Impact of FY-3A MWTS radiances on prediction in GRAPES with comparison of two quality control schemes

Abstract

The impact of Microwave Temperature Sounder (MWTS) radiances on the prediction of the Chinese Numerical Weather prediction (NWP) system-GRAPES (Global and Regional Assimilation and PrEdiction System) with comparison of two Quality Control (QC) schemes was researched. The main differences between the two schemes are cloud detection, O–B (brightness temperature difference between observation and model simulation) check and thinning. To evaluate the impact of the two QC schemes on GRAPES, a typhoon case study and cycle experiments were conducted. In the typhoon case study, two experiments were conducted using both the new and old QC schemes. The results show that outliers are removed in the new QC while they exist in the old QC. The analysis and the model forecast are subsequently generated after assimilating data from the two QC schemes. The model-predicted steering flows more southward with the new QC scheme, and as a result, the forecast track in the experiments is more southward, i.e., closer to the best track than the old scheme. In addition to the case study, four impact cycle experiments were conducted for 25-day periods. The results show that the new QC scheme removed nearly all the biases whereas the old scheme could not. Furthermore, the mean and standard deviation of analysis increments with the new scheme is much smaller than those of O–B. In contrast, the old scheme values are either slightly smaller or the same. Verifications indicate that forecast skill is improved after applying the new scheme. The largest improvements are found in the Southern Hemisphere. According to the results above, MWTS with the new QC scheme can improve the GRAPES forecast.

Keywords

FY-3 / MWTS / typhoon / GRAPES

Introduction

The FY-3A (Fengyun-3A) satellite, a polar-orbiting environmental research satellite, was successfully launched on May 27, 2008. The second in the FY-3 series, FY-3B, was successfully launched on November 5, 2010, carrying the same instruments as those on board FY-3A. The Microwave Temperature Sounder (MWTS) onboard FY-3A provided atmospheric temperature sounding data for the first time in China (Dong et al., 2009; Yang et al., 2009; Zhang et al., 2009). The MWTS is a four-channel radiometer that is similar to Advanced Microwave Sounding Unit-A (AMSU-A) channels 3, 5, 7, and 9 (You et al., 2012). Adjoint-based estimates of observation impacts on Numerical Weather prediction (NWP) (Baker and Daley, 2000) have demonstrated that the largest decrease is due to AMSU-A (Fourrié et al., 2002; Langland and Baker, 2004; Cardinali, 2009). Most NWP centers that made effective use of microwave temperature sounding data have claimed substantial reduction on forecast Root Mean Square (RMS) error (Derber and Wu, 1998; McNally et al., 2000; Bouttier and Kelly, 2001; Ahn et al., 2003; Okamoto et al., 2005).

Zou et al. (2011) have evaluated FY-3A MWTS against NOAA-18 AMSU-A. They found that MWTS data compare favorably with AMSU-A data in terms of global bias. Moreover, the MWTS data standard deviations are slightly larger than those of AMSU-A data. Wang et al. (2012) further demonstrated that the temperature dependence of FY-3A MWTS measurement biases is mainly introduced by a post-launch frequency shift found by Lu et al. (2010). MWTS biases are nearly constant with respect to scene temperature by incorporating the shifted frequencies in radiative transfer models. It is of great value to incorporate the MWTS data from the FY-3 series into NWP models.

MWTS radiance data have been assimilated into the Chinese NWP system Global and Regional Assimilation and PrEdiction System (GRAPES). The main components of GRAPES include (Chen et al., 2008; Xue et al., 2008): 1) three-dimensional variational assimilation (3D-Var), 2) a full compressible nonhydrostatical model core with a semi-implicit and semi-Lagrangian discretization scheme; 3) a modularized model physics package, and 4) global and regional assimilation and prediction systems. However, cloud contaminated data still remain after employing the current Quality Control (QC) scheme. In 2013, a new QC procedure for MWTS measurements was proposed by Li and Zou (2013). A cloud detection algorithm was incorporated based on the cloud fraction product provided by the Visible and InfrarRed Radiometer (VIRR) onboard the FY-3A. With the new method, cloud/rain affected radiances can be identified more efficiently.

In this study, both old and new QC methods were applied in the global version of GRAPES. A typhoon case study and four analysis/forecast cycle experiments for a 25-day period were carried out to show the MWTS observations could have a positive impact on NWP forecasts. A brief introduction to FY-3A MWTS datasets is provided in Section 2. Section 3 describes the two schemes. Section 4 gives the overview of the typhoon case, followed by the description of initialization, experimental setting, and results. Section 5 introduces four impact experiments and their results. A summary and discussion are presented in Section 6.

Description of FY-3A MWTS data

Table 1 shows channel characteristics of FY-3A MWTS. There are only 15 scene Field of Views (FOV) in each MWTS scan line. The horizontal resolution of the nadir FOV is 62 km. MWTS has 4 channels for atmospheric temperature profiling from the Earth’s surface to about 16 km (or 90 hPa) above. MWTS channel 1 is sensitive to surface emissivity and temperature, and atmospheric cloud liquid water. MWTS channels 2–4 provide the atmospheric temperature characteristic in the troposphere and low stratosphere. In both schemes, channels 2–4 radiance data in Level-1B format are employed. Channel 1 is not assimilated because model simulations of brightness temperature are still not accurate due to the uncertainty in the surface emissivity.

Introduction of the two QC schemes

In both QC schemes, FOVs were removed for: a) the two outmost (i.e., 1, 2, 14, and 15), b) channel 2 over sea ice and land, c) channel 3 over terrain at an altitude higher than 500 m, and d) coastal. Furthermore, cloud detection, brightness temperature difference between observation and model simulation (O–B) check, and thinning are different in the two QC schemes. These differences are described as follows.

In the old scheme, a simple O–B method is employed for cloud detection. The FOV is identified as cloudy if the absolute value of O–B for channel 1 (|O–B|_ch1) is greater than 3 K for FOVs over the ocean and 1.5 K over land. The O–B check removes the observations when the absolute value is greater than the product of k and σ_o. The values of k and σ_o for each channel are displayed in Table 2. A thinning distance of 150 km is applied in the old scheme.

In the new scheme, cloud detection is carried out using a cloud fraction product from the VIRR onboard FY-3A. Cloud fraction f_VIRR is calculated for an individual MWTS FOV. It is defined by the ratio of the total number of cloudy pixels to all VIRR pixels located in the MWTS FOV. An MWTS FOV with cloud fraction greater than a threshold of 37% is identified as cloudy scene (Li and Zou, 2013).

In the new scheme, a biweighting quality control procedure is implemented to remove outliers. This method can decrease the influence of outliers on the mean and the Standard Deviation (STD) of a set of data. First, the biweight mean (μ_bm) and biweight STD (σ_bsd) of the following variable are calculated (Lansante, 1996; Zou and Zeng, 2006):

x\equiv {\displaystyle \frac{\mathrm{\Delta }{T}_b}{T_b^{bg}}}={\displaystyle \frac{T_b^{obs}-T_b^{bg}}{T_b^{bg}}}.

Then, the Z-score for all data that pass all checks on cloud, scan edge rejection and surface types is calculated as:

Z_i={\displaystyle \frac{x_i-{\mu }_{bm}}{{\sigma }_{bsd}}},

where the subscript “i” indicates the i^th data. Data with a Z-score of more than two are removed. Considering the variations of the mean states of the atmosphere at different latitudes, the biweighting quality control is implemented in three separate latitudinal bands separately: tropics (30°N–30°S), middle latitudes (30°N–60°N, 30°S–60°S), and high latitudes (60°N–90°N, 60°S–90°S). Research indicates that radiance data that significantly deviate from the background fields can be removed (Zou and Zeng, 2006; Li and Zou, 2013).

In the new scheme, MWTS data are used without thinning for the coarse spatial distribution of MWTS FOVs.

A typhoon case study

Figure 1 shows the best track of typhoon Ma-on (No.1106) with a decrease in intensity from 0600 UTC, 12 July to 0000 UTC, 22 July 22 2011. Typhoon Ma-on (No.1106) was a powerful typhoon affecting Japan in July 2011 (RSMC Tokyo-Typhoon Center, 2012). Ma-on began as a tropical depression (TD) over the sea southeast of Marcus Island at 1200 UTC on 11 July 2011.The following day, it was upgraded to a tropical storm (TS) with an intensity of 0000 UTC and then strengthened to typhoon (TY) intensity northeast of the Mariana Islands at 0000 UTC on 14 July. Ma-on reached its peak intensity northeast of Okinotorishima Island at 1200 UTC on 16 July. It turned northward late on 17 July and made landfall in Shikoku with TY intensity at approximately 1400 UTC on 19 July. It transformed into an extra-tropical cyclone east of Hokkaido at 1200 UTC on 24 July, then dissipated east of Kamchatka seven days later.

The vortex initializing scheme originated from Wang et al. (1996). It includes the removal of an analyzed vortex and specification of a symmetric model vortex. The analyzed vortex is defined as a deviation from the environmental field. It is removed from the large-scale analysis using filtering operators. The Geophysical Fluid Dynamics Laboratory (GFDL) filtering operators are adopted (Kurihara et al., 1993). After filtering, an axially symmetric vortex is produced using the method proposed by Mathur (1991). The bogus field is obtained by implanting the specified vortex in the environmental field. Figure 2 shows the background field both with and without a bogus typhoon at 0000 UTC on 13 July 2011.

Three different data assimilation experiments are carried out (CONV, MWTS1, and MWTS2) using the global version of the GRAPES system. Only conventional observations are assimilated in CONV. MWTS data are assimilated in MWTS1 (MWTS2) with the old (new) QC scheme. The conventional observation contains a global set of surface and upper-air reports, including Radiosondes, SYNOP (surface SYNOPtic observations), Ship, Airep (Aircraft Report), and Atmospheric Motion Vectors from the Global Telecommunications System (GTS).

The model is initialized at 0000 UTC on 13 July 2011. The National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) 6-hour forecast field is used as the background field. It has a horizontal resolution of 1º×1º and 26 vertical levels. The highest vertical level is around 10 hPa. The observation is assimilated at 0000 UTC on 13 July 2011. A 120-hour forecast is conducted.

Figure 3 (Figure 4) shows cloudy and clear FOVs identified in MWTS2 (MWTS1) from 2100 UTC, 12 July to 0300 UTC, 13 July 2011. Cloudy or clear FOVs detected by The Meteorological Operational satellite A (MetOp-A) AMSU-A FOVs are displayed as black or grey dots. AMSU-A cloudy FOVs are detected based on the cloud liquid water path (LWP) estimated by AMSU-A channels 1 and 2 (Weng and Grody, 1994). Red dots indicate cloudy FOVs which have passed the cloud detection but are removed by the O–B check. Blue dots represent the data that passed all QC checks. These figures show that the O–B check can remove some residual cloudy FOVs. Compared with AMSU-A cloudy FOVs, no cloudy FOV remains in MWTS2 (Fig. 3) while several cloudy FOVs still exist in MWTS1 (Fig. 4).

In order to understand how the new QC scheme made an impact on the track prediction, the differences of steering flow between MWTS2 and MWTS1 are examined. The calculation of steering flow adopts the same method used in Wu and Zou (2008). An example of this type of flow separation is given in Fig. 5. The mean flow and the environmental flow are calculated from the initial forecast filed at 0000 UTC on 13 July in MWTS2. The mean flow is derived using the wind speed of 300 hPa, 400 hPa, 500 hPa, 600 hPa, 700 hPa, and 850 hPa. The formula is (Carr and Elsberyy, 1990; Velden and Leslie, 1991):

u_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}={\displaystyle \frac{{75u}_{300}+{100u}_{400}+{150u}_{500}+{175u}_{600}+{175u}_{700}+{150u}_{850}}{825}},

v_{\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{n}}={\displaystyle \frac{{75v}_{300}+{100v}_{400}+{150v}_{500}+{175v}_{600}+{175v}_{700}+{150v}_{850}}{825}}.

It is seen that typhoon Ma-on is characterized by a mean flow containing a seemingly asymmetric cyclone of speed exceeding 30 m·s^–1 (Fig. 5(a)). The environmental flow component in the mean flow (black solid arrow) in MWTS2 is successfully derived by the GFDL method (Fig. 5(b)). The environmental flow shows a northwestward steering effect. Environmental flow from MWTS1 is given in Fig. 5(c) which also points toward the northwest. The environmental and steering flow difference between MWTS2 and MWTS1 are shown in Fig. 5(d). The difference vector of the steering flow between MWTS2 and MWTS1 points to the southeast. The magnitude of this vector is approximately 0.7 m·s^–1.

Figure 6 displays steering flow of MWTS2, MWTS1, and the differences at 1800 UTC on 13 July and 1200 UTC on 15 July, respectively. The 18-hour forecast of the environmental flow of MWTS2 and MWTS1 are shown in Figs. 6(a) and 6(b). The steering flow is indicated as a black solid arrow. Steering flow differences between MWTS2 and MWTS1 are indicated as a blue solid arrow. It points to the southwest with an approximate magnitude of 0.3 m·s^–1. Figs. 6(c) and 6(d) are similar to Figs. 6(a) and 6(b), except for the 60-hour forecast. The steering flow difference also points to the southwest with a magnitude of 0.5 m·s^–1. These differences will cause the track to move southwestward.

Figure 7(a) shows the forecast tracks of CONV, MWTS1, and MWTS2, as well as the best track from the National Meteorological Center. Model-simulated tracks from CONV, MWT1, and MWTS2 experience a similar northward bias. Differences between the three forecast tracks are very small during the first 24 hours and greatly increase thereafter. The MWTS2 track is located south of the MWTS1 and closer to the best track. This difference is consistent with that for the model-predicted steering flow as seen in Fig. 6. Compared with CONV, MWTS data improves the track prediction. The maximum track error is reduced from 250 km in CONV to 202 km in MWTS2. The increase of the minimum Sea Level Pressure (SLP) in the first 60 hours is partly due to the coarse resolution of the GRAPES-3DVar system. The minimum SLP of MWTS2 is less than that of MWTS1 and CONV during the 84 to 120 hour forecast.

Analysis/forecast cycle experiments

Four different experiments are designed and the experimental settings are shown in Table 3. CONV1 (CONV2) assimilates MWTS using the old (new) scheme along with conventional observation. SAT1 (SAT2) are similar to CONV1 (CONV2) but with other satellite data added. These satellite data include radiance from NOAA-15/18 AMSU-A, MetOp-A AMSU-A, and COSMIC RO observations. These parallel experiments are carried out for the period of 7 July to 31 July 2011.

Figure 8 shows data counts of the observations which have passed all QC procedures and have been assimilated. Comparisons between CONV2 and CONV1 suggest that the new QC scheme allows more observation to be assimilated for all channels. The differences between SAT2 and SAT1 also support the results. In addition, more MWTS data are incorporated in SAT1 (SAT2) than in CONV1 (CONV2). The possible reason is that with other satellite data assimilations, the analysis field of SAT1 (SAT2) is better than CONV1 (CONV2). Therefore, the model simulation brightness temperature of the background field will more closely resemble the observation. In this condition, fewer outliers will be removed in the O–B check.

Figures 9–11 display the bias and STD of O–B and O–A (brightness temperature difference between observation and analysis) for channels 2–4 in CONV1 and CONV2. The time ranges from 7 July to 31 July 2011. Figures 12–14 are similar to Figs. 9–11 but for SAT1 and SAT2. These figures show that residual biases of O–B and O–A still exist in channel 4 in CONV1 and SAT1. The STD of O–B and O–A in CONV2 (SAT2) is smaller than that in CONV1 (SAT1) for all channels. This reveals that the adoption of the new scheme can improve the assimilation. The STD of O–A is much smaller than that of O–B in CONV2 and SAT2. However, it is almost the same, or only somewhat smaller, than that of O–B in CONV1 and SAT1. For example, O–A of channel 4 is comparable with O–B (Fig. 14). The contribution of MWTS to analysis is much smaller in the experiments with the old scheme.

Forecasts are verified against their own analyses. An overall measurement of the quality of medium-range forecasts in predicting a large scale weather system is given by the anomaly correlation coefficient (ACC) of the 500 hPa height forecast field. A key performance indicator for the forecast system is the forecast range at which the ACC drops to 60%. The results are shown in Fig. 15 in terms of mean 500 hPa anomaly correlations and root mean square for both the Northern (Figs. 15(a) and 15(b)) and Southern (Figs. 15(c) and 15(d)) Hemispheres. There is a small positive impact on the Northern Hemisphere and a more evident improvement in the Southern Hemisphere later in the forecast for SAT2 over the SAT1. In addition, Fig. 15 shows that when using the new QC scheme in CONV2, the anomaly correlation coefficient (ACC) of the 500 hPa height forecast field is also increased, although the improvement is not as significant as in SAT2. When all satellite data are added in SAT2, the 6-hr forecast field in SAT2 will be much better than in CONV2. Therefore, the quality control (such as O–B check), which is based on the background field, will retain more “good” observations in SAT2. For this reason, the MWTS radiance data have a more positive impact on the analysis and forecast in SAT2 than in CONV2. The verification also indicates improved forecasts when satellite data are used. Improved ACC and reduced RMS are found in SAT2 (SAT1) in comparison with CONV2 (CONV1).

Summary and discussion

In this research, two MWTS QC schemes were applied in the GRAPES system. A case study and four cycle experiments were conducted to show the new scheme could positively impact the assimilation and model forecast. The typhoon case study indicates that cloudy FOVs are removed in MWTS2, while some cloudy data still remain in MWTS1. In terms of the forecast track, MWTS1 is closer to the control experiment. The difference between MWTS2 and MWTS1 is consistent with the model-predicted steering flow difference. MWTS data produced an improvement in the track prediction.

Analysis/forecast cycle experiments were conducted for nearly four weeks. The assimilation with the new scheme was found to produce a smaller bias and STD of O–B and O–A. However, residual biases of O–B and O–A still exist in experiments using the old scheme. The STD of O–A is much smaller than that of O–B in CONV2 and SAT2, while it is almost the same, or only somewhat smaller, than that of O–B in the old scheme indicating that the contribution of MWTS to analysis is small in the old scheme. Verifications indicate that ACC of the 500 hPa height forecast field are increased and RMS are reduced in the new scheme. The greatest improvement is found in the Southern Hemisphere. ACC drops to 60% around forecast day 6 in the Southern Hemisphere. The 60% level is improved for approximately 4–5 hours with the new scheme. The impacts in the Northern Hemisphere are smaller but still important, with RMS errors lower with the new scheme. In all, the new scheme can improve model forecast when compared with the old scheme. The impact is largest in the Southern Hemisphere.

The study demonstrates the usefulness of the new QC scheme to NWP. Since the Chinese FY-3B satellite was successfully launched into an afternoon-orbit on 4 November 2010, the potential value of MWTS onboard both FY-3A and FY-3B will be further explored.

References

[1] AhnM H, KimM J, ChungC Y, SuhA S (2003). Operational implementation of the ATOVS processing procedure in KMA and its validation. Adv Atmos Sci, 20(3): 398–414

[2] BakerN L, DaleyR (2000). Observation and background adjoint sensitivity in the adaptive observation-targeting problem. Q J R Meteorol Soc, 126(565): 1431–1454

[3] BouttierF, KellyG (2001). Observing-system experiments in the ECMWF 4D-Var data assimilation system. Q J R Meteorol Soc, 127(574): 1469–1488

[4] CardinaliC (2009). Monitoring observation impact on short-range forecast. Q J R Meteorol Soc, 135(638): 239–250

[5] CarrL E, ElsberryR L (1990). Observational evidence for predictions of tropical cyclone propagation relative to environmental steering. J Atmos Sci, 47(4): 542–548

[6] ChenD H, XueJ S, YangX S, ZhangH L, ShenX S, HuJ L, WangY, JiL R, ChenJ B (2008). New generation of multi-scale NWP system (GRAPES): general scientific design. Chin Sci Bull, 53(22): 3433–3445

[7] DerberJ C, WuW S (1998). The use of TOVS cloud-cleared radiances in the NCEP SSI analysis system. Mon Weather Rev, 126(8): 2287–2299

[8] DongC H, YangJ, YangZ D, LuN M, ShiJ M, ZhangP, LiuY J, CaiB, ZhangW (2009). An overview of a new Chinese weather satellite FY-3A. Bull Am Meteorol Soc, 90(10): 1531–1544

[9] FourriéN, DoerenbecherA, BergotT, JolyA (2002). Adjoint sensitivity of the forecast to TOVS observations. Q J R Meteorol Soc, 128(586): 2759–2777

[10] KuriharaY, BenderM A, RossR J (1993). An initialization scheme of typhoon Models by vortex specification. Mon Weather Rev, 121(7): 2030–2045

[11] LanglandR H, BakerA L (2004). Estimation of observation impact using the NRL atmospheric variational data assimilation adjoint system. Tellus, Ser A, Dyn Meterol Oceanogr, 56(3): 189–201

[12] LansanteJ R (1996). Resistant, robust and non-parametric techniques for the analysis of climate data: theory and examples, including applications to historical radiosonde station data. Int J Climatol, 16(11): 1197–1226

[13] LiJ, ZouX (2013). A Quality Control Procedure for FY-3A MWTS Measurements with Emphasis on Cloud Detection Using VIRR Cloud Fraction. J Atmos Ocean Technol, 30: 1704–1715

[14] LuQ, BellW, BauerP, BormannN, PeubeyC (2010). An Initial Evaluation of FY-3A Satellite Data. ECMWF Technical Memoranda No.631, ECMWF, Shinfield Park, Reading, UK, ECMWF, 58

[15] MathurM B (1991). The National Meteorological Center’s Quasi-Lagrangian Model for hurricane prediction. Mon Weather Rev, 119(6): 1419–1447

[16] McNallyA P, DerberJ C, WuW, KatzB B (2000). The use of TOVS level-1b radiances in the NCEP SSI analysis system. Q J R Meteorol Soc, 126(563): 689–724

[17] OkamotoK, KazumoriM, OwadaH (2005). The assimilation of ATOVS radiances in the JMA global analysis system. J Meteorol Soc Jpn, 83(2): 201–217

[18] RSMC Tokyo-Typhoon Center (2012). Summary of the 2011 typhoon season. Forty forth session, Hangzhou, China, ESCAP/WMO Typhoon Committee

[19] VeldenC S, LeslieL M (1991). The basic relationship between tropical cyclone intensity and the depth of the environmental steering layer in the Australian region. Wea Forecasting, 6: 244–253

[20] WangG M, WangS W, LiuJ J (1996). A bogus typhoon scheme and its application to a movable nested mesh model. Journal of Tropical Meteorology, 12(1): 9–17 (in Chinese)

[21] WangX, ZouX L, Weng F Z, You R (2012). Using NWP models to remove frequency shift induced bias in FY-3A MWTS measurements. IEEE Trans Geosci Rem Sens, 50(12): 4860–4874

[22] WengF, GrodyN C (1994). Retrieval of cloud liquid water using special sensor microwave imager (SSM/I). J Geophys Res, 99(D12): 25535–25551

[23] WuY H, ZouX (2008). Test of a simple approach for using TOMS total ozone data in hurricane environment. Q J R Meteorol Soc, 134(635): 1397–1408

[24] XueJ S, ZhuangS Y, ZhuG F, ZhangH, LiuZ Q, LiuY, ZhuangZ R (2008). Scientific design and preliminary results of three-dimensional variational data assimilation system of GRAPES. Chin Sci Bull, 53(22): 3446–3457

[25] YangJ, DongC H, LuN M, YangZ D, ShiJ M, ZhangP, LiuY J, CaiB (2009). FY-3A: the new generation polar-orbiting meteorological satellite of China. Acta Meteorologica Sinica, 67(4): 501–509 (in Chinese)

[26] YouR, GuS Y, GuoY, WuX B, YangH, ChenW X (2012). Long-term calibration and accuracy assessment of the FengYun-3 Microwave Temperature Sounder radiance measurements. IEEE Trans Geosci Rem Sens, 50(12): 4854–4859

[27] ZhangP, YangJ, DongC H, LuN M, YangZ D, ShiJ M (2009). General introduction on payloads, ground segment and data application of Fengyun 3A. Frontiers of Earth Science, 3(3): 367–373

[28] ZouX, WangX, WengF, LiG (2011). Assessments of Chinese FengYun Microwave Temperature Sounder (MWTS) measurements for weather and climate applications. J Atmos Ocean Technol, 28(10): 1206–1227

[29] ZouX, ZengZ (2006). A quality control procedure for GPS radio occultation data. J Geophys Res, 111(D2): D02112

RIGHTS & PERMISSIONS

Higher Education Press and Springer-Verlag Berlin Heidelberg