Evaluation of the Long-term Performance of Microwave Radiation Imager Onboard Chinese Fengyun Satellites

The FY-3 series of Chinese polar-orbiting meteorological satellites has been launched in succession from 2008–21 (Dong et al., 2009; Yang et al., 2009; Lu and Gu, 2016; Zhang et al., 2019b, 2022), including FY-3A (May 2008), FY-3B (November 2010), FY-3C (September 2013), FY-3D (November 2017), and FY-3E (July 2021). The Microwave Radiation Imager (MWRI) is one of the primary sensors on the FY-3 satellites. The MWRI is a conical-scanning microwave imager with frequencies varying from 10.65 GHz to 89 GHz, which are sensitive to the surface, column water vapor, cloud, and precipitation (Tang and Zou, 2017; Li et al., 2022). MWRI observations have been used to study tropical cyclones and soil moisture (Zhang et al., 2015; Cui et al., 2016; Xian et al., 2021).

To guarantee and check high-quality MWRI measurements, many calibrations and validations have been performed since 2008 (Gu et al., 2021). Calibration and correction procedures for the MWRIs on FY-3A/B have been performed, and the linearity and non-linearity corrections were carefully considered (Yang et al., 2011a). Liu et al. (2014) introduced an antenna surface calibration system for MWRI/FY3C. Given the fact that nonlinear variation was an essential factor impacting the calibration accuracy, the nonlinear characteristics of MWRI/FY3D were investigated and an optimal calculation method was proposed (Chen et al., 2017; Chen et al., 2017; Dong et al., 2020). Short-term MWRI/FY3B BTs in 2010 were compared with observations from the Aqua Advanced Microwave Scanning Radiometer (AMSR-E) and radiative transfer model (RTM) simulations, showing a small bias of MWRI (Yang et al., 2013). MWRI/FY3B BTs in the early six months after launching were carefully checked. On-orbit MWRI BTs were stable, with a maximum fluctuation of the coldest reference values at all channels being no more than 1.8 K (Qiao et al., 2012). The geolocation errors of the MWRIs on FY3B/3C were analyzed and corrected (Chen et al., 2016; Tang et al., 2016; Song et al., 2017; Liu et al., 2021). The quality of MWRI/FY3C was evaluated by comparison with RTM results, showing a 1–2-K difference between ascending and descending orbits (Lawrence et al., 2017; Zeng and Jiang, 2021). Zhang et al. (2019a) compared the parameters of the calibration equation for MWRI/FY3C and found the high value of the hot load reflector was the main cause of the bias, which was corrected, and thereby the bias between ascending and descending orbits was reduced.

These previous studies were primarily based on short-term observations after launching an MWRI. Since there are long-term records from the MWRIs on the FY-3 serial satellites, it is highly critical to assess the long-term quality of MWRI data, which is a top priority for the retrievals of atmospheric and surface properties. Given the fact that the MWRI channels are nearly identical to those of the Advanced Microwave Scanning Radiometer-2 (AMSR2) instrument on the GCOM-W1 satellite (Takashi et al., 2016), the inter-comparison between MWRI and AMSR2 observations from 2012 to 2019 was investigated, which is, as far as we know, the first attempt to evaluate the long-term performance of MWRIs by using advanced AMSR2 measurements. Furthermore, we also used RTM simulations to investigate the causes for the bias between MWRI and AMSR2 measurements, which will shed new light on how to improve MWRI measurements in the near future.

Both the MWRI and AMSR2 are conical-scanning microwave imagers (Table 1), with nearly identical central frequencies from 10.65 GHz to 89 GHz. Due to very short-term MWRI observations from FY-3A and the larger Equator Crossing Time (ETC) gap between FY3C and GCOM-W1, the inter-comparison here was mainly focused on the FY3B and FY3D satellites.

The real-time, operational MWRI L1B radiance data from FY-3B/3D are available online (http://satellite.cma.gov.cn/PortalSite/Data/Satellite.aspx), including the calibrated BT and land cover at scanning pixels. The land cover is from the international geosphere biosphere program (IGBP) classification system. The operational L1B radiance data from the MWRIs were directly compared against the Global Precipitation Measurement (GPM) L1C AMSR2 Product. The latter is transformed from the equivalent AMSR2/GCOM-W1 L1B radiance data using the GPM Microwave Imager (GMI) as the reference standard. As the successor of AMSR-E/Aqua (launched in May 2002), AMSR2 includes an improved on-board calibration target, resulting in the reduction of annual BT variation and the improvement of BT stability; for instance, AMSR2 L1B brightness temperature has a precision or random error +/–0.3 K and an accuracy within +/–1.5 K (Ebuchi et al., 2021). After the GPM core satellite was launched in February 2014, the GMI was used as a calibration standard to inter-calibrate different imagers aboard other polar-orbiting satellites (Hou et al., 2014; Skofronick-Jackson et al., 2017). The GPM Intersatellite Calibration Working Group found that AMSR2 L1B exhibits a substantial cold-scene warm calibration bias with respect to the GMI (Berg et al., 2016) and established a calibration adjustment to generate GPM L1C AMSR2 data. Considering that GMI BT has high accuracy levels for all channels within 0.4 K and stability within 0.2 K (Wentz and Draper, 2016), the data quality from GPM L1C AMSR2 should be better than the original AMSR2/GCOM-W1 L1B radiance. Hence, GPM L1C AMSR2 radiance data is the optimal choice for comparing with MWRI data, which makes it possible for MWRI data to be tractable to the international reference.

MWRI and AMSR2 BT data from 1–5 July 2012–19 between 70°–135°E and 20°–50°N were compared. Hence, there are eight years of matching data for MWRI/FY3B (2012–19) and two years of matching data for MWRI/FY3D (2018–19). The BT observations were matched if the temporal/spatial difference between MWRI and AMSR2 observations is less than 15 minute/5 km. To mitigate uncertainties due to inhomogeneous atmosphere or land surface, an extra constraint was used in the matching process, i.e., the standard deviations of BT should be no more than 2 K in the area around the matched samples.

The BT differences at the eight channels from 10.65 GHz to 37.0 GHz (in both vertical and horizontal polarization channels, denoted as 10V, 10H, 19V, 19H, 23V, 23H, 37V, and 37H, respectively) were analyzed in this work. To see the BT difference clearly, the matching pairs were divided into two subgroups, i.e., over water and land, based on their land cover information. First, BT from both instruments at the eight channels over land (Fig.1) shows a good linear correlation (R varies from 0.94 to 1.0), and there is especially strong agreement between MWRI with AMSR2 at 19V onboard either FY3B or FY3D. A small standard deviation (STD) of BT difference (varying from 0.8 K to 1.3 K) occurs at the vertical polarization channels of MWRI on both FY3B and FY3D. Second, the agreement of AMSR2 and MWRI/FY3D appears better than that of AMSR2 and MWRI/FY3B, which is reflected in the obvious reduction of the mean bias error (MBE) from 2–4 K for MWRI/FY3B to 0–2 K for MWRI/FY3D. The STD for MWRI/FY3D is also smaller than that for MWRI/FY3B, particularly at the horizontal polarization channels.

The matched pairs over water are shown in Fig. 2. It can be seen that most MWRI BTs are smaller than AMSR2 measurements. In particular, the MBE at the vertically polarized channels exceeds 5 K and even reaches up to 10 K at 10V and 37V. This is almost three times larger than that at the corresponding horizontally polarized channels. In addition, the MBE at 23H (~8 K) is larger than that at the other three horizontal channels (2–3 K).

Figure 2.  Same as Fig. 1 except for the water group.

From the BT differences shown in Figs. 1 and 2, there are two interesting points that merit further discussion. First, the majority of MWRI BTs are smaller than AMSR2 measurements, no matter whether the MWRI is onboard FY3B or FY3D. Second, we can also see a clear improvement in the performance of the MWRI onboard FY3D as compared with that on FY3B. This is supported by the fact that the BT differences between MWRI/FY3D and AMSR2 over land are mostly less than 2 K; however, the differences are up to 2–4 K for MWRI/FY3B. The improvements observed from the MWRI onboard FY3D are much more outstanding at the horizontal polarization channels than that at the vertical channels. This implies that the relatively high-quality MWRI/FY3D observations would be beneficial for the correction of MWRI/FY3B BT measurements, thereby maintaining consistency in the long-term MWRI data.

Figure 1.  Scatter plots of matched BT at eight channels for AMSR2–MWRI/FY3B during 2012–19 (blue) and AMSR2–MWRI/FY3D during 2018–19 (red) over land. The corresponding statistics parameters are shown in the same color.

To evaluate the long-term performance of MWRI/FY3B BT data, the annual BT differences between AMSR2 and MWRI/FY3B for the years 2012–19 are shown in Fig. 3. The box plots of BT difference over land (filled) and water (unfilled) groups are clearly different at the vertical polarization channels, but they are similar to each other at the horizontal polarization channels. Over land, the annual median BT differences (1–5 July each year, hereinafter) at the horizontal and vertical polarization channels increase in absolute magnitude over time. Over water, it is still clear that the BT differences at the vertical polarization channels are much larger than those at the corresponding horizontal channels. In particular, at 10V and 37V, the median BT difference exceeds 10 K. The annual BT differences over water for channels of both polarizations increase slightly during 2012–19. The gradual increase in absolute magnitude of the annual BT difference over those eight years indicates a gradually increasing measurement uncertainty for the operationally calibrated MWRI on FY3B.

Figure 3.  Box plots of annual BT differences between MWRI/FY3B and AMSR2 during 2012–19 over land (filled) and over water (unfilled).

For AMSR2–MWRI/FY3D matching pairs in 2018–19, the variations of BT difference in each year are small (not shown), which implies stability of the MWRI on FY3D over those two years.

An obvious difference between the ascending and descending orbits for MWRI/FY3C was found by Lawrence et al. (2017), and the difference was found to be likely due to different errors in the hot load. This raises the question of whether this difference exists for the MWRIs on FY3B/3D. To investigate this question, we grouped the matching MWRI–AMSR2 samples from 2012–19 for FY3B and those from 2018–19 for FY3D based on MWRI ascending (A) and descending (D) orbits. Figure 4 shows the MBE of MWRI–AMSR2 BT differences for both A and D orbits over land and water. The BT difference for the descending orbit of MWRI/FY3B is generally smaller than that for the ascending orbit, independent of being over land or water. This phenomenon did not occur for MWRI/FY3D, reflecting the improvement of MWRI/FY3D in reducing the difference between ascending and descending orbits.

Figure 4.  Box plots of BT differences between MWRI and AMSR2 for the ascending and descending orbits of MWRI/FY3B (2012–19) and MWRI/FY3D (2018–19) over land (filled) and over water (unfilled).

In the above comparisons, substantial BT differences between MWRI and AMSR2 were found over water. The potential causes for these differences are explored here by using the simultaneous conical overpassing (SCO) method and RTM simulations.

The SCO method has been widely used to calibrate conically scanning instruments, such ss SSMIS and SSMI, and to remove biases in order to achieve consistency and traceability (Yan and Weng, 2008; Yang et al., 2011b). The SCO method is adopted from the Simultaneous Nadir Overpasses (SNO) technique developed by Cao et al. (2004). The SNO method has been used for the construction of long-term records of the Microwave Sounding Unit (MSU) and the Advanced Microwave Sounding Unit-A (AMSU-A) (Zou et al., 2006; Iacovazzi and Cao, 2008).

Considering that the MWRI and AMSR2 are conically scanning and have a good chance of matching due to their close ECT, the SCO method was used to explore the BT differences between AMSR2 and MWRI/FY3D during 1–3 July 2018. To obtain more matching in the high latitudes where the surface and atmosphere are relatively stable, the SCO constraint on the distance window from the two sensors is set to a ground distance of 5 km, and the time window is set to 60 s, although SNO (or SCO) events occur within a few seconds for succeeding satellites.

The SCO pairs of MWRI and AMSR2 over water mostly occurred at latitudes around ±75° on 1 July 2018. BTs from both instruments for SCO pairs are compared in Fig. 5, where colors are used to distinguish samples located in the north-polar (blue) and south-polar (red) regions. It can also be seen in Fig. 5 that the BT differences between MWRI and AMSR2 at 10V and 37V are more significant, and the MBEs at the eight channels are close to those obtained in the previous section. However, the matching number (~3200) is far less than that from the previous section, which implies that these BT differences are relatively stable and independent of the selected region.

Figure 5.  Scatter plots of the SCO pairs between MWRI/FY3D and AMSR2 at eight channels over water. The colors represent the pairs located in the north-polar (blue) and south-polar (red) regions.

RTM is a bridge that connects observations and theoretical radiation, and it has been widely used for satellite observation cross-validations (Goldberg et al., 2001; Clough et al., 2005; Lu et al., 2011). Hence, we combined the atmospheric profiles derived from ERA5 reanalysis data (1-h temporal resolution, 0.25° spatial resolution) with a fast microwave radiative transfer model, MWRT (Liu, 1998), to obtain the BT simulations for AMSR2 at 55° and MWRI at 53.1° incident angles, respectively. Firstly, BT observations over land from both instruments were compared with the corresponding model simulations, showing a slight change (<0.1 K) in BT simulations at different incident angles. Then, we focused on the substantial BT differences of MWRI–AMSR2 pairs at the vertical polarization channels over water.

Figure 6 shows the scatter plots comparing both BT simulations and the corresponding MWRI–AMSR2 observations at the vertically polarized channels over water. To clearly see the influence of incident angle on the simulations, BT simulations for AMSR2 (blue) and MWRI (red) are plotted in a panel for the same channel. Noted that we only selected ERA5 data at 0500 UTC 1 July 2018 to do the simulation testing, so the available matched samples are limited (~790). First, it can be seen in Figs. 6a–d that the simulated AMSR2 BTs (blue) at the four vertical channels are more consistent with the corresponding AMSR2 observations (with MBE less than 0.5 K and STD less than 1.3 K), which demonstrates that the AMSR2 BT simulations are reasonable and reliable. In the comparisons of AMSR2 observations with MWRI BT simulations (red), AMSR2 observations are higher by 2–4 K due to the 2° incident angle difference. Second, MWRI observations and both simulations are compared at four vertical polarization channels (Figs. 6e–h), showing that MWRI observations are generally lower than the BT simulations (red). Specifically, the observations are about 6 K lower at 10V and 37V. The corresponding BT differences between MWRI observations and AMSR2 simulations (blue) shown in Figs. 6e–h are close to the results derived from both BT observations.

Figure 6.  BT differences between MWRI–AMSR2 BT observations and the corresponding BT simulations (MWRT for short) for AMSR2 (blue, 55°) and MWRI (red, 53.1°) at the vertically polarized channels. The corresponding statistics parameters are shown in the corresponding color.

Therefore, using model simulations as a reference, we can see that the significant BT difference at the vertically polarized channels over water is mainly due to the different incident angles and the underestimated MWRI BT observations. To better understand the effect of both, the contribution of each to the total BT difference was calculated (Fig. 7). For the four vertically polarized channels, the underestimated MWRI observation accounts for 60% of the total difference at 10V and 37V, and the incident angle accounts for 40%; the proportions are reversed at 19V. For 23V, the contributions are equal at 50%.

Figure 7.  Ratio of the contribution of incident angle and underestimated MWRI to the MWRI–AMSR2 total BT difference.

The MWRI BT referenced in the previous sections is the operationally calibrated data. We further assessed the quality of the recalibrated MWRI BT [doi: 10.12185/NSMC.RICHCEOS.FCDR.MWRIRecalOrb.FY3.MWRI.L1.GBAL.POAD. NUL. 010KM. HDF.2021.2.V1, Wu et al. (2010)] to demonstrate the necessity and validity of recalibrations. The recalibrated MWRI BT is developed by the Laboratory of Radiometric Calibration and Validation for Environmental Satellites, National Satellite Meteorological Center, China Meteorological Administration. This dataset is based on the original level-0 MWRI data from the FY-3 series satellites and adopts the new and improved calibration algorithm based on the original MWRI operational calibration algorithm. The main algorithm improvements include a radiation correction of the microwave imager's hot mirror back lobe, a hot mirror emissivity correction, a heat source efficiency correction, and a receiver nonlinear correction.

Here, we chose the same period (2012 to 2019) of recalibrated MWRI BT from FY3B and FY3D and used the same treatments as before to match with AMSR2 observations and model simulations. We still focus on matching pairs over water due to the obvious BT difference as mentioned above. The differences are shown in Fig. 8 where the matched AMSR2 and two types of MWRI BTs [the operational calibration (blue) and the recalibration (red)] are plotted for each of the eight channels. It can be seen that the MBE of AMSR2–MWRI for the recalibrated version is obviously reduced at the vertically polarized channels. Specifically, at 10V and 37V, the difference is significantly reduced by more than 50%. Relatively, the changes at 18.7 GHz and 23.8 GHz are small with a considerable overlap between the operational and recalibrated data.

Figure 8.  Scatter plots of matched BT at eight channels for AMSR2–MWRI/FY3D in 2018–19 over water. The colors represent the operational (blue) and recalibrated (red) MWRI BT. The corresponding statistics parameters are shown in the corresponding colors.

To better understand the long-term performance of the recalibrated MWRI BT data, the annual BT difference between AMSR2 and recalibrated MWRI/FY3B during 2012–19 is shown in Fig. 9. Compared with the operationally calibrated MWRI results (Fig. 3), it can be clearly seen that over land the annual median BT differences at most of the channels are close to zero and have changed very little in eight years. Over water, the annual median BT differences at the vertical polarization channels are significantly reduced to 5 K and remain more constant over the eight-year period. These reduced and more stable BT differences over both land and water demonstrate the improvement of the recalibrated MWRI BTs, except for 23H, where the BT differences on land and water are still large.

Figure 9.  Same as Fig. 3 except for the recalibrated MWRI BT.

Using the model results from section 4.2 as a reference, the relationships between AMSR2 and the recalibrated MWRI BT observations are presented in Fig. 10. Firstly, the AMSR2 results shown in Figs. 10a–d are quite close to those shown in Figs. 6a–d, that is, AMSR2 observations are higher than MWRI BTs by more than 2–4 K due to the 2° incident angle difference. However, the recalibrated MWRI BT results seen in Figs. 10e–h show significant improvements. For example, the MWRI observations are closer to their BT simulations (red), especially at 10V and 37V where the MBE is reduced from 5–6 K (Figs. 6e–h) to 1.5 K, suggesting that the recalibrated MWRI BT has removed the underestimation at 10V and 37V seen in the operational MWRI BT over water. Relatively, the updated effect on 18V and 23V is weak at about 0.5 K. Additionally, the BT differences between MWRI observations and AMSR2 simulations (blue) shown in Figs. 10e–h also reflect the reduced BT difference compared to the corresponding results shown in Figs. 6e–h, especially at 10V and 37V since the obvious underestimation in the operational data has been corrected and only the BT difference caused by the 2° incident angle difference remains.

Figure 10.  Same as Fig .6 except for the recalibrated MWRI BT.

Several MWRIs on FY-3 serial satellites have been launched, providing rich data for monitoring cloud, precipitation, and ground conditions. To better understand the long-term performance of MWRIs, we carefully compared the operational MWRI BT from FY-3B/3D and AMSR2 observations from GCOM-W1 for the period 2012–19. The comparisons were made over land and water. The SCO method and RTM simulations were also used in the comparison, and a preliminary assessment of recalibrated MWRI data was made. The major conclusions are as follows:

1) On both the FY-3B and FY-3D satellites, the majority of MWRI BTs were smaller than AMSR2 measurements, especially at the vertical polarization channels over water where most MBEs exceeded 5 K and even reached 10 K at 10V and 37V.

2) Over land, the BT differences were reduced from 2–4 K from FY3B to about 0–2 K from FY3D, and the ascending and descending orbit difference for MWRI/FY3D is much smaller than that for MWRI/FY3B, reflecting the improvement of the MWRI onboard FY3D.

3) The operational MWRI BTs are significantly correlated the AMSR2 observations, and the BT differences between MWRI/FY3B and AMSR2 during 2012–19 gradually increased over time. However, the differences remained more stable between MWRI/FY3D and AMSR2 over the period 2018–19.

4) Similar BT differences were found over polar water by using the SCO method, which implies that these BT differences were more stable and independent of the selected region.

5) Substantial BT differences at the vertical polarization channels for MWRI–AMSR2 over water were partly due to the different incident angles; however, the underestimation of the operational MWRI BT over water remained a very important issue.

6) The recalibrated MWRI BT significantly improves upon the operational MWRI BT observations at the vertical polarization channels, especially the effective removal of the underestimation at 10V and 37V over water.

It is important to note that the current conclusions are obtained based on limited observations (e.g., randomly selected dates, the first five days in July) for the period 2012–19 over China and the polar region. We also need to examine the seasonal and spatial variations of BT differences, explore their causes using more MWRI and AMSR2 observations and other references, and consider the effect of cloud presence on observed pixels to better understand MWRI observation quality from the FY-3 series satellites.

Acknowledgements. This work is supported by the National Key R&D Program of China (Grant No. 2022YFF0801301) and the National Natural Science Foundation of China (Grant No. 41575033). Thanks to the China Meteorological Administration-National Satellite Weather Center for MWRI data support (2022YFF0801301). Thanks to the NASA GES DISC for the L1C AMSR2 data support. Thanks to ECMWF for ERA5 (hourly and 0.25°) data support. Thanks to Prof. Guosheng LIU working at Florida State University for MWRT model support.