Cloud fraction (CFR) and cloud top pressure (CTP) are significant cloud macro-physical properties and fundamental factors for the earth’s energy budget and hydrological cycle (Stephens, 2005; Butt et al., 2010; Dessler, 2011; Webb and Lock, 2013; Wang et al., 2015). They are also crucial for climate change studies and distinctly affect the accuracy of climate prediction due to their uncertainties in various climate modeling systems (Colman, 2003; Bony and Dufresne, 2005; Medeiros et al., 2008; Bony et al., 2015; Engström et al., 2015; Brient et al., 2016). Therefore, it is essential to obtain reliable and accurate information for these two cloud parameters. Cloud parameters are mainly derived from observations of ground-based radars and lidars (Hogan et al., 2001; Brooks et al., 2005; Hawkinson et al., 2005; Dong et al., 2006; Oue et al., 2016), airborne radars and lidars (Li et al., 2005; Lindstrot et al., 2006; Bedka et al., 2007; Settle and van de Poll, 2007), aircrafts (Wood and Field, 2000), satellites (Schiffer and Rossow, 1983; Rossow and Schiffer, 1991, 1999; Koelemeijer and Stammes, 1998; Koelemeijer et al., 2002; Zwally et al., 2002; Winker et al., 2010; Escrig et al., 2013; Xu et al., 2014; Håkansson et al., 2018), etc. Ground-based and aircraft observations can yield high temporal and spatial resolution cloud characteristics, but their spatial coverage is limited (Zhao et al., 2014). Satellites can provide data with a large regional coverage and even global coverage, and multiple observations as well as more cloud parameters that can be obtained per day over the same region. Polar orbiting satellites, such as NASA Terra, NASA Aqua [both of which have the Moderate Resolution Imaging Spectroradiometer (MODIS)], CloudSat, and Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO), provide high spectral and spatial resolution cloud products, but they can hardly capture diurnal variations of cloud parameters because the products are derived only a few times per day over the same region (Weisz et al., 2007; Wang and Zhao, 2017). Geostationary satellites such as the Himawari series, Geostationary Operational Environmental Satellite (GOES) series, Fengyun-2 (FY-2) series, and Fengyun-4A (FY-4A) can offer real-time monitoring of cloud characteristics over the fixed areas with the large regional coverage and high spatiotemporal resolution (Lu et al., 2008; Painemal et al., 2013; Yang et al., 2017; Chen et al., 2018; Wang et al., 2018).

Two generations of geostationary satellites have been developed in China. The FY-2 series of geostationary satellites is the first generation, enabling real-time monitoring with a focus on Asia. CFR derived from Visible Infrared Spin Scan Radiometer aboard FY-2C (VISSR/FY-2C) observations shows consistencies with those derived from other datasets over Southeast Asia (Jin et al., 2009). FY-4 series of geostationary satellites is the second generation, and the first satellite of this series (FY-4A) was launched on 11 December 2016. FY-4A has enhanced the monitoring capability compared to the FY-2 series. For instance, the Advanced Geosynchronous Radiation Imager (AGRI) aboard FY-4A (AGRI/FY-4A) takes full-disk images at a 15-min interval in 14 spectral bands with the 0.5–4-km resolution, while images available from VISSR/FY-2C are at a 30-min interval in 5 spectral bands with the 1.25–5-km resolution. In addition, more cloud products can be derived from AGRI/FY-4A, including the cloud mask, CFR, cloud top properties, cloud optical and microphysical properties, etc. Therefore, FY-4A can provide a greater variety of cloud products at the higher spatiotemporal and spectral resolution compared to the FY-2 series. Furthermore, FY-4A can obtain more accurate diurnal variations and spatial distributions of cloud characteristics over Asia than polar orbit satellites. Effective use of cloud products derived from AGRI/FY-4A observations can further improve our understanding of clouds. However, the quality and reliability of AGRI/FY-4A cloud products have not been evaluated yet.

Cloud products from MODIS/Terra and MODIS/Aqua, which are polar orbiting satellites, have been widely used and improved for more than 10 yr. They are considered as the relatively mature and reliable datasets (Zhao and Di Girolamo, 2006; Grandey et al., 2013; Gryspeerdt et al., 2016). Given the quality and reliability of MODIS in characterizing clouds, MODIS cloud products are often used to evaluate cloud products measured by other instruments. Previous studies (Weisz et al., 2007; Marchand et al., 2010) revealed that MODIS cloud top products are generally consistent with other cloud top products such as those from the Atmospheric Infrared Sounder (AIRS), CALIPSO, and International Satellite Cloud Climatology Project (ISCCP). Inconsistencies between them are mainly attributed to differences in the retrieval algorithm and instrument. Stammes et al. (2008) pointed out that the correlation coefficient between CFRs from the Ozone Monitoring Instrument (OMI) O₂–O₂ and MODIS is 0.92. Shang et al. (2018) indicated that the cloud detection and CFR derived from MODIS and Advanced Himawari Imager (AHI) are reliable over the Tibetan Plateau (TP). All these studies mentioned above suggest that the MODIS cloud products are reliable and of good quality, and thus can be used to evaluate the FY-4A cloud products.

This paper aims to evaluate the reliability of AGRI/FY-4A cloud products over East Asia (10°S–45°N, 60°–140°E) based on the comparison with MODIS/Terra and MODIS/Aqua datasets. We select CFR and CTP as representative variables of cloud products for comparison, since CFR is a vital component of cloud macro-physical properties and CTP is an important variable describing the cloud top height. The remainder of the paper is organized as follows. In Section 2, we describe the AGRI/FY-4A and MODIS cloud products as well as methods used in this study. Section 3 compares the CFR and CTP results of AGRI/FY-4A with those of MODIS/Terra. Sections 4 and 5 present a discussion of the results and summary of major conclusions.

The FY-4A geostationary satellite, a three-axis stabilized satellite, is located at 104.7°E above the equator with three advanced optical instruments aboard, including an AGRI, a Geostationary Interferometric Infrared Sounder (GIIRS), and a Lightning Mapping Imager (LMI). AGRI has 14 spectral bands from visible to infrared (0.45–13.8 μm) with high spatial (1 km for visible at nadir, 2 km for near-infrared, and 4 km for remaining infrared) and temporal (full-disk images at the 15-min interval) resolutions and provides more cloud products than the FY-2 series. In addition, a full-disc image offered by FY-4A covers a large area (80.6°N–80.6°S, 24.1°E–174.7°W). Detailed descriptions of FY-4A along with its products are given by Yang et al. (2017) and Min et al. (2017).

In this paper, we use the AGRI/FY-4A level-2 full-disc coverage CFR and CTP cloud products at the 4-km spatial resolution. In most cases, full-disc coverage CFR and CTP images are available at the 15-min interval. However, sometimes problems associated with instruments or observations may lead to longer time intervals. Since the FY-4A geostationary satellite uses the geostationary orbit nominal projection and the geographic coordinates are calculated by the WGS84 reference ellipsoid, the column and line numbers of nominal projection are transformed into the geographic latitude and longitude values here. Detailed instructions for the coordinate transformation can be downloaded from ( http://satellite.http://satellite.nsmc.org.cn/PortalSite/StaticContent/DocumentDownload.aspx?TypeID=3).

The NASA Terra and Aqua satellites were launched in December 1999 and May 2002, respectively. MODIS is a key instrument aboard both the Terra and Aqua platforms, which makes MODIS datasets more spatially complementary. The MODIS instrument has 36 spectral bands distributed between visible and infrared (0.41 and 14.5 μm), with 2 visible bands at the 250-m spatial resolution, 5 visible bands at the 500-m spatial resolution, and the remaining 29 visible and infrared bands at the 1-km spatial resolution. Cloud products from MODIS include the cloud mask, CFR, cloud top properties, cloud optical, microphysical properties, etc. Detailed descriptions of MODIS cloud products along with their algorithms are given by Ackerman et al. (1998, 2008), King et al. (2003), Platnick et al. (2003), Frey et al. (2008), and Menzel et al. (2008). In this study, CFR and CTP from MODIS/Terra and MODIS/Aqua level-2 cloud products are used. Their spatial resolutions are 1 or 5 km (at nadir), respectively, and temporal coverage are once or twice per day. Both instruments provide 288 product files per day, and each file includes data at the 5-min interval. MODIS cloud products are obtained from https://modis-atmos.gsfc.nasa.gov/.

Instead of directly using the daily or monthly mean datasets from AGRI/FY-4A, MODIS/Terra, and MODIS/Aqua, we have reconstructed new datasets over East Asia (10°S–45°N, 60°–140°E) based on the above products to calculate time lags between observations of the two instruments and avoid possible errors caused by the lower temporal coverage of MODIS products compared to that of AGRI/FY-4A over the same region.

Taking the MODIS/Terra level-2 CFR data as an example, the method is described as below. First, the numbers (1–4) of observation models for the two instruments are defined and listed in Table 1. AGRI/FY-4A provides full-disk images at the 15-min interval, while each MODIS/Terra scan covers a 5-min time interval. The observation model number is defined as “1”, “2”, “3”, and “4”, if the observation of AGRI/FY-4A or MODIS/Terra is from the following 0–15-, 15–30-, 30–45-, and 45–60-min time intervals, respectively. Second, the AGRI/FY-4A level-2 CFR data on specific days are transformed into new data with geographic latitude and longitude information. The MODIS/Terra level-2 CFR data over East Asia on these specific days are then selected and each scan is constructed to a 1° × 1° resolution record. Note that default values in individual scan are excluded. The exact time (determined based on the observation model number) and location of each scan from MODIS/Terra can be compared with those from AGRI/FY-4A. As shown in Fig. 1, there is a MODIS/Terra CFR scan over East Asia at 1005 UTC and the scan is constructed to a record with the 1° × 1° resolution (the blue box denotes the MODIS scanning region). An AGRI/FY-4A full-disk coverage CFR image in the closest time is at 1000 UTC (the black box) and it is also constructed to a data record with the 1° × 1° resolution over the same MODIS/Terra scanning region. Next, the two CFR data records from AGRI/FY-4A and MODIS/Terra over the same MODIS/Terra scanning region are compared and the time lag between them is calculated, defined as the absolute value of the observation model number from AGRI/FY-4A minus that from MODIS/Terra multiplied by 15 min (in the example here, |1–1| × 15 min = 0 min). Note that data from both AGRI/FY-4A and MODIS/Terra are not used if the time lag between them is more than 45 min for the scan. The daily mean CFR is the entire day’s average of the constructed AGRI/FY-4A and MODIS/Terra CFR at each grid over East Asia. In fact, there are about 30 scans over East Asia for a whole day and not all of them overlap, therefore the average is calculated by multiple scans in some regions and by only one scan in some other regions. The time lag on a specific day is calculated in the same way. Finally, the daily mean CFR data are averaged to produce the monthly mean data. The method described above can help us to compare geostationary satellite data with polar orbiting satellite data in a reasonable way and to obtain observation time lags between the two different datasets in the same region. A similar method can be found in Liu et al. (2017).

Figure 1 A schematic diagram of one AGRI/FY-4A full-disc image (black box) and one MODIS/Terra scan (blue box). The red box represents the study domain (East Asia).

Figure options

To assess the constructed MODIS/Terra level-2 CFR daily data, Figs. 2a–c show the constructed MODIS/Terra level-2 daily CFR data, MODIS/Terra level-3 daily CFR data (daily cloud products with the 1° × 1° resolution), and their difference on 9 August 2017. It is evident that Figs. 2a, b show similar spatial patterns. The proportion of areas where the absolute differences between the two datasets are less than 5% (20%) reaches 45.6% (75.1%) of the study area, and areas with absolute differences greater than 50% only account for 2.5% of the total study area. The results demonstrate that the constructed MODIS/Terra level-2 CFR daily data can well capture CFR characteristics. The constructed MODIS/Aqua level-2 CFR daily data are generally consistent with MODIS/Aqua level-3 daily data (results omitted). In addition, there are only slight differences between the constructed level-2 CFR daily data from MODIS/Terra and that from MODIS/Aqua, as shown in Figs. 2a, d, e. Areas where absolute differences between the two datasets are less than 5% (20%) account for 43.3% (72.3%) of the study area and areas with absolute differences greater than 50% only account for 4.6% of the study area.

Figure 2 (a) The constructed MODIS/Terra level-2 CFR daily data on 9 August 2017, (b) MODIS/Terra level-3 CFR daily data on the same day, and (c) their difference (Fig. 2a minus Fig. 2b). (d) The constructed MODIS/Aqua level-2 CFR daily data on the same day and (e) the difference (Fig. 2a minus Fig. 2d). Shaded areas in Figs. 2c, e are areas with the absolute difference greater than 5%. The black areas represent missing values.

Figure options

The consistency between the constructed MODIS/Terra level-2 and level-3 CTP daily data is also examined. The results show that the areas with absolute differences between the two CTP datasets less than 50 (200) hPa account for 30.5% (71.5%) of the study area. For MODIS/Aqua CTP, the areas with absolute differences between the constructed level-2 and level-3 daily data less than 50 (200) hPa account for 33.4% (77.5%) of the study area. The above results indicate that the constructed MODIS/Terra and MODIS/Aqua level-2 CTP daily data can well capture CTP characteristics. Further, the differences between the constructed level-2 CTP daily data from MODIS/Terra and that from MODIS/Aqua are also slight.

The above comparisons of the constructed level-2 CFR and CTP daily data with those from MODIS/Terra and MODIS/Aqua indicate that the constructed level-2 CFR and CTP daily data are reliable and can be used to evaluate the constructed FY-4A level-2 CFR and CTP daily data. In the following, we will only present the results of comparison between MODIS/Terra and AGRI/FY-4A CFR (CTP) since only slight differences exist between the constructed level-2 CFR (CTP) daily data from MODIS/Terra and that from MODIS/Aqua.

Figure 3 shows the CFR distributions from the constructed monthly MODIS/Terra and AGRI/FY-4A datasets (Figs. 3a, b), their differences (Fig. 3c), and the observation time lags between them (Fig. 3d) in August 2017. The spatial patterns of CFR from MODIS/Terra and AGRI/FY-4A agree well during the study period, although the AGRI/FY-4A CFR has relatively higher values in some regions than the MODIS/Terra CFR. Areas where absolute differences between CFRs from the two instruments are less than 5% (20%) account for 26.7% (79.0%) of the study region. The areas with absolute differences of CFR greater than 50% only account for 0.7% of the study region. The area with the longest time lag (15–20 min) is located in the Indian Ocean, where the maximum time lag is 16.6 min. Note that the area with the longest time lag does not correspond to that with the largest CFR discrepancy as shown in Figs. 3c, d. Large CFR discrepancies are concentrated at the high-elevation areas (elevation greater than 1000 gpm), especially in the vicinity of TP (28°–45°N, 75°–105°E). To investigate the contribution of long time lags to CFR discrepancies, Fig. 4 presents scatter plots of the two datasets in August 2017 in the latitude band of 0 to 10°S over Southeast Asia at different time lags. The two datasets generally become less consistent with the longer time lag. The linear correlation coefficient (R) between the two datasets is 0.89 when the time lag is from 0 to 5 min, while it falls to 0.74 when the time lag is from 15 to 20 min. The mean CFR difference (MD) is 7.4% when the time lag is from 0 to 5 min, while it increases to 11.1% when the time lag is from 15 to 20 min. Therefore, the time lag is one factor responsible for the CFR discrepancy and the contribution of the time lag to CFR discrepancy depends on the length of time lag and the rate of CFR change with time.

Figure 3 The constructed (a) AGRI/FY-4A and (b) MODIS/Terra level-2 CFR monthly data in August 2017, (c) their difference (Fig. 3a minus Fig. 3b), and (d) their observation time lag. Shaded areas in Fig. 3c are areas with the absolute difference greater than 5%. Contour lines denote the altitude (km).

Figure options

Figure 4 Scatter plots of AGRI/FY-4A and MODIS/Terra CFR monthly data in August 2017 within the latitude band of 0 to 10°S over Southeast Asia at different time lags. The points and solid lines denote the value of CFR and results of the least squares linear fitting respectively, which all pass the statistically significant test at the 99% confidence level. The legend gives the number of samples in the specified time lags (Samples), R, MB, RMSE, and MD. See the text for formulas used to calculate MD, MB, and RMSE.

Figure options

The formulas used to calculate MD, mean bias (MB), and root mean square error (RMSE) are as follows:

where y₁ denotes AGRI/FY-4A CFR, y₂ denotes MODIS/Terra CFR, and ${{y}}_1{\rm{'}}$ denotes regressing AGRI/FY-4A CFR onto MODIS/Terra CFR.

However, large CFR differences mainly appear in the high-elevation areas, especially in the vicinity of TP, where the mean time lag is 9.8 min, and 69% of the maximum CFR differences appear near the 5000-gpm height. Large CFR differences are not collocated with long time lags as shown in Fig. 3, and the inconsistency between the two datasets does not become worse with the longer time lag in the study region as shown in Fig. 5. The above results suggest that the time lag partly contributes to the inconsistency between the two datasets, while other factors like the elevated terrain may also be responsible for the inconsistency. The constructed daily distributions of CFR from the two datasets are examined to further test this conclusion. Daily distributions of CFR (Fig. 6) indicate that on some days in August 2017, time lags between observations of the two instruments are very small (approaching to 0 min), but large CFR differences still appear near the TP.

Figure 5 As in Fig. 4, but over East Asia at different time lags and regardless of time lags. The legend gives the number of samples in the specified time lags (Samples), R, MB, and RMSE.

Figure options

Figure 6 As in Fig. 3, but for daily data on 1 August 2017. Areas shaded in black represent missing values.

Figure options

To further explore other factors contributing to large CFR differences over East Asia, Fig. 7 shows the probability densities of CFR in the two datasets in August 2017 over East Asia. There are large probability density differences for CFR within 0–1% and 80%–85%, with their differences minus the mean value greater than twice the standard deviation. This result indicates that there exist large discrepancies in the probability densities of CFR under the clear-sky condition (CFR = 0). Further analysis shows that high values of the probability density under clear-sky in MODIS/Terra and low values of the probability density under clear-sky in AGRI/FY-4A are responsible for large CFR probability density discrepancies between the two products.

Figure 7 Probability densities of AGRI/FY-4A and MODIS/Terra CFR data over East Asia in August 2017. The CFR interval is 1%. The red and black lines denote AGRI/FY-4A and MODIS/Terra CFR data respectively.

Figure options

To demonstrate the above results, the AGRI/FY-4A CFR and MODIS/Terra CFR are separately considered under four different situations each day, i.e., Situation I: cloudy-sky CFR in AGRI/FY-4A (CFR greater than 1%) and clear-sky CFR in MODIS/Terra (CFR less than 1%), Situation II: cloudy-sky CFR in both AGRI/FY-4A and MODIS/Terra (CFR greater than 1%), Situation III: clear-sky CFR in AGRI/FY-4A (CFR less than 1%) and cloudy-sky CFR in MODIS/Terra (CFR greater than 1%), and Situation IV: clear-sky CFR in both AGRI/FY-4A and MODIS/Terra (CFR less than 1%). Daily mean values are then averaged to obtain the monthly mean values under the four different situations. Differences between the AGRI/FY-4A and MODIS/Terra products (AGRI/FY-4A CFR minus MODIS/Terra CFR) under the four situations are displayed in Fig. 8. Large CFR differences near the TP are obvious in Fig. 8a, which are the main results of different cloud detections by MODIS/Terra and AGRI/FY-4A in this region, where the clear-sky situation in MODIS/Terra actually corresponds to the cloudy-sky situation in AGRI/FY-4A. In addition, Situation II also partly contributes to large CFR differences near the TP (Fig. 8b). Percentage contributions to large CFR differences (greater than 50%) near the TP show that Situations I and II account for 61% and 39% of the differences respectively. Note that the results are very similar whether clear-sky is determined by the criterion that CFR is equal to 0 or less than 1%. Therefore, the difference in the cloud detection, i.e., clear-sky in MODIS/Terra and cloudy-sky in AGRI/FY-4A (Fig. 8b), is mainly responsible for large CFR differences near the TP. The incorrect correspondence between the low values of MODIS/Terra CFR and high values of AGRI/FY-4A CFR also partly contributes to large CFR differences between MODIS/Terra and AGRI/FY-4A products. Time lags are short and make a small contribution to the differences near the TP.

Figure 8 CFR differences under four situations in August 2017: (a) cloudy-sky CFR from AGRI/FY-4A minus clear-sky CFR from MODIS/Terra, (b) cloudy-sky CFR from AGRI/FY-4A minus cloudy-sky CFR from MODIS/Terra, (c) clear-sky CFR from AGRI/FY-4A minus cloudy-sky CFR from MODIS/Terra, and (d) clear-sky CFR from AGRI/FY-4A minus clear-sky CFR from MODIS/Terra. Contour lines denote the altitude (gpkm).

Figure options

Overall, the comparison between the two datasets indicates that the AGRI/FY-4A CFR data are spatially and temporally consistent with MODIS/Terra CFR data. Statistically, despite time lags, the R value between the two datasets is 0.83, MB is 8.6%, and RMSE is 11.2% (Fig. 5). However, large differences in CFR between the two datasets are found in the high-elevation areas, especially in the vicinity of TP (28°–45°N, 75°–105°E), where CFR from AGRI/FY-4A is higher than that from MODIS/Terra.

A method similar to that described in Section 3.1 is used to compare CTP from MODIS/Terra and AGRI/FY-4A measurements. Figure 9 shows spatial distributions of CTP from MODIS/Terra and AGRI/FY-4A measurements, differences between them, and observation time lags in August 2017 over East Asia. Spatial patterns of CTP from the two instruments are generally consistent with each other during the study period. A high (low) CTP denotes a high (low) value of pressure and a low (high) value of the geopotential height. The area where the absolute difference between AGRI/FY-4A CTP and MODIS/Terra CTP is less than 50 (200) hPa accounts for 19.0% (93.3%) of the study area. From the perspective of time lag, Fig. 10 shows that the two datasets generally become less consistent as the time lag increases, which indicates that the time lag is one factor that contributes to the CTP discrepancy. Note that for certain time lags, only a very small number of unrepresentative samples are available, which are not shown in Fig. 10.

Figure 9 As in Fig. 3, but for CTP monthly data. Shaded areas in Fig. 9c are areas with the absolute difference greater than 50 hPa. Areas shaded in black represent missing values.

Figure options

Figure 10 As in Fig. 5, but for CTP monthly data. For certain time lags, only a very small number of unrepresentative samples are available, so corresponding figures are omitted.

Figure options

The regions with large CTP discrepancies are mainly located in the eastern Iranian Plateau (IP; 25°–45°N, 60°–70°E) as shown in Fig. 9. A few samples with long time lags (from 20 to 35 min) and low R values are found there, which contribute to the large CTP discrepancy in the IP. Furthermore, the missing values of CTP from both the instruments are also found in this area, because there are more clear-sky grids in the IP than in the other regions during the study period (Fig. 3). The elevated terrain contribution to different cloud detections by the two instruments has been demonstrated in Section 3.1. With more clear-sky grids in the IP, the discrepancy in the cloud detection by the two instruments is larger than that in the other plateaus like the TP (Fig. 11). In addition, when clear-sky conditions are detected by MODIS/Terra, cloudy-sky conditions can be also detected by AGRI/FY-4A in both the IP and TP. In this situation, however, CFR in the IP is always less than that in the TP. Furthermore, the contribution of Situation Ⅱ (i.e., low CFR from MODIS/Terra and high CFR from AGRI/FY-4A) is smaller in the IP than in the TP. Thus, the CFR discrepancy over the IP is less than that over the TP.

Figure 11 The number of times when MODIS/Terra is clear-sky (CFR = 0) and AGRI/FY-4A is cloudy-sky (CFR > 0) in August 2017. Contour lines denote the altitude (km).

Figure options

In addition, we only compare real-time CTP in the two products, therefore the elevated terrain does not influence the quality of CTP products derived from observations of the two instruments themselves, but makes the number of compared CTP samples less in the IP than in the other regions during the study period. The comparison of the CTP sample number in the IP (25°–45°N, 60°–70°E) with that in the TP (25°–45°N, 80°–90°E) at different time lags during the study period is shown in Fig. 12. For certain time lags, only a very small number of unrepresentative samples are available, which are not shown in Fig. 12. The areas over the IP and TP selected for comparison are the same, whereas the total number of samples in the IP (1565) is much less than that in the TP (5563) and the consistency of the two datasets is also worse in the IP than that in the TP. The daily samples that can be used to average into monthly means in the IP are fewer than those in the TP, which is one reason for the large CTP discrepancy in the IP.

Figure 12 Scatter plots of AGRI/FY-4A and MODIS/Terra CTP monthly data in August 2017 over the (a) IP (25°–45°N, 60°–70°E; first row) and (b) TP (25–45°N, 80°–90°E; second row) at different time lags. The points and solid lines denote the value of CTP and results of the least squares linear fitting, which all pass the statistically significant test at the 99% confidence level. The Samples denote the number of samples in the specified time lags.R denotes the linear correlation coefficient. For other time lags, only a very small number of unrepresentative samples are available, socorresponding figures are omitted.

Figure options

Overall, the comparison between the two datasets indicates that the AGRI/FY-4A CTP measurements have a good spatial correspondence with MODIS/Terra CTP. As shown in Fig. 10, the correlation coefficient between the two products is 0.80, MB is 56.8 hPa, and RMSE is 80.0 hPa regardless of time lags. Areas with the greatest differences of CTP between the two datasets are found in the eastern IP (25°–45°N, 60°–80°E), where CTP from AGRI/FY-4A is higher than that from MODIS/Terra.

Previous work has shown that CFR derived from the first generation of the geostationary Fengyun satellite (FY-2C) shows consistencies with CFR derived from other datasets over Southeast Asia (Jin et al., 2009). The present study is to demonstrate that CFR (CTP) derived from the second generation of the geostationary Fengyun satellite (FY-4A) agrees well with those derived from MODIS/Terra over East Asia.

Though the CFR and CTP data from AGRI/FY-4A and MODIS/Terra generally agree well over East Asia in August 2017, areas of the largest discrepancy occur over plateau regions. Our analysis shows that time lags between the datasets cannot fully explain the occurrence of these discrepancies, and we suggest that terrain effects should be a possible factor. Liu (2009) pointed out that the accuracy of cloud detection is greatly affected over plateau areas by low temperatures in the underlying surface and snow cover there. Our work shows that the CFR discrepancies occur most often under MODIS clear-sky and AGRI/FY-4A cloudy-sky conditions (accounting for 61% of the CFR discrepancy that is greater than 50% near the TP). Even when clouds are detected by both MODIS/Terra and AGRI/FY-4A, the incorrect correspondence between high values of AGRI/FY-4A CFR and low values of MODIS CFR again is the reason, accounting for 39% of the CFR discrepancy, greater than 50% near the TP. The difference in CFR between MODIS/Terra and AGRI/FY-4A observations probably also results from either instrument differences or different algorithms for the CFR calculation. It is necessary to conduct more validation studies in the future to sort out the contributing factors in the CFR discrepancy.

Discrepancies in CTP derived from AGRI/FY-4A and MODIS/Terra are most prominent above the IP, but not above the TP. In this case, factors other than terrain are likely important. There are a few samples with long time lags (from 20 to 35 min) and low R values between the two datasets. There are also many samples with short time lags in the IP, but the consistency between the two datasets is still low. This is because there are many clear-sky grids over the IP during the study period, and thus fewer CTP daily data are available to be averaged into monthly means in the IP than in the other regions of East Asia.

This study has demonstrated a method to match measurements by polar orbiting satellites with those by geostationary satellites over the same region, by using techniques to determine the geospatial coincidence and account for time lags in the observations. The method is applied to compare CFR (CTP) level-2 daily data from AGRI/FY-4A and MODIS/Terra over East Asia in August 2017.

The results demonstrate that for both CFR and CTP, the data from AGRI/FY-4A and MODIS/Terra generally agree well over East Asia during the study period. The R value between the two datasets for CFR (CTP) is 0.83 (0.80), MB is 8.6% (56.8 hPa), and RMSE is 11.2% (80.0 hPa) regardless of time lags. In most cases, the consistency of CFR/CTP data from AGRI/FY-4A and MODIS/Terra is better with shorter time lags, when considering their time lags. This suggests that long time lags contribute to the inconsistencies that are observed between CFR and CTP datasets derived from the two instruments. However, large CFR/CTP discrepancies don’t always match up with long time lags, because of the influence of other factors. The agreement between CFR/CTP data is better to the south of 25°N than to the north of 25°N over plateau areas in East Asia, indicating that the plateau terrain plays an important role in discrepancies of CFR/CTP data from the two satellites. Large CFR discrepancies in the two datasets mainly appear in the vicinity of TP (28°–45°N, 75°–105°E), where high values of AGRI/FY-4A CFR incorrectly correspond to low values of MODIS/Terra CFR. For CTP, the largest discrepancies are above the IP, where again the values are higher for the AGRI/FY-4A dataset.

Despite some areas of discrepancies, our results demonstrate that observations from the second-generation geostationary FY-4A satellite can well capture the characteristics of clouds over East Asia. This platform offers a higher temporal coverage over the same region compared with instruments (such as MODIS) aboard polar orbiting satellites, and thus having advantages for the climate, meteorological, and hydrological studies. However, since the FY-4A data have been available for a relatively short period, only CTP and CFR products for a certain month are compared in this paper. The evaluation of long-term and seasonal variations with more observations and for a broader set of cloud properties should be conducted in the future.

Acknowledgments. We thank Zhe Xu, Xiaohu Zhang, Xi Wang, and Fu Wang from the National Satellite Meteorological Center (NSMC) of China Meteorological Administration for their guidance on the FY-4A cloud products. We also thank the NSMC for providing FY-4A datasets and the MODIS Atmosphere Group for providing MODIS datasets.