The Ka-band polarization Doppler radar, using a frequency of 35.075 GHz, situated at the IAP (39.967°N, 116.367°E), Beijing, China, was set up in 2010. The technical specifications of the Ka-band radar at the IAP are given in Table 1.
Parameter Ka polarization Doppler radar Ka-FMCW (wide pulse mode) Ka-FMCW (narrow pulse mode) Wavelength (mm) 8.55 8.569 8.569 Peak power (kW) 29 0.04 0.04 Pulse width (µs) 0.2 120 10 Transmitter type Magnetron Solid-state Solid-state Antenna gain (dB) 54 50.9 50.9 Noise factor (dB) 5.8 4.2 4.2 Scanning mode Vertically pointing Vertically pointing Vertically pointing Vertical resolution (m) 30 30 30
Table 1. Main technical specifications of the Ka radars at the IAP and YBJ sites.
The Ka-band radar at YBJ is one part of the Atmosphere Profiling Synthetic Observation System, which was the first ground-based observation system for profiling multiple atmospheric variables and constituents from the surface up to the thermosphere (Lu et al., 2018). The radar was set up in October 2017 and began continuous observations on 23 July the following year. This Ka-band radar, with a frequency-modulated continuous-wave (FMCW), has two pulse modes: a 120 µs wide pulse mode and a 10 µs narrow pulse mode. During radar operation, switching between the two modes takes place automatically to ensure that the cloud is detected with the highest possible accuracy. The technical specifications of the Ka-band radar at YBJ are also given in Table 1 [Ka-FMCW (wide pulse mode) and Ka-FMCW (narrow pulse mode)].
Except for special cases, the two Ka radars work 24 hours a day in vertically pointing mode. A threshold of −45 dBZ was used to identify the cloud in this study. For an arbitrary radar profile, it was considered to be cloudy if there were more than three radar bins with radar reflectivity greater than −45 dBZ (Huo et al., 2020a). For a cloudy profile, the CTH was determined as the height of the cloudy bin at the highest level. To facilitate comparison with satellite data, for clouds detected in a certain period (i.e., within 10 min or 15 min), the radar CTH was calculated as the average CTH of all cloudy profiles, but not for upper-level clouds if there were multilayer clouds present. It can be seen from Table 1 that the radar measures three profiles per second with vertically pointing mode and the vertical resolution is 30 m.
On 11 December 2016, the first of China’s new-generation geostationary meteorological satellite series, FY-4A, was successfully launched. It has been in operation by the National Satellite Meteorological Center, China Meteorological Administration (NSMC/CMA) since 1 May 2018. FY-4A has four payload instruments. The AGRI, with 14 spectral bands in the visible, near-infrared, and thermal infrared regions, is one of them, from which the CTH products are retrieved (Yang et al., 2017). The AGRI has three scanning modes: full disk (images of the whole Earth as seen from the satellite) every 15 min; Chinese area (3°–55°N, 70°–140°E) every 5 min; and target area (1000 km × 1000 km) every 1 min.
The AGRI uses radiances (temperature brightness) of two infrared split window channels of 10.8 µm (channel 12) and 12 µm (channel 13), as well as a 13.5 µm (channel 14) CO2 absorption channel, to retrieve the CTH by applying the Fengyun Cloud Top Height Algorithm (FCTHA) (Min et al., 2017; Wang and Zhao, 2020). The core of the FCTHA involves a one-dimensional variational method to retrieve the cloud top temperature based on the simulations of a radiative transfer model. The parameters applied include the brightness temperature of channel 12, the brightness temperature difference between channels 12 and 13, and the brightness temperature difference between channels 12 and 14. The CTH is obtained according to the atmospheric temperature profile obtained by a numerical prediction model. Moreover, the FCTHA performs a special process for multi-layer cloud pixels, and the CTH of the lower-layer cloud is estimated from other surrounding low clouds. The AGRI CTH product used in this study was obtained from the NSMC/CMA. The temporal resolution of the product is 15 min and the spatial resolution is 4 km.
As one of the new generations of Japanese geostationary meteorological satellites, the Himawari-8 satellite was successfully launched from Japan’s Tanegashima Space Center on 7 October 2014 and settled in geostationary orbit on 16 October. The Japan Meteorological Agency began operating the satellite on 7 July 2015 (Bessho et al., 2016). The satellite’s AHI is greatly improved over those of the MTSAT series (Multi-functional Transport Satellites—previous Japanese geostationary satellites) in terms of the number of bands, spatial resolution, and temporal frequency. The AHI has 16 spectral bands (three for visible, three for near-infrared, and ten for infrared) and observes the Japanese area and some other target or landmark areas every 2.5 min, and the entire full disk every 10 min, with a spatial resolution of 0.5–2.0 km. The scan ranges for full disk and the Japanese area are preliminarily fixed, while those for the target and landmark areas are flexible according to meteorological conditions.
The AHI CTH retrieval algorithm uses radiative transfer codes developed by the European Organisation for Meteorological Satellites and the temperature and humidity profile data obtained from a numerical weather prediction model to calculate the radiance at four infrared bands (6.2, 7.3, 11.2, and 13.3 µm) (Eyre, 1991; Iwabuchi et al., 2016). The algorithm includes the interpolation method, the CO2-slicing method, and the intercept method; the appropriate method is then selected according to the cloud type in the AHI cloud type product (Nieman et al., 1993; Schmetz et al., 1993; Mouri et al., 2016; Huo et al., 2020a; Letu et al., 2020). The interpolation method is used for opaque and fractional clouds and the intercept method is suitable for translucent clouds. In the case of optically thin (or translucent) cloud retrieval, the intercept method, the CO2-slicing method, and the interpolation method were used one after another until appropriate results were obtained. The AHI CTH product used in this study was the Himawari-8 Cloud Property data released through the P-Tree System of the Japan Aerospace Exploration Agency (JAXA). The temporal resolution of the product is 10 min and the spatial resolution is 5 km.
The retrieval of the satellite CTH is based on the radiance observed at the satellite, which includes the radiation emitted from the surface, the contribution from the atmosphere below the cloud, the cloud contribution, and the contribution from the atmosphere above the cloud— all the way to the top of the atmosphere (Liou, 2002), p. 403). Therefore, the errors of the satellite CTH retrieval may be partly produced by the surface radiation and the contributions by the atmosphere. This study mainly focuses on quantifying the CTH retrieval differences between different areas and analyzing the relationships of these differences with the physical properties of the underlying surface and atmosphere.
The 2-m temperature data are from the ERA5 dataset of the ECMWF (European Centre for Medium-Range Weather Forecasts). The surface emissivity data used in this paper are the Collection-6 MODIS Land Surface Temperature products (MOD11_L2) from the Aqua and Terra satellites. Finally, the aerosol optical depth (AOD) data used in this study are the Himawari-8 Aerosol Property data released through JAXA’s P-Tree System.
The satellites require several minutes to make a full disk scan and the field of view is larger than that of the radar. Data allocations between the satellites and radars are therefore required. In this study, we used a similar data collocation method as in Huo et al. (2020a). Due to the satellites’ viewing geometries, an AGRI CTH pixel has a fixed 4 km × 4 km spatial resolution and 15-min temporal resolution, whereas for an AHI CTH pixel the corresponding quantities are 5 km × 5 km and 10 min, over the IAP and YBJ site. Since the AGRI presents data about every 15 min, the Ka radar data within 15 min of the AGRI observation time are extracted and averaged (from observation start time to observation end time of the AGRI). The average AGRI CTHs of the four grids nearest to the sites are used for comparison. Because the AHI presents data every 10 min, the Ka radar data within 10 min of the AHI observation time are extracted and averaged (±5 min). The AHI CTHs nearest to the sites are used for comparison.
Figure 1 shows two cases from the AHI, AGRI, and radar on 8 August 2019 at YBJ, and on 16 June 2019 at IAP. It should be noted that the elevation of Beijing and YBJ are 0.04 km and 4.3 km, respectively. The satellite CTH is the CTH relative to mean sea level (NASA, 2020). For the convenience of comparison with radar, the satellite CTH at YBJ throughout this study was the original satellite CTH minus 4.3 km. In this paper, the CTH difference (CTHD) between radar and satellite (radar CTH minus satellite CTH) is calculated to quantify the discrepancy. The CTHD between the radar and the FY-4A satellite is termed CTHDrf, and that between the radar and the Himawari-8 satellite is termed CTHDrh.
2.1. Ka-band radar
2.2. FY-4A satellite
2.3. Himawari-8 satellite
2.4. Other data
2.5. Data collocation method
Here, the ratio of the radar observation time in a month to the total time was defined as the data acquisition rate (DAR). The DAR in each month at the two sites from February to August 2019 is shown in Fig. 2a. The IAP radar missed some observations in February, March, and April; most notably in April, the DAR was only 3.99%, while that of the YBJ radar was above 95% in all months except February. During the period, the average DAR of the IAP radar was 72.88% and that of the YBJ radar was 95.15%.
Figure 2. (a) Percentage of radar observation times to all times at YBJ and IAP in each month from February through August 2019, and (b) the data numbers expected and obtained from the two satellites at both sites during the same period. (c, d) The number of comparison cases from Himawari-8 and FY-4A for (c) all and (d) high-level clouds from February through August 2019 at the YBJ and IAP sites.
The AGRI instrument onboard FY-4A makes a full-disk observation every hour and three consecutive full-disk observations every three hours. Each full-disk observation takes 15 min, so FY-4A normally generates 40 full-disk CTH files a day. The Himawari-8 satellite requires 10 min to make a full disk observation and normally generates 144 full-disk CTH files per day. Figure 2b shows the expected data number and the actual available data number from FY-4A and Himawari-8 during the study period for both sites. For CTH comparison, only the effective CTHs (i.e., CTH > 0) retrieved by both satellite and radar (termed valid collocation in this paper) were selected. The number of valid collocations from FY-4A and radar was 2465 at IAP and 4754 at YBJ, accounting for 29.4% and 56.7% of all satellite observations (termed the collocation ratio), respectively. The number of valid collocations from Himawari-8 and radar was 4374 at IAP and 7473 at YBJ, accounting for 14.6% and 24.9% of all satellite observations, respectively. The collocation ratio of FY-4A is about two times that of Himawari-8 at both sites. Additionally, the collocation ratio of both satellites at YBJ is about two times that at IAP.
The data collocation ratio also changed between daytime and nighttime. Figure 2c illustrates that the collocation ratios from Himawari-8 at night for all clouds were 13.79% and 0.08% at IAP and YBJ, respectively, while they were 43.73% and 36.85% from FY-4A at IAP and YBJ, respectively. There was a clear discrepancy between the two satellites. The collocation ratio at night from Himawari-8 is significantly lower than that from FY-4A at both sites, which means that Himawari-8 might neglect more CTHs at night compared to the radar and FY-4A. A similar feature was also observed for high clouds (which will be analyzed below), as shown in Fig. 2d. The reason for this discrepancy will be explained in sections 4.1 and 5.
Table 2 shows the quantified statistics of CTHD at both sites and Figs. 3a–d show scatterplots of the CTHs retrieved from radar and satellites. Figure 4a shows the probability density distributions of the CTHDs. Statistically, the CTHDrf ranged from −8.49 km to 14.20 km; the mean of the CTHDrf was 0.06 km; the standard deviation (STD) of the CTHDrf was 1.90. The CTHDrh ranged from −11.58 km to 9.96 km; the mean of the CTHDrh was −0.02 km; the STD of the CTHDrh was 2.40. Statistically, the CTHDrf ranged from −9.81 km to 14.84 km; the mean CTHDrf was 0.93 km; the STD of the CTHDrf was 2.24. The average CTH at IAP is about 3 km higher than that at YBJ. The CTHDrh ranged from −12.92 km to 11.26 km; the mean CTHDrh was 0.99 km; the STD of the CTHDrh was 2.37.
Site Satellite Mean radar CTH (km) Mean satellite CTH (km) Correlation coefficient Mean CTHD (km) Median CTHD (km) IQR of CTHD (km) STD of CTHD (km) YBJ, Tibet FY-4A 4.11 4.06 0.61 0.06 0.06 2.14 1.90 Himawari-8 4.05 4.06 0.57 −0.02 −0.07 2.41 2.40 IAP, Beijing FY-4A 7.88 6.95 0.72 0.93 0.69 2.24 2.24 Himawari-8 7.99 7.00 0.69 0.99 0.69 2.27 2.37
Table 2. Statistics of the CTHD at the YBJ and IAP sites.
Figure 3. Scatterplots of radar and satellite CTHs at both sites: (a) YBJ radar and FY-4A; (b) YBJ radar and Himawari-8; (c) IAP radar and FY-4A; (d) IAP radar and Himawari-8. A one-to-one line is included in each figure for comparison.
Figure 4. The probability density distribution of the CTHD: (a) average CTHD at both sites; (b) CTHDrh for different cloud levels at YBJ; (c) CTHDrf for different cloud levels at YBJ; (d) CTHDrh for different cloud levels at IAP; (e) CTHDrf for different cloud levels at IAP.
From Table 2 it can be seen that the correlation coefficients of FY-4A and Himawari-8 with radar CTHs were 0.61 and 0.57 at YBJ, and 0.72 and 0.69 at IAP, respectively, all of which demonstrate good agreement between radar and both satellites. FY-4A and Himawari-8 retrievals agree quite well, with a difference of 0.08 km between the CTHDrf and the CTHDrh at YBJ, and the CTHDrh being 0.06 km higher than the CTHDrf at IAP, on average. The CTHs at YBJ are distributed symmetrically along the diagonal lines shown in Figs. 3a and b, and the distributions in Fig. 4a are not positively skewed. Different from the results at YBJ, most CTHs at IAP lie above the diagonal lines, as shown in Figs. 3c and d, and the distributions in Fig. 4a appear positively skewed, which means the satellite CTH retrievals at YBJ are more reasonable and both satellites underestimated the CTHs when compared with the radar at IAP.
It should be noted that the statistical results at YBJ contain some satellite CTHs lower than zero (see Figs. 3a and b). This is because the satellite CTH used here was equal to the original satellite CTH minus 4.3 km of surface altitude. The negative CTHs of FY-4A and Himawari-8 accounted for 2.27% and 8.91% of all collocations at YBJ, respectively. Obviously, these negative data have uncertainties.
Previous research has shown that the performances of satellite retrievals of CTH vary with the different cloud levels (Wang et al., 2018b). In this paper, comparisons of CTH retrievals are discussed and characterized according to the cloud-base height (CBH) observed by radar, wherein clouds with CBH ≥ 6 km are defined as high-level clouds, < 6 km but ≥ 2 km as mid-level clouds, and < 2 km as low-level clouds. Of all 4754 FY-4A and radar collocations at YBJ, the percentages of high-, mid-, and low-level cloud collocations were 5.70%, 64.64%, and 29.66%, respectively. Meanwhile, of all 7473 Himawari-8 collocations at YBJ, the percentages of high-, mid-, and low-level clouds collocations were 6.40%, 62.68%, and 30.92%, respectively. Of all 2465 FY-4A collocations at IAP, the percentages of high-, mid-, and low-level cloud collocations were 53.47%, 31.20%, and 15.33%, respectively. Among all 4374 Himawari-8 collocations at IAP, the percentages of high-, mid-, and low-level cloud collocations were 55.69%, 29.61%, and 14.70%, respectively. Tables 3 and 4 show the statistics of the CTHDs for different levels of clouds at the two sites, and the probability density distributions of the CTHDs for high-, mid-, and low-level clouds are shown in Figs. 4b–e.
Cloud type Radar CBH Satellite No. of samples Mean
High CBH ≥ 6 km FY-4A 271 0.99 0.40 2.95 2.26 Himawari-8 478 0.67 0.18 2.98 2.48 Middle 2 km ≤ CBH < 6 km FY-4A 3073 0.40 0.38 2.13 1.77 Himawari-8 4684 0.31 0.12 2.46 2.32 Low CBH < 2 km FY-4A 1410 −0.86 −0.64 2.00 1.77 Himawari-8 2311 −0.81 −0.61 2.49 2.34
Table 3. CTHDs for different levels of clouds at YBJ, Tibet.
Cloud type Radar CBH Satellite No. of samples Mean
High CBH ≥ 6 km FY-4A 1318 1.80 1.45 2.27 2.15 Himawari-8 2436 1.79 1.32 2.83 2.47 Middle 2 km ≤ CBH < 6 km FY-4A 769 0.38 0.39 1.73 1.79 Himawari-8 1295 0.29 0.36 1.72 1.77 Low CBH < 2 km FY-4A 378 −0.96 −0.55 1.57 1.79 Himawari-8 643 −0.61 −0.34 1.48 1.67
Table 4. CTHDs for different levels of clouds at IAP, Beijing.
The average CTHDrf of high-, mid-, and low-level clouds at YBJ was 0.99 km, 0.40 km, and −0.86 km, respectively, and the average CTHDrh was 0.67 km, 0.31 km, and −0.81 km, respectively. That is, both satellites underestimated the CTH of the high- and mid-level clouds but overestimated the CTH of the low-level clouds at YBJ when compared with radar. The average CTHDs of mid-level clouds were the smallest, all being < 0.5 km for both satellites, and the average CTHDs of high- and low-level clouds were higher. It can be concluded that the retrieval accuracy of the two satellites for high- and low-level clouds is worse than that for mid-level clouds at YBJ.
The average CTHDrf of high-, mid-, and low-level clouds at IAP was 1.80 km, 0.38 km, and −0.96 km respectively, and the average CTHDrh was 1.79 km, 0.29 km, and −0.61 km, respectively. That is, like the results at YBJ, both satellites underestimated the CTH of the high- and mid-level clouds but overestimated the CTH of the low-level clouds at IAP when compared with radar. The average CTHDs of mid- and low-level clouds were all < 1.0 km, and the average CTHDs of mid-level clouds were the smallest, all being < 0.5 km for both satellites; however, the average CTHDs of high-level clouds were the biggest, with 1.80 km for FY-4A and 1.79 km for Himawari-8. Therefore, the retrieval accuracy of the two satellites for high-level clouds is the worst at IAP, while that for mid-level clouds is the best.
In general, high clouds are thinner and more transparent, making it harder to meet the blackbody assumption for clouds, ultimately resulting in poor accuracy in top-height retrieval. As for low clouds, infrared-based, space-borne measurements have inherent difficulties in detecting the CTH because of the uncertainties in the assumed temperature profiles in the lower atmosphere, especially within the boundary layer (Huang et al., 2019).
Clouds with greater optical thickness are more likely to be regarded as blackbodies, which meets the assumptions required for CTH retrieval using infrared radiance. Thicker clouds generally have greater optical thickness. The relationship between CTHDs and cloud depth was examined in this study, and the results are presented in Fig. 5. It can be seen that the CTHDs decreased as the cloud depth increased at both YBJ and IAP. Figure 6a shows that the mean depth of high-level clouds was smaller than that of mid- and low-level clouds at both sites. Compared to optically thick cloud, it is more difficult to identify thin cloud from the surface (background) since it is somewhat transparent to the infrared radiation from the atmosphere and surface below. Also, optically thin clouds complicate CTH retrieval since the blackbody assumptions cannot be met (Hollars et al., 2004; Weisz et al., 2007). Although FY-4 and Himawari-8 have used specialized approaches to improve the accuracy regarding the CTH of thin clouds, the analysis here shows that the current retrieval performance is still worse for thin clouds than it is for thick clouds.
Figure 5. The CTHDs associated with various cloud depths at both sites: (a) CTHDrfs with cloud depth at YBJ; (b) CTHDrhs with cloud depth at YBJ; (c) CTHDrfs with cloud depth at IAP; (d) CTHDrhs with cloud depth at IAP.
Figure 6. The (a) cloud depth of different levels of clouds, (b) proportions of different levels of clouds, and (c) proportions of high-level clouds, high-thin clouds, and single-layer high-thin clouds, at the YBJ and IAP sites.
Previous research has shown that comparisons of CTH retrievals from satellite and radar will become extremely complicated when multilayer clouds exist, particularly for times when a thin cloud overlays a thick cloud, satellite retrievals place the CTHs somewhere between the upper boundaries of the two cloud layers (Hollars et al., 2004; Weisz et al., 2007; Tan et al., 2019). In addition, when retrieving the CTH of the broken clouds by satellite infrared remote sensing, the infrared radiation below the cloud layer can penetrate through the cloud, resulting in the radiation measured by the satellite being compromised by the background radiation below the cloud (Fan et al., 2017). Thus, multilayer clouds, or broken clouds, cause CTH retrieval uncertainty. We also investigated the change in CTHDs with multiple cloud layers and cloud fractions, and the statistical results showed that the CTHD demonstrated slight increases which are deemed insignificant for multilayer clouds and broken clouds at both sites (for brevity, figures not shown in this paper). This result is the same as that reported by Huo et al. (2020b). It can be concluded that the cloud fraction and the presence of multiple cloud layers are not critical factors resulting in CTHD when compared with other causes, such as the cloud height and cloud depth.