FENGYUN-4A Advanced Geosynchronous Radiation Imager Layered Precipitable Water Vapor Products’ Comprehensive Evaluation Based on Quality Control System

: A physical retrieval algorithm has been developed for deriving the layered precipitable water vapor (LPWs) product from infrared radiances of the Advanced Geosynchronous Radiation Imager (AGRI) onboard FengYun-4A (FY-4A), the ﬁrst of the new generation of Chinese geostationary weather satellites (FengYun-4, or FY-4 Series). The FY-4A AGRI LPWs are evaluated with different types of reference datasets based on Quality Control System (QCS), including those from Himawari-8 AHI (Advanced Himawari Imager), MODIS (Moderate Resolution Imaging Spectrora-diometer), Radiosonde, ERA5 (European Centre for Medium-Range Weather Forecasts Reanalysis v5), NCEP (National Centers for Environmental Prediction) reanalysis and CMA (China Meteorological Administration) forecast product from global medium range numerical weather prediction (NWP) system. QCS is one of the important components of FY-4A ground segment, which mainly focuses on the satellite products’ evaluation and validation. It is found that the AGRI LPW product has a good agreement with different evaluating sources and the quality is favorable and stable. With the capability of frequent (5-min interval) observations over the East Asia and Western Paciﬁc regions, the AGRI LPW products can be used to investigate the atmospheric temporal and spatial variations in the pre-landfall environment for typhoons. The QCS is a useful tool to monitor, evaluate, and validate the AGRI LPW products.


Introduction
Fengyun-4 (FY-4) is the latest generation of China's geostationary meteorological satellites for high-impact weather (HIW) event monitoring, warning, and forecasting, with the first FY-4A launched on 11 December 2016 [1]. Compared with the existing operational FY-2 series, the FY-4 series have a longer operating lifetime and improved capabilities for weather and environmental monitoring, including a new capability for vertical temperature and moisture sounding of the atmosphere with its high-spectral-resolution infrared (IR) sounder, the Geostationary Interferometric Infrared Sounder (GIIRS). For the coming 15 years, the three-axis stabilized FY-4 series will offer the imaging capabilities with fulldisc coverage every 15 min or better, and the option for more rapid regional and mesoscale observation modes. FY-4 represents an unprecedented expansion in China's geostationary remote sensing capabilities. FY-4A is experimental, and the subsequent satellites in the FY-4 series will be operational. The remaining satellites of this series are planned to be launched in the near future with FY-4B launched in 2021 [2]. For the FY-4 operational Clouds, air temperature * S/N is signal to noise ratio. ** ρ is typical reflectance. *** NE∆T is noise equivalent temperature difference.
The products from AGRI of FY-4C and after will have an IR spatial resolution of 2 km, together with those from other international geostationary satellites, will provide near global quantitative datasets with high spatial and temporal resolutions for weather and climate applications. Atmospheric precipitable water vapor plays an important role on the Earth's weather and climate, especially under the global warming condition. It not only affects the radiation balance and atmospheric circulation, but also impacts the intensity of precipitation by changing the moisture transport. The atmospheric water vapor content is an important parameter for weather forecasting, a key factor for studying the global water cycle, climate change, the energy circulation of the earth-atmosphere system, and the input parameters for aerosol and surface temperature retrieval [7][8][9]. Water vapor is an important factor in climate as the main feedback factor associated with the radiation effect and moisture dynamics [10]. It is the most abundant greenhouse gas, and the increase of tropospheric water vapor is closely related to climate warming [7]. Therefore, accurate estimate of atmospheric water vapor content is important towards the predictability of rainfall and our understanding of many feedback processes.
Over the past few decades, many techniques have been developed to obtain atmospheric precipitable water vapor (PW), or the vertically integrated water vapor content, including in situ measurements from radiosonde observation (RAOBs) [11], retrievals from ground-based sun photometry [12], microwave radiometers [13][14][15], Raman radar [16,17], Global Positioning System (GPS) receivers [18][19][20], and satellite remote sensing [21,22]. Among them, in situ radiosonde represents the most traditional PW measurement technology. Because the radiosonde can obtain PW with controlled accuracy, it is widely used as the reference database to evaluate other PW retrieval techniques [23,24]. But limitations such as high cost, sparse spatial distribution, large bias in the upper troposphere, and low temporal and spatial resolution (launched only twice a day) prevent the usability of radiosonde for monitoring the wide range of weather process and short-term weather forecasting. Ground-based GPS, ground-based sun photometer and microwave radiometers measurements are three kinds of reliable and common techniques to retrieve PW with high precision and accuracy as well as high temporal resolution. However, these ground-based observations are still fairly sparse for most areas.
Satellite observations can be used as a more effective technology to monitor the spatial distribution and temporal changes of PW on a large scale. For example, the MODerate resolution Imaging Spectroradiometer (MODIS) onboard TERRA and AQUA provides PW observations with high spatial resolution [25][26][27][28]. Over the past 20 years, the MODIS PW products have been extensively evaluated using ground-based PW observations over most areas of the world [29][30][31][32][33][34], and used in weather, climate, and hydrological cycle studies [26,28,31]. Other satellite-based PW products include those from Atmospheric InfraRed Sounder (AIRS), Cross-track Infrared Sounder (CrIS), and Infrared Atmospheric Sounding Interferometer (IASI) [35]. Due to low spatial and temporal resolutions of low earth orbit (LEO) satellites, there are about 50% data missing over high-altitude area (i.e., Tibetan Plateau) from 0.05 • × 0.05 • spatial resolution of MODIS data [36]. In addition, measurements from sounders with coarser spatial resolution have limited capability on detecting small scale atmospheric moisture spatial variations [37].
Compared with LEO based PW products, geostationary orbit (GEO) offers the unique value with much higher temporal resolution. For example, the ABI onboard the new generation GOES-16/-17 satellite has fine temporal (30 s to 10 min) and spatial resolutions (0.5-2 km), and 16 spectral bands. GOES-16 started a new era for the U.S. geostationary satellite observing system. However, due to the lack of an infrared sounder on the spacecraft, the ABI is used to continue the legacy atmospheric profile (LAP) products [4] that the previous GOES Sounder [38] provided for nowcasting and weather forecasting [39]. Those LAP products from ABI include the legacy atmospheric vertical moisture profile (LVM), legacy atmospheric vertical temperature profile (LVT), total precipitable water (TPW), layered precipitable water (LPW), and derived atmospheric stability indices (DSI) over each 5 × 5 ABI pixels box area with clear-sky IR band radiances [40]. The GOES-R LAP products have been evaluated by National Weather Service (NWS) forecasters at the NOAA Hazardous Weather Testbed (HWT) and received positive feedbacks for severe storm nowcasting and forecasting.
The LPW product retrieval algorithm for FY-4 AGRI was introduced in Zhang et al. (2019) [41]. The algorithm is similar to that used for the operational GOES Sounder profile Atmosphere 2022, 13, 290 4 of 22 retrieval [42] and is responsible for the retrieval of atmospheric layer integrated moisture for AGRI field of regard (FOR) (an M × M pixel box). The product generation takes IR brightness temperatures (BTs) from all AGRI IR channels as inputs along with numerical weather prediction (NWP) forecast output as background. The one-dimensional variational (1Dvar) based physical retrieval algorithm modifies the temperature and moisture profiles based on the differences between the observed and the calculated BTs using Newton's nonlinear iterative method [43]. The LPW products are derived from the moisture profile retrievals, including TPW and PW at three atmospheric layers. TPW is the amount of liquid water (in the unit of cm) if all the atmospheric water vapors in the column was condensed. Layer perceptible water (LPW) provides information on the water vapor contained in a vertical column of unit cross-section area in three layers in the troposphere in sigma coordinate: LPW low (LPWL): 0.9-surface (SFC), LPW middle (LPWM): 0.7-0.9, LPW high (LPWH): 0.3-0.7. The FY-4A AGRI clear sky LPW products are available in 12 km resolution (an FOR of a 3 × 3 box area) for full-disc or China region.
It is important to point out that GIIRS is the primary instrument to provide water vapor (and temperature) profile information with its thousands of sounding channels. However, comparing to AGRI, there are several limitations in the FY-4A GIIRS water vapor products for nowcasting and forecasting, including spatial coverage, temporal and spatial resolutions. AGRI covers different sectors including full disc, north hemisphere, China area and focus domains, while GIIRS only covers the rectangle area of 3 • N-55 • N and 60 • E-137 • E. AGRI has a temporal resolution better than 15 min, much better than GIIRS' 2 h. While the spatial resolution of AGRI LPW products is 12 km, comparable to GIIRS' 16km, the AGRI LPW products have better retrieval yields than what would have been achieved from GIIRS. This is because AGRI retrievals are performed if there are less than 9 cloudy IR pixels within an FOR. Such higher temporal and spatial resolution data are important in the climate sensitive regions such as the high-altitude mountain terrain region of Tibetan Plateau in particular where there is limited number of ground-based observations [44]. It should be noted that both spatial and temporal resolutions will be improved for moisture soundings from the future Fengyun-4 GIIRS measurements.
Comprehensive evaluation of FY-4A AGRI LPW products has been carried out and the results are presented in this paper. Section 2 describes the data used for comparisons in this study. Section 3 introduces the Quality Control System of FY-4A ground segment. Section 4 explains the evaluating method. And Section 5 presents results and discussions, followed by conclusions in Section 6.

Data
To evaluate FY-4A AGRI LPW products, different types of water vapor data were collected, compared and analyzed, including Himawari-8 AHI Layered Perceptible Water data, RAOB data, ECMWF Reanalysis v5 (ERA5) data, National Centers for Environmental Prediction (NCEP) reanalysis, and China Meteorological Administration (CMA) forecast product from NWP system. Brief introductions of these data are shown below.

FY-4A AGRI LPW Product
The AGRI can complete a full disk scan every 13 min, a regional scan covering China every 5 min, and a meso scan every 2 min. To reduce temporal redundancy and balance between the full disks, regional, and meso scans, the AGRI is configured to have 40 full disks and 165 China regional LPWs files for each day. The full disk observations are made every nominal hour plus 15 min before and after the nominal 3-h (00, 03, 06, etc.), which is specially designed for atmospheric motion vector (AMV) applications. Three consecutive regional observations (also for AMV) are made at 30 min after every nominal hour and 15 min before and after nominal hour when there are no full disc observations [1]. Since 18 January 2019, LPW products of FY-4A are publicly available. The data used in this study are from the time period from 23 October 2019 to 30 May 2020 from the domains of full disk and China region.

Himawari-8 AHI LPW Data
The AHI LPW retrievals are also based on the operational GOES Sounder profile retrieval algorithm [42], which is a 1Dvar retrieval method based on optimal estimation theory [40,45]. The Quasi-Newton iteration technique is employed in the 1Dvar solution, and the University of Wisconsin-Madison's (USA) SeeBor emissivity database is applied as a prescribed input [46]. Since the IR bands of AHI are similar to those of ABI [4], the employed bands are also the same as those of the ABI. After the atmospheric temperature and moisture profiles have been retrieved, the derived products are calculated from the profiles, and the methods for obtaining these products are described in the ABI/GOES-R legacy atmospheric profile (LAP) retrieval Algorithm Theoretic Basis Document (ATBD) [47]. The LAP retrieval algorithm [43,45] has been tested and validated using the spinning enhanced Visible and Infrared Imager (SEVIRI) and the GOES Sounder as proxy [48], as well as the AHI [49] and ABI [40] observations. While the LPWs are not operational LAP products, they can be calculated from the retrieved moisture profiles using the same definition of the three layers in sigma coordinates. The AGRI LPWs will be inter-compared with those from AHI to ensure comparable quality from the two satellites.

MODIS PW Product
The MODIS Precipitable Water product consists of column water-vapor amounts. During the daytime, a near-infrared algorithm is applied over clear land areas of the globe and above clouds over both land and ocean. Over clear ocean areas, water-vapor estimates are provided over the extended glint area. An infrared algorithm for deriving atmospheric profiles is also applied both day and night for Level 2. There are two MODIS Water Vapor data product files: MOD05, containing data collected from the Terra platform; and MYD05, containing data collected from the Aqua platform. The Level 2 data are generated at the 1-km spatial resolution of the MODIS instrument using the near-infrared algorithm during the day, and at 5-km resolution (a 5 × 5 box) both day and night using the infrared algorithm when at least nine FOVs are cloud free. The infrared-derived precipitable water vapor is generated as one component of product MOD07/MYD07, and simply added to product MOD05/MYD05 for convenience. The solar retrieval algorithm relies on observations of water-vapor attenuation of reflected solar radiation in the near-infrared MODIS channels so that the product is produced only over areas where there is a reflective surface in the near IR [25][26][27][28].

RAOB Data
There are two RAOB datasets used in this study. For FY-4A AGRI China region, the RAOB data was obtained by the L-band sounding system (1675 MHz) of the CMA, named as "soundings", and those collected from the University of Wyoming website [50] were named as TK. It contains the observation data of the height, temperature, dew point temperature, wind direction, wind speed of the 89 Chinese radiosonde observation stations obtained by the National Meteorological Information Center (NMIC) through the domestic communication system of China. All these stations only cover Chinese territory. The CMA sounding system is composed of Global Telecommunications System-1 (GTS1) digital electronic radiosonde, secondary wind-finding radar and a ground-check set. It is widely used to measure the air pressure, temperature, relative humidity, and wind from the ground to about 30 km at radiosonde sites across China. The accuracy of the measured pressure, temperature, and relative humidity is 1-2 hPa, 0.2-0.3 • C and 4-5%, respectively [51]. The radiosonde LPW was derived by integrating the specific humidity using the same algorithm for AGRI with the RAOB observed surface pressure. The RAOB LPW will be used as reference for the evaluation purpose at 00 and 12 UTCs.

ERA5 Reanalysis Data
ERA5 is the fifth generation ECMWF atmospheric reanalysis [52]. The ERA5 was produced using the 4D-Var data assimilation in CY41R2 of ECMWF's Integrated Fore- cast System (IFS), with 137 hybrid sigma/pressure (model) levels in the vertical, with the top level at 0.01 hPa. The ERA5 is preferred over the previously widely used ECMWF Re-analysis Interim (ERA-Interim), which is the third-generation global atmospheric reanalysis from ECMWF [53], because of higher temporal resolution of 1 h and higher spatial resolution of 31 km. The ERA5 complements the RAOB measurements in both time and space, making it possible to evaluate satellite sounding products on extended spatial and temporal scales.

NCEP Data
The NCEP/NCAR Reanalysis project uses a state-of-art analysis/forecast system to perform data assimilation using past data from 1948 to the present [54]. A large subset of this data is available from Physical Sciences Laboratory (PSL) in its original 4 times daily format and as daily averages. The local ingestion process took only the 0 Z, 6 Z, 12 Z, and 18 Z reanalyzed values. Those four times were used to make the daily time series and monthly means in this study.

CMA T639 Products
T639L60 global medium range NWP model products have high resolutions [51]. Its global horizontal resolution reaches 30 km with 60 vertical layers, and its model top reaches 0.1 hPa. The analysis data of T639 were used in this paper. The data files adopt grib2 format, which are disseminated daily at 00:00 and 12:00 UTC.

Quality Control System of FY-4A Ground Segment
Quantitative remote sensing products' quality evaluation and validation is an important part of the FY-4A ground segment and the basic technical goal of the product quality control system. Satellite remote sensing products are derived by processing the data obtained by satellite remote sensors through scientific methods. The quality of the products directly affects the quantitative applications of remote sensing data. In high-altitude and mountain terrain areas, especially when the estimated parameters are of the same order with the measurement uncertainties, maintaining quality control of satellite retrieval products are important [44]. It is very necessary to carry out the evaluation and validation of the accuracy of satellite remote sensing products. An operational processing system for products quality control, named QCS of FY-4A, has been developed.

Method
In this study, the temperature and humidity profiles of RAOB data, the layer water vapor products of Himawari-8 AHI, NCEP, T639, ERA5, and TPW of Aqua/Terra MODIS are used to calculate bias, standard deviation (STD), root mean square error (RMSE) and correlation coefficient (CC). These indicators will be used to characterize the quality of the FY-4A AGRI LWP products.
Because the spatial resolution and observing time of the AGRI LPW products and the reference evaluation data are different, according to their own characteristics, appropriate methods are needed to do the spatial and time matching, which will be introduced later. After data preprocessing and matching, the FY-4A AGRI LPW products are compared with the reference datasets for evaluation. Some statistical indicators will be calculated. This will be done for different latitudes, different time periods, and user-defined regions. Finally, the visualization of the evaluation results of various statistical indicators are shown in the form of geographical distribution maps, histograms, and statistical tables.

Temporal Matching
The radiosonde data is obtained twice a day (0:00 and 12:00 UTC). Based on the observation time of the RAOB data, the time matching threshold is set to ±60 min, and the nearest FY-4A AGRI LPW products within the time interval is searched for matching the RAOB data. The time resolution of Himawari-8 AHI LPW data is 10 min. Based on the observation time of Himawari-8 AHI LPW data, the time matching threshold is set to ±5 min, and the nearest FY-4A AGRI LPW product within the time interval is selected for matching. For TERRA/AQUA MODIS PW data, also based on the observation time of MOIDS, the time interval is set to ±5 min, and the nearest FY-4A AGRI LPW product within the time interval is selected. The global model water vapor profile data (include ERA-5, NCEP and T639 data) is generally generated 4 times a day (0 Z, 6 Z, 12 Z, and 18 Z) except ERA-5. Based on the generation time of the model data, the time interval threshold is set to ±60 min, and the nearest FY-4A AGRI LPW product within the time interval is searched for matching data. For ERA-5, the time threshold is set to ±30 min.

Spatial Collocation
For radiosonde data, based on the FY-4A pixels as a center, the sounding stations were searched in a 20 km radius area. The sounding data of the matched station were compared with AGRI LPW products. Same as radiosonde data, the FY-4A pixels were also used as the center to search the Himawari-8 AHI and MODIS pixels in a 5 km radius area. All matched Himawari-8 AHI and MODIS pixels were compared with AGRI LPW products. The global model water vapor profile data (include ERA-5, NCEP and T639 data) used in this study were with different spatial resolution, 2.5 • grid for NCEP, 0.25 • grid for ERA-5 and 0.28125 • grid for T639 data. FY-4A products were resampled to the same spatial grid as global model water vapor profile data separately to complete the spatial collocation.

Results
The QCS automatically collocates the AGRI LPW product with three types of evaluating data, including radiosonde data, model data and satellite data, and generates various evaluation results in near real time. Figures with statistical results are automatically generated from those evaluations and are continuously monitored to ensure the quality of the AGRI LPW products. This study shows the monthly evaluation results for the month of November 2019, and more comprehensive statistics for longer period from 23rd October 2019 to 15th November 2020. These are done for both full disc and the China region. All figures and statistic results in this paper were generated by QCS.

Evaluation with Radiosonde Data
Monthly mean bias distribution maps of TPWs and LPWL by radiosonde data for November 2019 were shown in Figure 1. From Figure 1a,c, positive bias was mainly located in low latitude area. From Figure 1b,d, the large bias stations were mainly located in southeast China area. That is reasonable because that is the region with more moisture than the rest part of China. Figures 2 and 3 show scatter plots and histograms of TPW and LPWL between AGRI and radiosonde data. From these Figures, it can be seen that FY-4A TPW and LPWL products were with good consistency to radiosonde data, especially for TPW.  Monthly statistical analysis of differences between AGRI TPW/LPWL and radiosonde data were listed in Table 2. The time coverage of the data pairs was November 2019.   Monthly statistical analysis of differences between AGRI TPW/LPWL and radio-339 sonde data were listed in Table 2. The time coverage of the data pairs was November 2019. 340  Figure 4 shows the 342 12-hourly time series of the TPW/LPWL bias and STD compared with different radio-343 sonde dataset from 23rd October 2019 to 20th February 2020. The black lines in each sub-344 plot showed the mean bias and the grey area showed the STD of each data pair. From 345 Figure 4 we can see that the bias was around 0 and STD was within ±0.3 cm for whole 346 comparing period except some outliers. A good sounding product should have temporally stable biases. Figure 4 shows the 12-hourly time series of the TPW/LPWL bias and STD compared with different radiosonde dataset from 23rd October 2019 to 20th February 2020. The black lines in each subplot showed the mean bias and the grey area showed the STD of each data pair. From Figure 4 we can see that the bias was around 0 and STD was within ±0.3 cm for whole comparing period except some outliers.  34 with RAOB. 35 35 Same as comparisons with radiosonde data, the evaluation results with model data 35 for November 2019 were also shown in geographic map, scatter plots, histograms and 35 time series for 3 LPWs. From Figure 5a, 5c and 5e, the consistent significant negative bias 35 can be seen in Tibet, India, Arabian, Africa when compared with T639. In Figure 5b

Evaluation with Model Data
Same as comparisons with radiosonde data, the evaluation results with model data for November 2019 were also shown in geographic map, scatter plots, histograms and time series for 3 LPWs. From Figure 5a,c,e, the consistent significant negative bias can be seen in Tibet, India, Arabian, Africa when compared with T639. In Figure 5b two big bias areas with opposite values in symmetrical areas across equator around Southeast Asia, are identified. This phenomenon didn't appear in the bias distribution map with T639 products (Figure 5a). From Figure 5b,d,f, the consistent significant negative bias can be seen over Pacific Ocean off southeast Asia and the south Indian Ocean, and positive bias over Australia and nearby oceans when compared with NCEP. Figures 6 and 7 show scatter plots and histograms of differences between AGRI and model data for LPWL/LPWM/LPWH. From these figures, FY-4A LPW products were with good consistency to model data. From Figure 7, for NCEP data low layer, there were bi-peak shape in histogram. That is because two big bias area with opposite values in symmetrical area across equator showed in Figure 5b. The monthly detailed evaluation statistics were listed in Table 3.  Figures 6 and 7 show scatter plots and histograms of differences between AGRI and 364 model data for LPWL/LPWM/LPWH. From these figures, FY-4A LPW products were with 365 good consistency to model data. From Figure 7, for NCEP data low layer, there were bi-366 peak shape in histogram. That is because two big bias area with opposite values in sym-367 metrical area across equator showed in Figure 5b. The monthly detailed evaluation statis-368 tics were listed in Table 3.    Figure 8, we can see that AGRI LPW products were with very good consistency to T639, NCEP and ERA5, not only in bias but also in STD. For T639 data, the biases are around 0 cm and the STD were less than 0.2 cm, except for LPWH around 0.4 cm; for NCEP data there were also with a bias around 0 cm, but the STD were less than 0.3 cm in all three layers. For ERA5 data, only MID layer was compared and also with good consistency. Figures 6 and 7 show scatter plots and histograms of differences between AGRI and 364 model data for LPWL/LPWM/LPWH. From these figures, FY-4A LPW products were with 365 good consistency to model data. From Figure 7, for NCEP data low layer, there were bi-366 peak shape in histogram. That is because two big bias area with opposite values in sym-367 metrical area across equator showed in Figure 5b. The monthly detailed evaluation statis-368 tics were listed in Table 3.

387
The AGRI TPW and LPW products were also evaluated by inter-comparing with 388 other satellite retrieval products from AHI and MOD05. The results for November 2019 389 were shown in geographic maps, scatter plots, histograms, and time series. Because 390 MOD05 only has TPW product. The comparison between AGRI and MOD05 was only 391 carried out with TPW. The comparison with AHI is also focused on TPW. The monthly 392 accumulated bias distribution map of TPW was shown in Figure 9. From Figure 9a, we 393 can see that there is an area with large positive bias around equator and ocean area. This 394 is likely caused by MOD05 because AGRI TPW shows no such pattern when compared 395 with different model data. And the positive bias appears to be related to cloud contami-396 nation. To verify that, Figure 10 shows the mean brightness temperature of FY-4A AGRI 397 10.8 um every 6 h from 1st to 30th November 2020, totally 121 time observing data. The 398 region with positive bias in Figure 9a is clearly an indication of more likelihood of being 399 affected by clouds with colder BTs than surrounding area. The MOD05 algorithm, as a 400 result, underestimates TPW of the region, leading to positive bias in Figure 9a.

Inter-Comparison with Other Satellite Retrieval Products
The AGRI TPW and LPW products were also evaluated by inter-comparing with other satellite retrieval products from AHI and MOD05. The results for November 2019 were shown in geographic maps, scatter plots, histograms, and time series. Because MOD05 only has TPW product. The comparison between AGRI and MOD05 was only carried out with TPW. The comparison with AHI is also focused on TPW. The monthly accumulated bias distribution map of TPW was shown in Figure 9. From Figure 9a, we can see that there is an area with large positive bias around equator and ocean area. This is likely caused by MOD05 because AGRI TPW shows no such pattern when compared with different model data. And the positive bias appears to be related to cloud contamination. To verify that, Figure 10 shows the mean brightness temperature of FY-4A AGRI 10.8 um every 6 h from 1 to 30 November 2020, totally 121 time observing data. The region with positive bias in Figure 9a is clearly an indication of more likelihood of being affected by clouds with colder BTs than surrounding area. The MOD05 algorithm, as a result, underestimates TPW of the region, leading to positive bias in Figure 9a.  Table 4. From these Figures and Tables, it can be seen that FY-4A TPW 407 product was with good consistency to different evaluation sources. For AHI and MOD05, 408 they are two types of satellite-based data, the monthly accumulated samples were very 409 large numbers and with smaller bias.   Table 4. From these Figures and Tables, it can be seen that FY-4A TPW 407 product was with good consistency to different evaluation sources. For AHI and MOD05, 408 they are two types of satellite-based data, the monthly accumulated samples were very 409 large numbers and with smaller bias.    December 2019 for AHI. The black lines in each subplot showed the mean bias and the 417 grey area showed the STD of each data pair. From Figure 13 we can see that the bias and 418 STD between AGRI and AHI were almost 0 for all data pairs. The AGRI and AHI agree 419 with each other so well mainly because they use the same physical retrieval algorithm 420 and the same NWP background. So, the differences between them should be mainly 421 caused by the differences in the two sensors, which appears to be small. For, AGRI and 422 MOD05, the bias and STD change rapidly in different dates, the mean bias was around 423 0.2 cm and the STD was within 0.3 cm, respectively.   Figure 13 shows the hourly time series of the TPW bias from comparisons with the reference dataset from 27 October to 3 December 2019 for MOD05 and 22 October to 7 December 2019 for AHI. The black lines in each subplot showed the mean bias and the grey area showed the STD of each data pair. From Figure 13 we can see that the bias and STD between AGRI and AHI were almost 0 for all data pairs. The AGRI and AHI agree with each other so well mainly because they use the same physical retrieval algorithm and the same NWP background. So, the differences between them should be mainly caused by the differences in the two sensors, which appears to be small. For, AGRI and MOD05, the bias and STD change rapidly in different dates, the mean bias was around 0.2 cm and the STD was within 0.3 cm, respectively.

430
The FY-4A AGRI retrieved sounding products have been comprehensively evaluated 431 using various reference sources automatically based on QCS of FY-4A, including in-situ, 432 model data, and other satellite measurements. Evaluation report can be generated auto-433 matically by each month. 434 The FY-4A AGRI LPWs product is integrated from the moisture sounding profile for 435 three layers in sigma coordinate: surface-0.9, 0.9-0.7, 0.7-0.3 and total precipitable water 436 vapor. With the higher temporal resolution from AGRI, the LPWs products are now avail-437 able for 40 full disk observations and 165 China area observations daily. It provides fore-438 casters moisture information in TPW, lower, middle, and upper troposphere. From this 439 study, the quality of AGRI LPWs products is favorable and stable. These evaluation re-440 sults show that the AGRI LPWs products have reasonable quality, and are ready for op-441 erational use. And the QCS is a useful tool to monitor, evaluate, and validate the AGRI 442 LPW products. The AGRI LPWs products are available in NetCDF via National Satellite 443 Meteorological Center's (NSMC) FENGYUN Satellite Data Center [55].  Data Availability Statement: The T639 data were obtained via http://data.cma.cn/data/de-454 tail/dataCode/F.0003.0001.html, (accessed on 29 January 2022). ERA-5 products used are available 455 from ECMWF (https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5, accessed on 29 456 January 2022). NCEP Reanalysis data provided by the NOAA/OAR/ESRL PSL, Boulder, Colorado, 457 USA, from their Web site at https://psl.noaa.gov/data/gridded/reanalysis/, accessed on 29 January 458 2022. Himawari-8 L1 Gridded data is provided by the Japan Meteorological Agency 459 (https://www.data.jma.go.jp/mscweb/en/himawari89/cloud_service/cloud_service.html, accessed 460 on 29 January 2022). Radiosonde data are available from the University of Wyoming 461 (http://weather.uwyo.edu/upperair/sounding.html, accessed on 29 January 2022) and the China Me-462 teorological Administration (http://data.cma.cn/data/cdcindex/cid/565b6087618affa1.html, accessed 463 on 29 January 2022).

Conclusions and Discussion
The FY-4A AGRI retrieved sounding products have been comprehensively evaluated using various reference sources automatically based on QCS of FY-4A, including in-situ, model data, and other satellite measurements. Evaluation report can be generated automatically by each month.
The FY-4A AGRI LPWs product is integrated from the moisture sounding profile for three layers in sigma coordinate: surface-0.9, 0.9-0.7, 0.7-0.3 and total precipitable water vapor. With the higher temporal resolution from AGRI, the LPWs products are now available for 40 full disk observations and 165 China area observations daily. It provides forecasters moisture information in TPW, lower, middle, and upper troposphere. From this study, the quality of AGRI LPWs products is favorable and stable. These evaluation results show that the AGRI LPWs products have reasonable quality, and are ready for operational use. And the QCS is a useful tool to monitor, evaluate, and validate the AGRI LPW products. The AGRI LPWs products are available in NetCDF via National Satellite Meteorological Center's (NSMC) FENGYUN Satellite Data Center [55].