analyses and the nature run yield nearly identical results at day 3 and beyond.
To conclude, a HEO constellation dedicated to meteorological observations should contribute in a major way to the Earth Observing System. This study provides only a partial view of the potential benefits for weather prediction. High temporal radiance assimilation over the circumpolar area (hourly as opposed to ~6 h) is another promising area of exploration. The study could be pursued with sensitivity experiments linked to parameters such as thinning and assigned observation error, the inclusion of spatial error correlation effects, the use of 4DVAR assimilation, and longer periods. Such studies, based on both real and simulated data, provide significant insight with regard to the optimum assimilation of current and future data types.
Acknowledgments
This work benefited from developments at Environment Canada for a PREMIER impact study funded by the European Space Agency. We thank ECMWF and the Joint OSSE program for providing the nature run. We acknowledge the early contribution of Dr. J. W. Kaminski and Dr. R. Errico to the OSSE setup. We thank Mrs. Iriola Mati for her contribution to the preparation of the AMV data files. We also thank Jacques Hallé for his support in the modeling of PCW radiances from model output.
REFERENCES
Aparicio, J., G. Deblonde, L. Garand, and S. Laroche, 2009: Signature of the atmospheric compressibility factor in COSMIC, CHAMP, and GRACE radio occultation data. J. Geophys. Res., 114, D16114, doi:10.1029/2008JD011156.
Charron, M., and Coauthors, 2012: The stratospheric extension of the Canadian Global Deterministic Medium-Range Weather Forecasting System and its impact on tropospheric forecasts. Mon. Wea. Rev., 140, 1924–1944.
Cotton, J., 2012: Fifth analysis of the data displayed on the NWP SAF AMV monitoring website. NWP-SAF-MO-TR-027, EUMETSAT, 42 pp. [Available online at http://cimss.ssec.wisc.edu/iwwg/iwwg.html.]
Davies, R., A. Horvath, C. Moroney, B. Zhang, and Y. Zhu, 2007: Cloud motion vectors from MISR using sub-pixel enhancements. Remote Sens. Environ., 107, 194–199.
Dworak, R., and J. R. Key, 2009: Twenty years of polar winds from AVHRR: Validation and comparison with ERA-40. J. Appl. Meteor. Climatol., 48, 24–40.
Errico, R. M., R. Yang, M. Masutani, and J. Woollen, 2007: Estimation of some characteristics of analysis error inferred from an observation system simulation experiment. Meteor. Z., 16, 695–708.
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.
EUMETSAT, 2007: MTG mission requirements document. EUM/MTG/SPE/06/0011, 133 pp. [Available online at www.eumetsat.int/groups/pps/documents/document/pdf_mtg_mrd.pdf.]
Garand, L., O. Pancrati, and S. Heilliette, 2011: Validation of forecast cloud parameters from multispectral AIRS radiances. Atmos.–Ocean, 49, 121–137, doi:10.1080/07055900.2011.567379.
Gauthier, P., C. Charrette, L. Fillion, P. Koclas, and S. Laroche, 1999: Implementation of a 3D variational data assimilation system at the Canadian Meteorological Centre. Part 1: The global analysis. Atmos.–Ocean, 37, 103–156.
Hartung, D. C., J. A. Otkin, R. A. Peterson, D. D. Turner, and W. F. Feltz, 2011: Assimilation of surface-based boundary-layer profiler observations during a cool season observation system simulation experiment. Part II: Forecast assessment. Mon. Wea. Rev., 139, 2327–2346.
Heilliette, S., Y. Rochon, L. Garand, and J. W. Kaminski, 2013: Assimilation of infrared radiances in the context of observing system simulated experiments. J. Appl. Meteor. Climatol., 52, 1031–1045.
Hernandez-Carrascal, A., N. Bormann, R. Borde, H.-J. Lutz, J. Otkin, and S. Wanzong, 2012: Atmospheric motion vectors from model simulations. Part 1: Methods and characterisation as single-level estimates of wind. ECMWF Tech. Memo. 677, 31 pp. [Available online at www.ecmwf.int.]
Hoover, B., D. Santek, M. Lazzara, R. Dworak, J. Key, C. Velden, and N. Bearson, 2012: High latitude satellite derived winds from combined geostationary and polar orbiting satellite data. Proc. 11th Int. Winds Workshop, Auckland, NZ, Coordination Group for Meteorological Satellites. [Available online at http://cimss.ssec.wisc.edu/iwwg/iwwg.html.]
Kidder, S. Q., and T. H. Vonder Haar, 1990: On the use of satellites in Molniya orbits for meteorological observations of middle and high latitudes. J. Atmos. Oceanic Technol., 7, 517–522.
Lahoz, W. A., R. Brugge, D. R. Jackson, S. Migliorini, R. Swinbank, D. Lary, and A. Lee, 2005: An observing system simulation experiment to evaluate the scientific merit of wind and ozone measurements from the future SWIFT instrument. Quart. J. Roy. Meteor. Soc., 131, 503–523.
Masutani, M., and Coauthors, 2010a: Observing system simulation experiments. Data Assimilation: Making Sense of Observations, W. Lahoz, B. Khattatov, and R. Ménard, Eds., Springer-Verlag, 647–679.
Masutani, M., and Coauthors, 2010b: Observing system simulation experiments at the National Centers for Environmental Prediction. J. Geophys. Res., 115, D07101, doi:10.1029/2009JD012528.
Matricardi, M., and R. Saunders, 1999: Fast radiative transfer model for simulations of Infrared Atmospheric Sounding Interferometer radiances. Appl. Opt., 38, 5679–5691.
Moll, P., K. Karbou, C. Faccani, N. Saint-Ramond, and F. Rabier, 2012: Data impact studies in the global NWP model at Meteo-France. Proc. Fifth Workshop on the Impact of Various Observing Systems on NWP, Sedona, AZ, WMO, 1b(3). [Available online at http://www.wmo.int/pages/prog/www/OSY/Reports/NWP-5_Sedona2012.html.]
Payan, C., and J. Cotton, 2012: Collaborative satellite winds impact study. International Winds Working Group Final Rep., 47 pp. [Available online at http://cimss.ssec.wisc.edu/iwwg/iwwg.html.]
Privé, N. C., R. M. Errico, and K.-S. Tai, 2013: Validation of the forecast skill of the Global Modeling and Assimilation Office observing system simulation experiment. Quart. J. Roy. Meteor. Soc., 139, 1354–1363.
Reale O., J. Terry, M. Masutani, E. Andersson, L. P. Riishojgaard, and J. C. Jusem, 2007: Preliminary evaluation of the European Centre for Medium-Range Weather Forecasts (ECMWF) nature run over the tropical Atlantic and African monsoon region. Geophys. Res. Lett., 34, L22810, doi:10.1029/2007GL031640.
Rochon, Y., J. W. Kaminski, S. Heilliette, L. Garand, J. de Grandpré, and R. Ménard, 2012: Observation system simulation experiments for the PREMIER mission. Proc. Fifth Workshop on the Impact of Various Observing Systems on Numerical Weather Prediction, Sedona, AZ, WMO, 3c(1). [Available online at http://www.wmo.int/pages/prog/www/OSY/Reports/NWP-5_Sedona2012.html.]
Schmit, T., M. M. Gunshor, W. P. Menzel, J. J. Gurka, J. Li, and A. S. Bachmeir, 2005: Introducing the next-generation Advanced Baseline Imager on GOES-R. Bull. Amer. Meteor. Soc., 86, 1079–1096.
Shimoji, K., and M. Hayashi, 2012: Study of relationship between spatial and temporal image resolutions for AMV derivation of next generation satellites. Proc. 11th Int. Winds Workshop, Auckland, NZ, Coordination Group for Meteorological Satellites. [Available online at http://cimss.ssec.wisc.edu/iwwg/iwwg.html.]
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, 1927–1948.
Trishchenko, A. P., and L. Garand, 2012: Observing polar regions from space: Advantages of a satellite system on a highly elliptical orbit versus a constellation of low Earth polar orbiters. Can. J. Remote Sens., 38, 12–24.
Trishchenko, A. P., L. Garand, and L. D. Trichtchenko, 2011: Three-apogee 16-h highly elliptical orbit as optimal choice for continuous meteorological imaging of polar regions. J. Atmos. Oceanic Technol., 28, 1407–1422.
Velden, C., and Coauthors, 2005: Recent innovations in deriving tropospheric winds from meteorological satellites. Bull. Amer. Meteor. Soc., 86, 205–223.
Winker, D. M., M. A. Vaughan, A. Omar, Y. Hu. K. A. Powell, Z. Liu, W. H. Hunt, and S. A. Young, 2009: Overview of the CALIPSO mission and CALIOP data processing algorithms. J. Atmos. Oceanic Technol.,26, 2310–2323.
WMO, 2009: WMO vision for the GOS in 2025. Recommendation 6.1/1 (CBS-XIV), 7 pp. [Available online at http://www.wmo.int/pages/prog/www/OSY/WorkingStructure/documents/CBS- 2009_Vision-GOS-2025.pdf.]
