CRA-40/Atmosphere—The First-Generation Chinese Atmospheric Reanalysis (1979–2018): System Description and Performance Evaluation

Atmospheric reanalysis reproduces the past atmos-pheric conditions through the assimilation of historical meteorological observations and the fixed version of a numerical weather prediction (NWP) model and a data assimilation (DA) system. Since the mid-1990s, several agencies (mainly NOAA, NASA, ECMWF, and JMA) have successively conducted four rounds of global atmos-pheric reanalysis as listed in Table 1. Note that ECMWF named their latest reanalysis as ERA5 (i.e., the 5th gene-ration) with consideration of the FGGE reanalysis (Beng-tsson et al., 1982). These reanalysis products have been widely used in weather, climate, and related research and applications (e.g., Trenberth et al., 2011; Serreze et al., 2012). They are also used increasingly for business applications such as in energy (e.g., Cannon et al., 2015) and agriculture (e.g., Toreti et al., 2019) sectors. Most of these reanalysis systems are continuously running as operational “climate data assimilation system (CDAS)^①” for climate monitoring. More recent reanalysis products usually have been produced with more advanced NWP models and DA systems at higher resolutions and with more optimal use of observations.

In November 2013, China Meteorological Administration (CMA) launched the global reanalysis project, which is led by the National Meteorological Information Centre (NMIC) with participation of several other national centers of CMA and also external research institutes and universities. The goal was to produce China’s first-generation global atmospheric and land reanalysis (RA) products for a 40-yr (1979–2018) period (so this reanalysis is named CRA-40; here CRA actually refers to CMA-RA) and then the reanalysis system will become operational as a CDAS to support climate monitoring activities at the National Climate Centre (NCC) of CMA.

At the time of proposing CRA-40, the CMA-developed GRAPES-GFS was not mature and the CMA’s operational global NWP system was based on an old version of ECMWF’s spectral model configured at T_L639 spectral resolution (~31 km) and NCEP’s Gridpoint Statisti-cal Interpolation (GSI) DA system. The original plan was to use operational T639 + GSI at that time for CRA-40 atmospheric reanalysis. However, during the early stage of project execution it was found that the T639 model exhibited significant temperature bias near the model top, which greatly limited the optimal use of satellite radiance data. GRAPES-GFS began operational on 1 June 2016 with a 3DVar DA scheme (Wang et al., 2017) and was then upgraded to 4DVar on 1 July 2018 (Zhang et al., 2019). However, GRAPES-GFS’s DA system cannot assimilate satellite radiance data from old satellite sensors and relatively low model top (36 km) of GRAPES-GFS also restricts the optimal assimilation of satellite radiances. Therefore, it was decided to replace the T639 model with NCEP’s publically available Global Spectral Model (GSM), while keeping GSI as atmospheric DA system for CRA-40.

An interim version of GFS/GSI-based CRA (CRA-Interim) for a 10-yr period (2007–2016) at a 6-h time interval and a T_L574 spectral resolution (34 km) was produced in February 2018. A 10-yr (2007–2016) CRA-Interim/Land dataset with the same time and spatial (34-km) resolutions as CRA-Interim was also produced by forcing Noah land surface model using observation blended CRA-Interim precipitation and near surface temperature, pressure, humidity, wind, and radiation directly from CRA-Interim (Liang et al., 2020). Evaluation of CRA-Interim shows some advantages in terms of clouds (Yao et al., 2020a), precipitation (Li et al., 2020), upper troposphere water vapor (Xue et al., 2020), and the ability of depicting global energy cycle (Zhao et al., 2019). In late 2019, CRA-40/Atmosphere and CRA-40/Land were completed. This article documents CRA-40/Atmosphere. Section 2 describes the model and DA system used in CRA-40 as well as the production streams. Section 3 provides some details about the observations assimilated in CRA-40. Section 4 presents the DA diagnostics of CRA-40, followed by the evaluation of some aspects of the products in Section 5. The paper is concluded in Section 6.

As mentioned earlier, CRA-40 adopted NCEP’s GSM and the community GSI DA system for atmospheric reana-lysis. Configurations of GSM model and GSI DA system will be briefly described in the following subsections.

The atmospheric prediction model used by CRA-40 is GSM version 14, operational at NCEP from July 2017 to June 2019. The GSM-v14 was made publicly available by the NCEP’s Central Operation (NCO). This is the last operational implementation of GSM before the transition to FV3-based GFS at NCEP in June 2019. GSM-v14 was configured with T_L1534 spectral resolution (~13-km grid spacing at the equator) at the NCEP’s operation, but CRA-40 configured it at a reduced spectral resolution T_L574 (~34-km grid spacing at the equator) to lower the computational cost. In the vertical, there are 64 hybrid sigma pressure layers (Sela, 2009) with the model top around 0.27 hPa (~55 km). The GSM dynamic core adopts a two time-level semi-implicit and semi-Lagrangian time discretization (Sela, 2010). The semi-Lagrangian advection calculation and physics processes are performed on a linear, reduced Gaussian grid in the horizontal domain.

GSM’s prognostic variables consist of vorticity, divergence, logarithm of surface pressure, specific humidity, virtual temperature, mixing ratios of cloud liquid water and cloud ice (Zhao and Carr, 1997), and O₃ (McCormack et al., 2006) (see https://dtcenter.ucar.edu/GMTB/v4.1.0/sci_doc/GFS_OZPHYS.html). GSM-v14 replaced spectral file output (sigma files) with new NEMSIO (https://www.emc.ncep.noaa.gov/emc/pages/infrastructure/nems.php) binary files on model native grid. More details about physical parameterization schemes of GSM and specific changes in GSM-v14 can be found at https://www.emc.ncep.noaa.gov/emc/pages/numerical_forecast_systems/gfs/documentation.php. SST and sea ice are important lower boundary conditions for any atmospheric reanalyses. CRA-40 uses daily SST and sea ice data (0.25° × 0.25°) from CFSR for the period from 1979 to 2014. As CFSR did not continue after December 2014, we use NCEP operational daily Real-Time Global SST (RTG SST, 8 km × 8 km) and sea ice analysis (0.5° × 0.5°) for more recent years from 2015 to 2018.

The community version 3.6 of GSI (Wu et al., 2002; Kleist et al., 2009; Shao et al., 2016) was adopted for the final production of CRA-40/Atmosphere (http://www.dtcenter.org/community-code/gridpoint-statistical-interpolation-gsi/community-gsi-version-3-6-enkf-version-1-2), which was released in September 2017 by the Developmental Testbed Center (DTC). While more advanced DA techniques such as hybrid-3DEnVar or hybrid-4DEnVar are available in GSI, CRA-40 used 3DVar for the sake of its simplicity and computational efficiency. This choice is also constrained by the requirement that the production needs to be completed within a relatively short time period (several months) for all 40-yr products.

The background and analysis files of the GSI-3DVar contain the grid point fields (instead of spectral coefficients) at a 34-km horizontal resolution and in the NEMSIO format. The analysis was carried out at each 0000, 0600, 1200, and 1800 UTC with assimilation of conventional observations, satellite retrievals, and satellite radiance data (see more details in Section 3). GSI-3DVar was run in the FGAT mode, in which assimilation time window is 3-h centered at analysis times and the three background files at −3-, 0-, and 3-h come from a 9-h GSM forecast initialized from the previous cycle’s analysis. Variational bias correction (VarBC) was applied to clear-sky satellite radiances (Zhu et al., 2014) and aircraft temperature data (Zhu et al., 2015). However, cloudy AMSU-A radiance assimilation capability (Zhu et al., 2016) available in GSI version 3.6 was not utilized. For radiance DA, version 2.2.3 of the Community Radiative Transfer Model (CRTM) was used. Channel selection of satellite radiances for various sensors over the reanalysis period is based upon CFSR, which is also a GSI-based reanalysis for its atmospheric component.

GSI observation interface was modified to allow the assimilation of atmospheric motion vectors (AMVs) from Chinese FY-2 geostationary satellites, and GNSS radio occultation bending angle data and MWHS-2 microwave radiances (Jiang et al., 2020) from Chinese FY-3C polar-orbiting satellite. Another important development within GSI is to interface the model level ERA-Interim reanalysis data so that GSI can serve as a pre-evaluation tool for observations by comparing ERA-Interim with observations over the 40-yr period. The addition of this functionality in GSI and the observation pre-evaluation step prior to the reanalysis production greatly helped to implement CRA-40’s own quality control and blacklisting strategy for conventional observations, which is approved to be crucial for the success of CRA-40.

The analysis variables of GSI consist of stream function, unbalanced part of velocity potential, temperature and surface pressure, and normalized relative humidity as well as O₃ with the background error covariance model following Wu et al. (2002). The GSI’s static background error covariance statistics file of CRA-40 is taken from the NCEP’s operational implementation of GFS-v14 and is not changed over the 40-yr reanalysis period. Note that the inner loop minimization of GSI hybrid-4DEnVar incremental analysis at the NCEP’s GFS-v14 operation was done at T574 resolution (so the resolution of static background error covariances) even though the GSM-v14’s forecast background is at T1534 resolution. Observation errors were configured with relevant files in the community GSI package (Hu et al., 2017). It is worth mentioning that temperature and humidity at 2 m and zonal and meridional winds at 10 m are model diagnostic variables and are not analyzed by GSI. The analysis of those screen-level variables will come from CRA-40/Land.

CRA-40 covers a 40-yr period from January 1979 to December 2018. The CRA-40 production is divided into 10 parallel streams with a 4.5-yr period for each stream. This enables the whole 40-yr production being completed within one year as required by the project. There is a 6-month overlap between two adjacent streams and the first six months’ data of each stream are discarded so that different streams’ data can be joined together without discontinuity. For instance, stream-10 is run from 1 July 2014 to 31 December 2018. However, only the data between 1 January 2015 and 31 December 2018 in stream-10 are kept for the final delivered products and the data between 1 July 2014 and 31 December 2014 in the final products are actually from stream-9, which is run from 1 July 2010 to 31 December 2014. This multiple stream production strategy is common practice for most reanalysis projects even though the length of each stream and overlap period can be different among different reanalyses. Due to computational and storage cost, no longer range forecasts are carried out. Model level analysis is postprocessed into pressure level products by using UPP (https://dtcenter.org/community-code/unified-post-processor-upp). A list of CRA-40/Atmosphere products is given in Appendix A.

Conventional observations assimilated in CRA-40 are based on NCEP’s Prepbufr data used in CFSR (1979–2014, https://rda.ucar.edu/datasets/ds099.0) (Saha et al., 2010) and GDAS (2015–2018, https://rda.ucar.edu/datasets/ds735.0), but supplemented with additional data collected by NMIC.

For synoptic observations over land (SYNOP), CRA-40 combined data from CFSR/GDAS, Integrated Surface Database (ISD; Smith et al., 2011), and more than 2400 CMA’s surface synoptic stations. No METAR data are used in CFSR before June 1997. All the hourly METAR data used in CRA-40 come from ISD. METAR data from 1997 to 2009 in CFSR contain more observations than ISD, and therefore the counts of surface pressure (Ps) used in CRA-40 are slightly less than CFSR. Surface observations over the ocean (buoy and ships) combined data from CFSR/GDAS, ICOADS R3.0 (Freeman et al., 2017), NMIC archived global and regional datasets, and the North–East Asian ocean observations from Japan Oceanographic Data Center (JODC; https://near-goos1.jodc.go.jp). While multiple observed parameters are contained in these datasets, only surface pressure (P_s) observations are assimilated over land. Over the ocean, surface pressure, temperature, humidity, and wind observations are all assimilated in GSI.

Radiosonde observations combined six data sources: (1) NCEP’s CFSR/GDAS; (2) NCEI’s IGRA V2.0 dataset (Durre et al., 2018); (3) about 120 radiosonde stations operated by CMA; (4) about 80 pilot balloon stations operated by CMA from 1980 to 1991; (5) NMIC archived global radiosonde data from GTS; and (6) some field campaign datasets collected and archived by NMIC. Radiosonde temperature observations are bias corrected following RAOBCORE 1.4 (Haimberger, 2007) for non-Chinese stations, and Chen et al. (2021) for Chinese stations.

Aircraft data (AMDAR, ACARS, and AIREP) are based upon the integrated global aircraft dataset developed by NMIC (1973–present, Liao et al., 2021), which combined data from CFSR/GDAS, NCAR archived reports (from October 1999, https://rda.ucar.edu/datasets/ds351.0), NCEI’s worldwide aircraft reports (1973–1998, DSI-6380; NCDC, 2002), AMDAR reports from the Centre for Environmental Data Analysis (CEDA) archive of Met Office (https://catalogue.ceda.ac.uk/uuid/33f44351f9ceb09c495b8cef74860726), and NMIC archived Chinese AMDAR data (from May 2003). Variational bias correction is applied to aircraft temperature observations in GSI (Zhu et al., 2015).

Global wind profiler and pilot/pibal data are mainly from CFSR/GDAS and mostly available over US. Chinese wind profiler data are supplemented from 2007 with more than 100 stations in 2018.

Duplicated data from multiple data sources are removed according to the priority order predetermined for different data sources. Each type of conventional data is converted into the NCEP’s “BUFR DUMP” file format, which is then preprocessed into NCEP’s “Prepbufr” file format along with the GFS background file to perform in-fly quality control and is subsequently assimilated in GSI. The monthly blacklists of radiosonde, aircraft, and surface data are pre-generated by evaluating the data quality for each station, airplane, ship, and buoy at different times. The differences between observations and the ERA-Interim reanalysis are considered as “quality feedback information” to estimate the quality of conventional observations. The blacklists are determined depending on the quality feedback information and carefully manual diagnostic.

Availability of different types of conventional observations used in CRA-40 is given in Table 2. Figure 1 shows the monthly counts for all conventional observations (P_s, t, q, u) used in CRA-40 and CFSR. The two reanalyses differ mainly in the additional observations over China, the METAR data usage, and the assimilation of aircraft humidity.

Figure 1.  Monthly counts of conventional observations for (a) surface pressure, (b) temperature, (c) specific humidity, and (d) wind used in CRA-40 (red line) and CFSR (black line) from 1979 to 2014.

Satellite observations represent a majority of dataset assimilated in “satellite-era” reanalyses (typically referred to those reanalyses from 1979), including radiances from microwave and infrared sensors onboard polar-orbiting and geostationary platforms, GPS radio occultation bending angle data, retrieved AMVs, surface ocean winds, and O₃ concentration, as listed in Table 3 and Figs. 2 and 3. While satellite radiances and retrieved O₃ data are from CFSR/GDAS BUFR files, CRA-40 assimilated the latestly reprocessed GPSRO data by UCAR’s COSMIC, AMVs by CIMSS, EUMETSAT, JMA, CMA, and surface ocean winds by KNMI. Unlike conventional observations, quality control procedures of satellite data are mostly performed within GSI. For reprocessed retrieval products (AMVs, surface ocean winds), quality flags provided by the processing agencies are also kept when converting raw retrieval data into GSI’s BUFR format files and used in data assimilation.

Figure 2.  Timeline of conventional data, AMVs (i.e., SATWND), ocean surface winds, ozone, and GPSRO data used in CRA-40.

Figure 3.  Timeline of satellite radiance data using in CRA-40, including those from various microwave (MSU, AMSU-A, AMSU-B, MHS, MWHS-2, ATMS, and SSMIS) and infrared sensors (HIRS2–4, AVHRR3, AIRS, IASI, and CrIS) onboard various polar-orbiting satellites (e.g., n* for NOAA series, f17 for DMSP 17 satellite). Infrared radiances from geostationary GOES (g*) sounders are indicated by sndr*, while those from Meteosat-10 (m10) are indicated by SEVIRI.

The 40-yr evolution of the root-mean-square (RMS) of the first-guess departures (OMB) and analysis departures (OMA) from observations is shown in Fig. 4. The statistics are averaged over globe and over all vertical levels for upper air observations. Generally, the fitting to observations is improved over time, which is more pronounced from surface pressure and upper air temperature with the RMS of OMB (OMA) reduced from ~1.35 (1.05) hPa in 1979 to ~1.0 (0.8) hPa in 2018 for surface pressure and from ~2.1 (1.65) K in 1979 to ~1.6 (1.35) K in 2018 for upper air temperature. The improving fit over time is believed to be resulted from more and more satellite radiances assimilated. This can be partly verified with a larger improving fit to surface pressure over ocean (not shown), where satellite data are the dominant dataset.

Figure 4.  The 30-day mean (thin curves) and 360-day mean (thick curves) of the root-mean-square of first-guess (blue) (OMB) and analysis (red) (OMA) departures from observations in CRA-40 for (a) upper-air zonal wind from radiosondes, dropsondes, and PILOTs; (b) upper-air temperature from radiosondes and dropsondes; (c) upper-air specific humidity from radiosondes and dropsondes; and (d) surface pressure from SYNOP, buoys, ships, and METAR.

The change of fit to radiosonde specific humidity observations over time is relatively small, which might be an indication of relatively small impact of satellite radiances from humidity-sensitive channels. Geer et al. (2017) demonstrated the growing impact of microwave humidity sounders with all-sky assimilation approach. However, only clear-sky radiances were assimilated in CRA-40. For upper air zonal wind, there is an apparent trend of degraded analysis fit over time from ~1990 (red line in Fig. 4a) although the first-guess fit keeps improving or remains in a similar level. While this wind analysis fit degradation trend can be an aspect to be improved in the future version of reanalysis, it is also likely an indication of assimilation effect of more and more satellite radiances, which are directly sensitive to temperature and moisture and thus reduce the analysis fit to wind observations. Note that the first-guess fit to wind observations, which is usually a better indication of data assimilation performance than the analysis fit, does not exhibit an obvious trend of degradation.

Pressure–time diagrams for monthly and globally averaged mean analysis increments (i.e., analysis minus background) for specific humidity, zonal wind, and temperature are displayed in Fig. 5. With bias-corrected observations and unbiased model forecast, mean increments should be small. Systematic biases of increments typically indicate an average conflict between observations and the model background. Abrupt changes in mean analysis increments are usually related to changes in the observing system.

Figure 5.  Pressure–time diagrams of monthly and globally averaged analysis increments for (a) temperature (K), (b) zonal wind (m s⁻¹), and (c) specific humidity (mg kg⁻¹) on a logarithmic pressure scale.

The pattern of temperature increments (Fig. 5a) is very similar to that of ERA5 [see Fig. 16a of Hersbach et al. (2020)], e.g., opposite increments at the top of the stratosphere indicating a model bias with respect to anchoring satellite observations which peak at those heights, positive increment band around 200 hPa, and negative increment band below 200 hPa. Abrupt change of temperature increments at the upper stratosphere in late 1998 is most likely related to the introduction of NOAA-15 AMSU-A in September 1998 (Fig. 3). Mean zonal wind increments (Fig. 5b) are relatively small with a maximum of ~0.1 m s⁻¹ between 1 and 3 hPa. For specific humidity (Fig. 5c), sharper change from 2000 is related to the introduction of NOAA-15 AMSU-B (Fig. 3).

Pressure–time diagrams for monthly and globally averaged root-mean-square errors (RMSEs) of CRA-40, CFSR, and JRA-55 against ERA5 for geopotential height, zonal wind, temperature, and specific humidity are shown in Fig. 6. The RMSE of geopotential height (Fig. 6a) at high-levels (10–100 hPa) for CRA-40 is larger than that of CFSR, except for the period of 1998–2010. JRA-55 has a sudden increased RMSE for geopotential height at the layer of 10–100 hPa from 2000 to 2007. For zonal wind, all of three reanalyses have larger and periodically changing RMSEs above 100 hPa, reflecting the zonal wind high uncertainty at upper levels likely due to the lack of wind observations aloft. Similar to the time variation of analysis increment shown in Fig. 5a, there are also abrupt changes of temperature RMSE between 10 and 30 hPa in 1998, which may also be related to the evolution of observing systems. For specific humidity, CRA-40 has the smallest RMSE among three reanalyses when taking ERA5 as a reference (Fig. 6d).

Figure 6.  Pressure–time diagrams of three reanalyses’ monthly and globally averaged root-mean-square errors (RMSEs) for (a) geopotential height (gpm), (b) zonal wind (m s⁻¹), (c) temperature (K), and (d) specific humidity (g kg⁻¹) against ERA5. The RMSEs of CRA-40, CFSR, and JRA-55 are shown from top to bottom in each subplot.

The trend analysis of near surface air temperature at 2 m (t2m) is one important application of reanalysis data. Figure 7 shows 12-month running averages of global mean t2m anomalies (K) from 1979 to 2018 for CRA-40/Atmosphere. The global warming trend is well reflected in Fig. 7. While t2m is part of CRA-40/Atmosphere products, it is worth noting that the t2m is not an analy-sis variable of GSI and it is actually from the model’s 6-h forecast background. Therefore, we suggest that users employ the t2m (also for humidity at 2 m and wind at 10 m) product from CRA-40/Land, in which those near surface variables are produced by merging the corresponding model background from CRA-40/Atmosphere with global near surface observations using the EnOI method (Liang et al., 2020). Further details on CRA-40/Land will be presented in a separate article.

Figure 7.  Twelve-month running averages of global-mean near surface air temperature (T2m) anomalies (K) from 1979 to 2019 for CRA-40/Atmosphere. The T2m anomaly is computed relative to the mean from 1981 to 2010.

Figure 8 shows the climatology of global precipitation distributions in CRA-40, ERA5, JRA-55, and GPCP-v2.3 (2.5° × 2.5°) as an observational dataset. The global precipitation patterns of three reanalysis datasets match well with that of GPCP, while the precipitation magnitudes of three reanalyses are larger than that of GPCP. Over the ocean (e.g., ITCZ over Pacific), the magnitude of CRA-40 agrees more closely with that of ERA5, while JRA-55 exhibits stronger rainfall than both CRA-40 and ERA5. In some regions over land (e.g., central Africa, Southeast Asia, and north of South America), CRA-40 generally overestimates the precipitation compared to ERA5 and JRA-55, which is mainly caused by the sudden increase of precipitation over land since 1998 (not shown). Zhang et al. (2012) investigated the similar abrupt increase in precipitation from CFSR around 1998, and pointed out that it is related to the spin down of the increased initial moisture introduced by the assimilation of ATOVS data. It is also worth noting that these excessive precipitation regions in CRA-40 apparently correspond to regions with low correlation between ERA5/ERA-Interim and observed precipitation [see Fig. 24 of Hersbach et al. (2020)].

Figure 8.  Thirty-year (1981–2010) averaged precipitation rates (mm day⁻¹) in (a) CRA-40, (b) ERA5, (c) JRA-55, and (d) GPCP-v2.3.

Figure 9 displays the time series of 360-day running average of global mean precipitation (mm day⁻¹) from CRA-40, ERA5, JRA-55, and GPCP-v2.3 over 1979–2018. The magnitudes of mean daily precipitation are consistent with those shown in Fig. 8. JRA-55 has the strongest global mean daily precipitation with a mean value of 3.28 mm day⁻¹, while ERA5 with a mean value of 2.91 mm day⁻¹ agrees most closely with GPCP (2.69 mm day⁻¹). CRA-40 with a value of 3.12 mm day⁻¹ is roughly in the middle of JRA-55 and ERA5. Compared to ERA5 and JRA-55, one noticeable difference in CRA-40 is the increased precipitation amount from 2011, which is apparently associated with the wetter troposphere moisture increments in Fig. 5. The cause of excessive water vapor and precipitation from 2011 remains unclear, but we suspect that it might be related to the deficiency of satellite radiance DA for some sensors and/or channels. The mean evaporation of CRA-40 is lower than precipitation at most of years except for 2007–2009 and similar feature of evaporation vs. precipitation is also present in ERA5 (Hersbach et al., 2020). Unlike precipitation, the evaporation does not exhibit abnormal increase from 2011.

Figure 9.  Time series of 360-day running average of global mean precipitation rates (mm day⁻¹) from CRA-40 (red solid line), ERA5 (blue line), JRA-55 (green line), and GPCP-v2.3 (gray line). The red dash line is the corresponding evaporation from CRA-40.

The QBO of the zonal mean wind is a main feature of the tropical lower stratosphere, which is mainly manifested in the cyclical changes of east–west wind, with an average period of about two years (Wallace, 1973). Studies have shown that QBO is related to the Walker circulation, ENSO, tropospheric circulation, and climate change (Li and Long, 1992; Pawson and Fiorino, 1999). Figure 10 shows the zonal mean of monthly-mean zonal wind averaged between 5°S and 5°N as a function of pressure and time for ERA5, CRA-40, JRA-55, and CFSR. The results show that the QBO period and the corresponding vertical and time evolution characteristics are similar among four reanalyses. The main issue with CRA-40 is that the amplitudes of the westerly phase near 10 hPa and the easterly phase near 20 hPa are too strong when compared to other reanalyses. The issue is apparently reduced after 2005, when more AMSU-A radiance data available to constrain high-level state. We suspect that this deficiency is mainly caused by the GFS model (e.g., relatively low model top). Saha et al. (2010) pointed out that CFSR (based upon an older version of the GFS model) had a similar issue in an earlier version of the production, which was alleviated by assimilating ERA-40 stratospheric wind profiles above 20 hPa from 20°S and 20°N as bogus observations until December 1998.

Figure 10.  Time–pressure cross-section of zonal-mean of monthly-mean zonal wind (m s⁻¹) averaged between 5°S and 5°N for (a) ERA5, (b) CRA-40, (c) JRA-55, and (d) CFSR.

The hemispherical atmospheric circulation is highly related to the variability of the Tibetan Plateau heat source, and consequently influences greatly on the Asian monsoon, water and energy cycle, and precipitation variation (Zhao and Chen, 2001; Zhou et al., 2009). Based on the ECMWF reanalysis data from 1958 to 2001, Zhou et al. (2009) proposed a basic physical model to explain the influence of the Tibetan Plateau heat source on Northern Hemispherical atmospheric circulation. Following Zhou et al. (2009), we reproduced the summer zonal–vertical atmospheric circulation climatology (averaged from 1980 to 2010) across 15°–50°N, using the global reanalysis datasets from ECMWF (ERA5), NCEP (CFSR), JMA (JRA-55), and CMA (CRA-40). Figure 11a shows the longitude–pressure cross-section of zonal–vertical (15°–50°N) mean averaged summer (June, July, and August; JJA) wind of ERA5 covering 1980–2010. It can be seen that the updraft in the Tibetan Plateau separated into two parts. One part is uplifted to the lower stratosphere, moving westward and sinking over Europe, forming an anti-clock zonal–vertical circulation in the upper troposphere and the lower stratosphere. The other part flows eastward in the troposphere and sinks near the eastern Pacific Ocean. It is worth mentioning that, some of the eastward flow merges into the weak North American updraft, and continues to move eastward and then sink over the eastern Atlantic. Similar zonal–vertical atmospheric circulation characters are also found in CRA-40, CFSR, and JRA-55 (figure omitted). To see the details more clearly, Fig. 11b shows the differences between CRA-40 and ERA5, while Figs. 11c and 11d for JRA-55 and CFSR, respectively. It can be seen that, CRA-40 is closer to ERA5 than JRA-55 and CFSR over the Tibetan Plateau.

Figure 11.  Longitude–height cross-section of summer (June, July, and August) zonal–vertical atmospheric circulation climatology (1980–2010) between 15° and 50°N (a) for ERA5 and the difference (b) between CRA-40 and ERA5, (c) between JRA-55 and ERA5, and also (d) between CFSR and ERA5.

Atmospheric water vapor transport is one of the most important features in the East Asian summer monsoon (EASM) system, and it has significant influences on summer rainfall patterns in China (e.g., Park and Schubert, 1997; Zhou and Yu, 2005; Zhao et al., 2007). Figure 12a shows the 40-yr (1979–2018) mean of summer (JJA) water vapor transport (in kg m⁻² s⁻¹) over East Asia at 850 hPa from ERA5. Three major water vapor pathways can be seen clearly for normal climatological conditions in East Asia, as pointed out by Zhou and Yu (2005). The first one is the strong transport from the Arabian Sea and the Bay of Bengal to eastern China; the second comes from the tropical western Pacific Ocean; and the third is a cross-equator flow straddling 105°–150°E. Figures 12b–d are similar to Fig. 12a, but show differences of CRA-40, JRA-55, and CFSR relative to ERA5, respectively. CRA-40 has smaller differences from ERA5 than JRA-55 and CFSR for the first and second transport pathways. The spatial correlations of CRA-40, JRA-55, and CFSR with ERA5 are 0.98, 0.98, and 0.97, and the root-mean-square errors are 77.1, 84.7, and 84.3 kg m⁻² s⁻¹ respectively.

Figure 12.  Forty-year mean of JJA water vapor transport (kg m⁻² s⁻¹) at 850 hPa (a) from ERA5, and (b) the differences between CRA-40 and ERA5, (c) between JRA-55 and ERA5, and (d) between CFSR and ERA5. The colors indicate the magnitude of the water vapor flux vector.

Xue et al. (2020) evaluated the monthly variation of upper tropospheric water vapor (UTWV) in several reanalysis products using near-global homogenized water vapor channel (6.5 μm) radiances from geostationary satellites. They pointed out that the UTWV from ERA5, JRA-55, and CRA-40 agreed more closely to observations when compared to that from ERA-Interim, CFSR, and MERRA-2. Using MODIS data as the reference, Yao et al. (2020b) showed that CRA-40 slightly underestimated cloud cover, but captured well the influences of El Niño–Southern Oscillation (ENSO) on cloud spatiotemporal distributions like ERA5. With a newly reconstructed 35-yr daily solar radiation dataset for the Great Wall and Antarctica stations, Zeng et al. (2022) demonstrated that CRA-40 followed a relatively consistent trend with observations with a correlation coefficient of 0.977, but overestimated the solar radiation with a mean bias of 6.90 MJ m⁻² during the austral summer half-year. The assessment of surface air temperature over China against urbanization-bias-adjusted observations indicated that CRA-40/Land performed the best among eight global reanalysis products for the period 1979–2015 in reproducing various aspects of climatological features in China (Zhang et al., 2021). The evaluation of surface air temperature over the Tibetan Plateau by Yang et al. (2021) showed that CRA-40/Land is spatially consistent with 22 in situ observations, and has a comparable performance to ERA-Interim.

A new global atmospheric reanalysis CRA-40 produced by CMA for the first time is documented. The original 6-h model output covers a period from 1979 to 2018, with a horizontal resolution of T_L574 (~34 km) and 64 levels spanning from surface to 0.27 hPa. A large number of reprocessed satellite datasets and widely collected conventional observations were assimilated during the production. It is worth mentioning that, the 13-yr reprocessed AMV products from FY-2C/D/E/G are used in global reanalysis for the first time. Dense conventional observations over China (~120 radiosonde and ~2400 synoptic stations) are also used.

The original model level output is postprocessed into 47 pressure level products listing in Appendix A using UPP. All the variables are then interpolated to four horizontal resolutions in longitude–latitude projection, including 0.25° × 0.25°, 0.5° × 0.5°, 1.0° × 1.0°, and 2.5° × 2.5°. Daily and monthly products are also produced based on the 6-h products. All these products are publicly released in http://data.cma.cn/CRA. CRA-40/Atmosphere is now running operationally in near real time as a climate data assimilation system for the climate monitoring. In conjunction with CRA-40/Atmosphere, a land product is also produced through a single simulation using the Noah land surface model driven by observations adjusted near the surface. User is recommended to use t2m, q2m, u10, v10, and precipitation from the CRA-40/Land products, as they have merged SYNOP observations using EnOI method. Details on the production of CRA-40/Land will be presented in a separate article.

The next version of CMA’s global reanalysis will be based on CMA Global Forecast System (CMA-GFS) with the 4DVar DA technique. More satellite radiance data in early years (e.g., those from SSU) will be included to improve high-level temperature analysis. The reprocessed radiances from FY-3 series will also be used in the next generation reanalysis of CMA.

Acknowledgments. As with other reanalyses, CRA-40 is a collective work of many people from multiple organizations. We express our gratitude to all members involved in the funding projects from the National Meteorological Information Centre (NMIC), National Climate Centre (NCC), National Meteorological Centre (NMC), and National Satellite Meteorological Centre (NSMC) of CMA, and the Institute of Atmospheric Physics (IAP) of Chinese Academy of Sciences, as well as the Nanjing University of Information Science and Technology (NUIST).

We wish to thank the Central Operations (NCO) of NCEP for making the GSM model publicly available, Developmental Testbed Center (DTC) of NCAR for providing the GSI community software, and also the Research Data Archive (RDA) of NCAR for providing the CFSR input observations.