Added Benefit of the Early-Morning-Orbit Satellite Fengyun-3E on the Global Microwave Sounding of the Three-Orbit Constellation

Satellite observations are an important source of meteorological observation data, and direct assimilation of satellite observations has made an essential contribution to improving the timeliness and accuracy of global numerical forecasts (Eyre et al., 1994; McNally et al., 2000; Bormann, 2017; Li et al., 2022). Among numerous satellite instruments, the microwave sounder has made the most noticeable contribution toward improving numerical forecasts (Fourrié et al., 2002; Gelaro et al., 2010; Bormann, 2019; Li et al., 2021).

As early as the 1990s, the National Centers for Environmental Prediction (NCEP) of the United States and the European Center for Medium-Range Weather Forecasts (ECMWF) implemented the direct assimilation of microwave sounding unit (MSU) radiance data into the operational forecasting system (Andersson et al., 1994; Derber and Wu, 1998). Since the new generation MSUs, namely the Advanced Microwave Sounding Unit A (AMSU-A) and Advanced Microwave Sounding Unit B (AMSU-B), were first carried onboard the fifteenth satellite of National Oceanic and Atmospheric Administration (NOAA-15) in May 1998, MSUs have gradually been replaced by AMSUs.

Similar to AMSU-A/B, the microwave-sounder data onboard China’s Fengyun polar-orbiting satellites have also positively impacted numerical forecasting (Lu and Bell, 2012; Li and Zou, 2013, 2014; Li and Liu, 2016a). On 26 May 2008, Fengyun-3A (FY-3A) was successfully launched (Yang et al., 2009; Dong et al., 2009; Zhang et al., 2009) and carried China’s first-generation microwave temperature sounder (MWTS-1) and microwave humidity sounder (MWHS-1). MWTS-1 has four channels, and the frequencies of these channels correspond to channels 3, 5, 7, and 9 of AMSU-A (You et al., 2012). MWHS-1 has five channels, and its detection capability is equivalent to AMSU-B and Microwave Humidity Sounding (MHS). In 2013 and 2017, FY-3C and FY-3D were successively launched, carrying the second-generation microwave sounders, MWTS-2 and MWHS-2. MWTS-2 has 13 channels, and its performance is similar to AMSU-A (Dong et al., 2009; Zhang et al., 2009). Compared with MWHS-1, MWHS-2 has eight more temperature channels with frequencies around 118 GHz.

Many studies have assessed the data quality of MWTS/MWHS onboard the Fengyun series polar-orbiting satellites in detail (Lu et al., 2010; Guan et al., 2011; Gu et al., 2012; Lu and Bell, 2012; Wang and Zou, 2012; Li and Liu, 2016b; Qin and Zou, 2016; Han and Hou, 2020; Kan et al., 2020), and proposed many targeted quality control schemes and assimilation methods (Qin et al., 2013; Li et al., 2016, Dong et al., 2017; Lawrence et al., 2018; Duncan and Bormann, 2020). Many researchers have shown that the direct assimilation of MWTS/MWHS data can remarkably improve the analysis results of the Chinese Meteorological Administration global forecast system (CMA-GFS) and has an overall positive effect on numerical forecasts (Li et al., 2016). Recently, the FY-3D MWTS data have been applied to the CMA-GFS operational system.

However, the assimilation of polar-orbiting satellite data is limited by insufficient satellite-data coverage. One polar-orbiting satellite can provide only two observations daily for a fixed point, but the operational global data assimilation system conducts at least one six-hour interval of assimilation per day.

The local equator crossing time (LECT) of existing operational polar-orbiting meteorological satellites is concentrated around 1000 local solar time (LST; mid-morning satellites) or 1400 LST (afternoon satellites). For mid-morning satellites, there are Meteorological Operational satellites A/B (MetOp-A/B), and for afternoon satellites, there are NOAA-18, NOAA-19, Suomi National Polar-Orbiting Partnership (Suomi-NPP), and other satellites. Among the four polar-orbiting satellites launched by China, FY-3A/C are mid-morning satellites, while FY-3B/D are afternoon satellites. Within the six-hour assimilation window, there are always satellite observation gaps in the case of using only the available mid-morning and afternoon satellites. The shortcomings of insufficient data coverage are particularly evident when forecasting fast-moving extreme weather.

The justification for having at least three operational polar-orbiting satellites rather than two has been supported by many numerical weather prediction (NWP) impact studies over the last decade (Eyre and English, 2008). The Further Observing System Experiment (OSE) studies have shown the benefits of the early morning orbit using the NOAA-15 AMSU-A microwave-sounding instrument as a proxy. (Di Tomaso and Bormann, 2011; Zou et al., 2016). Other scientists have used the Observing System Simulation Experiment (OSSE) system of the USA Joint Center for Satellite Data Assimilation to carefully evaluate the value of the instrument payload located in the early morning satellite in numerical prediction (Riishojgaard et al., 2012).

Fengyun-3E (FY-3E), China’s first early-morning polar-orbiting satellite, was successfully launched by China on 5 July 2021. It is the fifth satellite in the Fengyun-3 series, whose descending orbit crosses the equator at 0540 LST. It effectively supplements the gap of polar-orbiting satellite observations in the 6-hour assimilation window, making up for the shortage of global observation data (Zhang et al., 2022). Not only does the satellite have a different operating orbit, FY-3E also carries the microwave temperature sounder, MWTS-3, which is more advanced than that of FY-3D, as well as the improved MWHS-2. The data evaluation results show that the error characteristics and various types of noise inherent to the MWTS-3 and MWHS-2 data are generally less than those onboard FY-3D, which can meet the requirements of data assimilation applications (Mao et al., 2022; Qian et al., 2022).

In this study, the assimilation module for FY-3E microwave-sounding data is established and added to the CMA-GFS system that only assimilates the microwave-sounding data of mid-morning and afternoon satellites. The effect of adding the early-morning satellite into the three-orbit polar-orbiting system is evaluated by comparing the results of the one-month cycle assimilation and forecast results. It should be noted that after September 2021, China’s operational numerical forecasting system was officially renamed CMA-GFS (Chen et al., 2008; Xue et al., 2008).

The remainder of this paper is organized as follows. The general details of the FY-3E MWTS-2 and MWHS-2 radiance observations are described in section 2. Section 3 demonstrates the change in the coverage ratio caused by adding observation data of FY-3E MWTS-3 and MWHS-2. Section 4 introduces the CMA-GFS four-dimensional variational assimilation (4D-Var) system. The assimilation experiment setup, data, and preprocessing scheme are presented in section 5. The preliminary quality assessment of FY-3E MWTS-3 and MWHS-2 is also introduced there. The numerical results are shown in section 6. Finally, conclusions and discussions are presented in section 7.

FY-3E MWTS-3 and MWHS-2 have 17 and 15 channels, respectively, and can detect the atmospheric temperature and humidity information from the ground to the upper atmosphere. Both instruments have the same 2 700-kilometer-wide scanning track, which is wider than that of MWTS-2 (2250 km) and other similar instruments such as the AMSU-A (2300 km), Advanced Technology Microwave Sounder (ATMS) (2500 km) and MHS (2180 km). The field of view (FOV) of a single MWTS-3 scan line is increased to 98, which is eight more than that of MWTS-2 and also higher than that of AMSU-A (30) and ATMS (96). The FOV of a MWHS-2 single scan line is also 98, slightly more than that of MHS (90).

To meet the needs of data assimilation, two channels with frequencies of 23.8 and 31.4 GHz have been added to MWTS-3 for the first time. The data of these two channels can be used to retrieve cloud liquid water path (CLWP) and serve as the basis for cloud detection. In addition, two vertical channels are added, namely channel 6 (53.246 ± 0.08 GHz) and channel 8 (53.948 ± 0.081 GHz), thus making MWTS-3 more capable of vertical detection.

The MWHS-2 onboard FY-3E slightly differs from the MWHS-2 onboard FY-3C and FY-3D. Its window channel frequency is changed from 150 GHz to 166 GHz. In addition, except for channel 2, the sensitivity of each channel has nearly doubled, and the calibration accuracy has also been improved (Zhang et al., 2022). Compared with the MHS and the water vapor channels of ATMS, the MWHS-2 has more high-frequency temperature-sounding channels with frequencies around 118 GHz, which is a unique feature of MWHS-2.

Based on the transmittance coefficients prepared by the National Satellite Meteorological Center of CMA, the weighting functions (WF) of the FY-3E MWTS-3 and MWHS-2 are calculated by the Radiative Transfer for TIROS Operational Vertical Sounder (TOVS) v12 (RTTOV-12) based on the US standard atmospheric profile, as shown in Fig. 1. The spectral range of MWTS-3 is 23.8–57 GHz. It can be seen from the WF distribution (Fig. 1a) that the peak WFs of channels 1–4 are mainly located on the ground, and other channels are above the ground, which can detect the atmospheric temperature at different heights. Channel 17 has the highest peak value, up to about 2 hPa. The newly added channels 1 and 2 mainly detect the water vapor content. The peaks of the WFs of the new channels 6 and 8 are 700 hPa and 500 hPa, respectively. Channel 3 is a window channel.

Figure 1.  WFs of (a) FY-3E MWTS-3 and (b) MWHS-2 calculated using the U.S. standard atmosphere profile.

This study assesses the impact of adding the FY-3E data into the three-orbit operation system. The MetOp-B is selected as the mid-morning satellite, and the FY-3D as the afternoon satellite. Figure 2 shows the spatial distribution of the MetOp-B AMSU-A, FY-3D MWTS-2, and FY-3E MWTS-3 observations from 2300 UTC 23 September to 0300 UTC 24 September, and during 0300–0900 UTC, 0900–1500 UTC, and 1500–2100 UTC 24 September 2021. As can be seen, the scanning area of the FY-3E microwave data can better cover the mid-latitude blank area that cannot be covered by MetOp-B and FY-3D. For the analysis time of 0000 UTC (Fig. 2a, assimilation window is 2100 UTC 23–0300 UTC 24) and 1200 UTC (Fig. 2c, assimilation window is 0900–1500 UTC), observations from the FY-3E mainly cover the blank area in 120°–70°W and 60°–100°E. While for the analysis time of 0600 UTC (Fig. 2b, assimilation window is 0300–0900 UTC) and 1800 UTC (Fig. 2d, assimilation window is 1500–2100 UTC), the blank area to be filled is in 120°E–120°W and 40°W–40°E.

Figure 2.  Spatial distribution of MetOp-B AMSU-A (green), FY-3D MWTS-2 (blue), and FY-3E MWTS-3 (red) observations during (a) 1800 UTC 23–0300 UTC , (b) 0300–0900 UTC, (c) 0900–1500 UTC, and (d) 1500–2100 UTC 24 September 2021.

The added percentage of data coverage by FY-3E MWTS-3 and MWHS-2 is calculated based on data from 14 to 23 September 2021. The statistical method divides the globe into 1° × 1° grids. For each assimilation window, the percentage of the number of grids with only FY-3E data to the number of global grids is regarded as the added percentage. On average, MWTS-3 and MWHS-2 can increase the coverage of polar-orbiting satellite data by 14.8% and 10.58%, respectively. Since the FOV number of MHS is much more than that of AMSU-A and ATMS, the improvement of FY-3E MWTS-3 is more obvious than that of FY-3E MWHS-2.

The CMA-GFS 4D-Var system is the analysis system that has been operationally used in China since 1 July 2018 (Zhang et al., 2019). The horizontal resolution of the CMA-GFS 4D-Var system is 0.25º × 0.25º. The model top is approximately 0.1 hPa, with 87 vertical layers. The RTTOV-12 is used to simulate the satellite radiance (Saunders et al., 1999) in CMA-GFS. Currently, the background covariance matrix used in this study is calculated using the NMC (National Meteorological Center) method (Parrish and Derber, 1992; Wu et al., 2002).

The CMA-GFS system conducts global 4D-Var assimilation four times daily, with 6-h assimilation windows starting at 0300, 0900, 1500, and 2100 UTC, generating a 0.25° × 0.25° global analysis at the beginning of each assimilation window. After that, the analysis field is used as the initial condition, and a 3-hour forecast is made to generate the initial condition of the model at the start time of the four operational forecasts.

Currently, the CMA-GFS can directly assimilate data from the following sources: radiosondes, surface synoptic observations, ship observation data, aircraft reports, Atmospheric Motion Vectors (AMVs), NOAA-15/18/19 AMSU-A, and MHS, MetOp-A/B AMSU-A, MHS, and IASI (Infrared Atmospheric Sounding Interferometer), Suomi-NPP ATMS, FY-3C/D MWHS-2 and Micro-Wave Radiation Imager (MWRI), FY-3D MWTS-2 and Hyperspectral Infrared Atmospheric Sounder-2 (HIRAS-2) radiance data, and the FY-3C/D GNSS Radio Occultation Sounder (GNOS).

Four groups of experiments are conducted. The control experiment (CTRL) assimilates the conventional observation data, the MetOp-B AMSU-A/MHS, and the FY-3D MWTS-2/MWHS-2. The conventional observations include surface and upper-air reports, such as radiosondes, surface synoptic observations, ship and aircraft reports, and AMVs from the Global Telecommunications System. The setting of the sensitivity experiment (TEST) is the same as the CTRL but with the addition of FY-3E MWTS-3/MWHS-2 radiance data.

To further clarify the value of FY-3E data to operational forecasting, we also added a second group of experiments, consisting of the operational forecasting results called CTL2, and an experiment that adds FY-3E microwave detection data to the operational forecasting system, which we called TEST2.

The dynamic framework of the forecast model adopts a set of non-hydrostatic fully compressible equations (shallow atmosphere approximation), a three-dimensional reference atmosphere, a predictor-corrected semi-implicit semi-Lagrangian algorithm, a three-dimensional vector discretization of the momentum equation plus W damping, a vertical non-interpolation SISL (a two-time layer Semi-Implicit, Semi-Lagrangian algorithm) to solve the potential temperature prediction equation, and a high-precision conservative scalar advection PRM (Piecewise Rational Method). The physical processes used include a two-parameter cloud physics scheme, cloud macro physics and forecast cloud scheme, a NSAS (New Simplified Arakawa-Schubert) cumulus convection parameterization, RRTMG (rapid radiative transfer model for general circulation models) short-wave radiation, CoLM (Common Land Model) land surface process, NMRF (New Medium-Range Forecast) boundary layer parameterization, sub-grid terrain gravity wave drag and small-scale terrain turbulence drag. Other detailed information about the forecast system can be found in Shen et al. (Shen et al., 2020).

The period of the rolling experiment is from 24 September to 25 October 2021. The CMA-GFS forecast system makes four forecasts per day, with the forecast starting times of 0300, 0900, 1500, and 2100 UTC, noting that a 240-h forecast is made each time. The initial condition is provided by its data four-dimensional assimilation system. The assimilation window of the assimilation system is six hours. For example, the initial condition of 0300 UTC assimilates all conventional and satellite observations in the 0300–0900 UTC period. A total of 128 240-h forecasts have been made, and all analysis results in this study are based on these forecasts.

The considerable uncertainties of the background surface temperature and the surface emissivity in the microwave band lead to large errors in the brightness temperature simulation of near-ground channels. Therefore, this study does not use the data from those near-ground channels (channels 1–5). Furthermore, considering that CMA-GFS has a relatively large bias from 10 hPa to the model top (compared with ERA5 reanalysis, the average bias at 3 hPa is about 0.7 K), the data from those high-level channels (channels 16-17) are not used. In addition, MWTS channels 6–8 are also not used in this study due to the problems described in Qian et al. (2022). In summary, this study only assimilates channels 9–15 of the FY-3E MWTS-3 and channels 2–6, 11, and 12 of the MWHS-2.

For the data observed by the MetOp-B and FY-3D satellites, this study follows the original setting in the CMA-GFS system. The data observed by AMSU-A and MHS onboard MetOp-B and the MWTS-2 and MWHS-2 onboard FY-3D have been assimilated operationally into the CMA-GFS system. In this evaluation, we only assimilate the data from channels 5–14 of the AMSU-A, channels 11–13 of the MHS, channels 7–13 of the MWTS-2, and channels 2–6 and 11–12 of the FY-3D MWHS-2.

Satellite observations are susceptible to cloud contamination. Although microwave radiation can penetrate most non-precipitating clouds, large water and ice particles can still scatter or absorb microwave radiation. Due to the insufficient simulation effect of the location and internal particle structure of clouds, the simulation error of the microwave data in the cloud area is significantly larger than that in the clear sky. Various technical solutions have been internationally built to assimilate the cloud-affected observations from microwave temperature/humidity sounders and microwave imagers (Geer and Bauer, 2010; Geer et al., 2018; Li et al., 2020; Duncan et al., 2022). However, for microwave sounders with frequencies in the 50–60 GHz range, the current method of the CMA-GFS assimilation system is to exclude the cloudy observation data. Therefore, we are also developing all-weather data assimilation methods for satellite data based on the CMA-GFS operational system.

Two channels, sensitive to water content, have been added to the MWTS-3 for the first time, which have been used internationally to retrieve CLWP for non-precipitating clouds over ocean areas (Weng and Grody, 1994). According to Weng et al. (2003), The CLWP can be calculated by observed brightness temperatures of the two AMSU window channels at 23.8 ($ {T}_{\mathrm{b}23}) $and 31.4 GHz ($ {T}_{\mathrm{b}31}) $ and the sea surface temperature (T_s). The specific calculation formula is:

where

Here, μ represents the cosine of the local zenith angle, τ_o is the optical thickness, ε represents the sea surface emissivity, $ {\kappa }_{\mathrm{v}} $ and $ {\kappa }_{\mathrm{l}} $ are the absorption coefficients of water vapor and cloud liquid water, respectively. For detailed information about parameter specifications, please refer to Weng et al. (2003).

In this study, the detection threshold is specified as 0.02 kg m⁻². Figure 3 shows the spatial feature of the CLWP retrieved from the MWTS-3 data. The blue and green dots on both sides of the figure represent the MetOp-B AMSU-A and FY-3D MWTS-2 pixels, respectively, and the red dots are for the MWTS-3 data over the ocean with a retrieved CLWP less than 0.02 kg m⁻². To evaluate the correctness of cloud detection results, the reflectance data (shading in Fig. 3) of channel 7 (12 μm) of the Medium Resolution Spectral Imager-Low Light onboard FY-3E is selected to represent the spatial distribution of clouds. The comparison shows that the detected clear-sky data are mainly distributed in the high-reflectance area. In contrast, the MWTS-3 data in the low-reflectance area have been removed, which shows that the cloud detection method is generally effective.

Figure 3.  Spatial distribution of FY-3E MWTS-3 clear pixels (red), MetOp-B AMSU-A pixels (blue), FY-3D MWTS-2 pixels (green), and the brightness temperature (units: K) of MERSI channel 7 during 0300–1500 UTC 24 September 2021.

Over the land, the window channel’s O-B is used to perform the cloud detection for the MWTS-3. When the O−B of the window channel of a pixel is greater than 1.5 K, it is considered to be a cloudy pixel of MWTS-3.

For the FY-3E MWHS-2, the cloud ice water path (IWP) is retrieved using the observed brightness temperatures of the two AMSU channels with frequencies of 89.0 and 166 GHz, respectively (Weng et al., 2003). The formula is as follows:

where Ω is the scattering parameter, μ represents the cosine of the local zenith angle, $ {\rho }_{i} $ is the bulk volume density of particles, mostly a given constant, and $ {D}_{\mathrm{e}} $ is the particle effective diameter.

The scattering parameter Ω is decided by cloud single-scattering albedo ($ \omega $), asymmetric factor ($ g $), and optical thickness (τ) as follows:

The effective particle diameter $ {D}_{\mathrm{e}} $ can be calculated using:

where ${\varOmega }_{\mathrm{N}}$ is the normalized scattering parameter, which only depends on the effective particle size parameter $ {x}_{\mathrm{e}} $, and the complex refractive index $ m $. For the specific parameters, please refer to Weng et al. (2003).

When the IWP is greater than 0.02 kg m^–2, it is determined that this data is in a cloudy area. Over land, when the O−B of the window channel 10 is greater than 1.5 K, it is identified as a cloudy pixel. In the assimilation system, channels 9 and 10 only retain clear-sky pixels.

For each FOV of the MWTS-3 data, when it is identified as the FOV on the cloud, only the data of channels 11–15 are used because the peaks of their WFs are located in the upper troposphere, and the influence of clouds can be ignored. Since the standard deviation of the O−B over sea ice is much greater than that over the ocean, the channel 9 data over the sea ice is excluded. The sea ice area is defined as the area with sea surface temperatures lower than 271.45 K. For the same reason, the MWTS-3 data of channels 9 and 10 over high land (terrain higher than 500 m) are also excluded. Because of the remarkable decrease in the simulation accuracy caused by the limb effect, the outermost 10 FOVs (i.e., FOVs 1–10, 89–98) on each scan line are not used for all channels. The increase of orbital width further improves the data coverage of FY-3E, but due to the uncertainty of observation error in the orbital edge, the impact of the increased orbital width on the assimilation effect of FY-3E microwave-sounding data needs further investigation.

For MWHS-2 data, the following quality control (QC) steps are conducted: (i) removing the observations of channels 5−6 and 11−12 if they are identified as cloudy data; (ii) removing observations with a FOV covering the coastline; (iii) excluding the data of the 10 outermost FOVs; (iv) for channels 5−6 and 11−12, excluding the data over sea ice or land; (v) when the terrain is higher than 500 m, excluding the data of channel 5.

For the MWTS-3 and MWHS-2 data, the data that pass QC are thinned to a horizontal interval of 120 km. It means that only one observation is reserved in every 120 km range.

Many studies have found that satellite data tends to contain two types of systematic bias. One is the bias caused by the scanning angle, and the other is the bias that depends on the air mass attributes. In the early stages, the bias was corrected for each scanning angle separately (Harris and Kelly, 2001), which was proven to be an effective correction scheme by operational applications (Derber and Wu, 1998; McNally et al., 2000). However, the static bias correction method has difficulty addressing the bias drifts with the weather system. Therefore, air mass and variational bias correction methods have been established (Derber and Wu, 1998; Dee, 2004, 2005). After selecting appropriate bias prediction factors, the correction coefficient is obtained by statistical methods or updated in the data assimilation process to achieve the dynamic bias correction.

Currently, the bias correction scheme used in the CMA-GFS system includes two steps. The first step is the scanning bias correction, in which the systematic bias caused by the scanning angle difference from the nadir measurement is subtracted from each observation. The second step addresses the air mass bias correction. In this study, two prediction factors are selected for the air mass bias correction of MWTS-3 data, which include the background atmospheric thicknesses of the 1000–300 hPa and 200–50 hPa layers. Regarding the MWHS-2 data, four prediction factors are adopted: the atmospheric thicknesses of the 1000–300 hPa, 200–50 hPa, and 50–10 hPa layers, and the atmospheric water vapor content in the 1000–300 hPa layer. The correction coefficients are calculated from the observations two weeks before the data assimilation time. Then, the air mass bias can be calculated and subtracted from the data by combining the prediction factors and the corresponding coefficients in the assimilation.

Figure 4 gives the O-B mean value and standard deviation of the FY-3E microwave-sounding data before and after bias correction. Before correction, the biases of those selected MWTS-3 channels are basically within ±2.0 K, and the biases of channels 10 and 15 are slightly larger. After correction, the bias of each channel is close to zero. The bias correction also reduces the standard deviation. All STDs are reduced to less than 1 K, with the most remarkable reductions in channels 9, 10, and 15. The detection ranges of channels 11–14 are in the mid-upper levels with relatively good background accuracy, and the O−B standard deviations are also lower than those of other channels, all within 0.5 K.

Figure 4.  (a) Bias and (b) STD of the O-B for FY-3E MWTS-3 channels from 24 September–3 October 2021. Panels (c)–(d) are similar to (a)–(b), but for MWHS-2.

The frequencies of the FY-3E MWHS-2 channels 2–6 are around 118 GHz, and their peak WFs are all above 300 hPa. These channels have a remarkable negative bias (over −1.5 K). Channel 2 has the highest peak WF (about 20 hPa), and its O-B bias is the largest among all channels, lower than –3 K. The frequencies of channels 11 and 12 are close to the water vapor absorption band, with a slight bias between ±0.5 K. However, the O-B STDs of these channels are obviously greater than those of the 118 GHz channels. Overall, the O-B STD of all channels is about 2 K. After the bias correction, the bias of each channel is near zero. The O-B STDs of channels 2–6 are within 1.2 K, and those of channels 11 and 12 are within 1.8 K.

After bias correction, the observations with O-B greater than twice the observation error are also removed. According to the results of Qian et al. (2022), the observation errors of MWTS-3 channels 9, 10, and 12–14 are all set to 0.55 K, and the errors of channels 11 and 15 are 0.4 and 1.1 K, respectively. As for MWHS-2, the error of channel 2 is 1.4 K, the errors of channels 3–6 are 0.8, 0.6, 0.5, and 1.0 K, respectively, and the errors of channels 11–12 are both 2 K.

Figure 5 shows the spatial distribution of O−B and the difference between observation and analysis (O−A) for the FY-3E MWTS-3 channel 11 and FY-3E MWHS-2 channel 6 during 0300–1500 UTC on 24 September 2021. For comparison, the MetOp-B AMSU-A and FY-3D MWTS-2 data are also shown in small and big grey dots, respectively. As can be seen, after data assimilation, the O−As are all within ± 0.3 K and are generally smaller than the O−Bs, indicating an improved analysis. The results prove that the data assimilation is effective, and the observation information of the FY-3E MWTS-3 and MWHS-2 can be well integrated into the analysis results.

Figure 5.  Spatial distribution of (a, c) O–B and (b, d) O–A for (a, b) channel 11 of MWTS-3 and (c, d) channel 6 MWHS-2 from FY-3E during 0300–1500 UTC on 24 September 2021. Pixels of MetOp-B AMSU-A (light blue dots) are overlaid with FY-3D MWTS-2 (pink dots) for comparison.

To investigate the accuracy of the analysis results, the root mean square error (RMSE) is calculated for the 500-hPa geopotential height difference between the analysis fields and the ERA5 reanalysis data (Fig. 6). The RMSE is reduced by the additional assimilation of the FY-3E microwave data at each assimilation time; this is more obvious in the southern hemisphere and tropical regions. The most apparent reduction of RMSE can be as large as 10% (Fig. 6a). The assimilation of FY-3E microwave data in the northern hemisphere has a relatively neutral effect, which may be due to the large amount of available conventional observation data.

Figure 6.  The daily RMSE of geopotential height analysis differences between CTL and ERA (black), TEST and ERA (red) in the (a) Southern Hemisphere, (b) Northern Hemisphere, and (c) Tropics at 500 hPa from 24 September–25 October 2021.

Figure 7 shows the RMSEs for the wind speed in the southern hemisphere and in the northern hemisphere from 24 September to 25 October 2021. The conclusion is similar to that of Fig. 6. After assimilating the FY-3E microwave-sounding data, the RMSE of wind speed at each height decreases, especially in the southern hemisphere. The most obvious reduction of RMSE is near 300 hPa in the southern hemisphere, up to about 5%.

Figure 7.  The difference in RMS of U-wind analysis between CTL and ERA (black), TEST and ERA (red) in the (a) Southern and (b) Northern Hemisphere from 24 September–25 October 2021. Panels (c) and (d) are for the V-wind.

The anomaly correlation coefficient (ACC) of the 500-hPa geopotential height is an important parameter to measure the forecast level of the medium-range numerical forecast system. The corresponding forecast time with the ACC greater than 0.6 is defined as the predictable time. Figure 8 shows the mean ACC of the geopotential height at 500 hPa for CTL and TEST experiments from 24 September to26 October 2021. As shown in Fig. 8, the ACC of the TEST experiment is bigger than that of the CTL nearly all of the time, and the predictable time has been increased by 6–7 hours, which is about 3% of the predictable time of the CTRL. From the comparison of operational experiments (Figs. 8c–d), it can be seen that adding FY-3E microwave-sounding data can improve the ACC of the current operational forecasting system, especially in the southern hemisphere, which experiences a significant positive impact from the 120-hour forecast, demonstrating statistical significance since the 168-hour forecast.

Figure 8.  Mean ACC of the 500-hPa geopotential height for (black) CTL/CTL2 and (red) TEST/TEST2 experiments for the (a, c) Northern and (b, d) Southern Hemisphere, from 24 September to 25 October 2021.

Figure 9 is for the comprehensive scorecard. The figure shows the ACCs and RMSEs of 10-day forecasts at different levels for multiple regions (the northern hemisphere, the southern hemisphere, East Asia, and the tropics) and multiple physical quantities (geopotential height, potential temperature, zonal and meridional wind) in the world. The red symbols indicate that the forecasts of TEST are better than those of the CTL, and the big red triangle indicates remarkable improvement. The green symbols represent the opposite result. From the comprehensive scorecard, the FY-3E microwave data makes a remarkable positive contribution to the forecasts in the southern hemisphere and tropics and a slight positive contribution to the forecasts of the northern hemisphere and East Asia.

Figure 9.  The score card for TEST against CTL.

As China’s first early-morning-orbit satellite, FY-3E effectively complements the gaps within the polar-orbiting satellite observation. Based on the CMA-GFS 4D-Var assimilation system, this study assesses the added value of the microwave sounders onboard FY-3E to the three-orbit constellation for polar-orbiting satellites.

The results show that the FY-3E microwave-sounding data can positively contribute to the numerical forecast after effective preprocessing. Compared with the data coverage obtained solely using mid-morning and afternoon satellites, the FY-3E MWTS-3 data can improve the spatial coverage of microwave temperature sounding data by about 14.8%, and the MWHS-2 can increase that of the microwave humidity sounding data by about 10.6%. The analysis of assimilation results and forecasts shows that based on the MetOp-B, AMSU-A/MHS, and FY-3D MWTS-2/MWHS-2, the additional assimilation of FY-3E microwave sounding data can reduce the error of geopotential height, potential temperature, wind speed and specific humidity in the analysis field. The ACC and RMSE of the 500-hPa geopotential height forecasts have also been improved.

It is noteworthy that only the conventional data and the microwave-sounding data of the three polar-orbiting satellites are assimilated in this study. In the following study, more operational assimilated observations are to be added. At the same time, the contribution difference between the MWTS-3 data and the MWHS-2 data will be further investigated. Moreover, more channels, especially the two newly added channels, will be included to further improve the impact of the FY-3E microwave-sounding data assimilation in the CMA-GFS operational system.

Acknowledgements. The authors want to acknowledge the National Satellite Meteorological Centre of the CMA for providing the satellite data. This study was jointly supported by the National Key Research and Development Program of China (Grant No. 2022YFC3004002), the Fengyun Application Pioneering Project (FY-APP-2021.0201) and FY-3 Meteorological Satellite Ground Application System Project [FY-3(03)-AS-11.08].

Data availability statement. Satellite data sets and data for WFs are available at https://pan.baidu.com/s/1k4Ed5ormHexjLbV5ZN8AbA?pwd=7yc9. (for Figs. 1, 2, and 4). ERA5 reanalysis data can be obtained from https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5. The analytical and forecast results for this study have been uploaded to https://pan.baidu.com/s/1W4QDvm2gZV1zC4W5LvFECg?pwd=7wt5. Some of the above links may require a sign-in, which can be obtained free of charge.