Effect of Assimilating Himawari-8 Atmospheric Motion Vectors on Forecast Errors over East Asia
1. Introduction
The future state of the atmosphere can be predicted by integrating the numerical weather prediction (NWP) model forward in time, from a chosen initial state. During this process, the uncertainty associated with the initial condition is one of the factors leading to the forecast error (Rabier et al. 1996; Kim et al. 2004). Many studies have been performed to quantify the effect of initial conditions on the improvement of the forecast accuracy (Langland et al. 1999; Baker and Daley 2000; Fourrie et al. 2002; Kim and Jung 2006; Kim et al. 2008; Kim and Jung 2009a,b; Kim et al. 2013; Martinet et al. 2014). The analysis state, considered closest to the true state, is used as the initial condition for the NWP model. During data assimilation (DA), the analysis state is produced by minimizing the sum of each error associated with the background state of the atmosphere and observations, weighted by the presumed error of the model background and the observation, respectively. The optimal observation system can contribute to the production of more exact initial conditions and the improvement of the forecast accuracy. Therefore, it is necessary to quantitatively analyze the impact of each observation to design the observation system appropriately.
Observation system experiments (OSEs) and the forecast sensitivity to observations (FSOs) have been used to diagnose the observation impact on forecasts (Kim et al. 2017). The OSE is a method used to perform DA with observation systems and analyze the effect of selected components of observation systems on forecasts (Langland et al. 1999; Cardinali et al. 2003; Buizza et al. 2007; Kelly et al. 2007; Jung et al. 2010, 2012). The observation system can be modified by excluding particular components of the system or introducing new, novel components to the system. In OSEs, the impact of observations can be measured by differences between the control forecast based on an observation system and forecasts based on modified observation systems. Because the observation systems for OSEs differ from the control experiment (or from one another) (Gelaro and Zhu 2009), the correlations between the observations change for every experiment. In addition, the computational costs of OSEs are high because independent experiments have to be performed for each modified observation system. However, in contrast to the adjoint-based observation-impact technique, the observation impact can be analyzed in OSEs for long-term forecasts representing a nonlinearly developing atmospheric state because the adjoint model is not used. Using the adjoint of the NWP model and DA system, the FSO independently analyzes the impact of each observation on the forecast error reduction (Baker and Daley 2000; Kim and Kim 2014, 2017). The observation impact can be estimated by calculating the adjoint integration once; hence, the computational costs decrease and each observation impact is calculated in an identical system (Tremolet 2007; Gelaro and Zhu 2009). However, the FSO is mostly valid from 1 to 3 days because of the tangent linear assumptions, which makes it inappropriate to apply for long-term forecasts (Cardinali 2009).
Atmospheric motion vectors (AMVs) are observation data included in the observation system of the operational NWP centers (Velden et al. 2005; Wu et al. 2015). The AMVs are produced by tracking the movement of clouds and water vapor in successive satellite images (Schmetz et al. 1993; Velden et al. 1997). In particular, the AMVs produced by using geostationary satellite images (geoAMVs), which have relatively high temporal resolution, are useful for mesoscale atmospheric flow identification (Wang et al. 2006; Wu et al. 2014). The AMVs can sample the ocean region, which is typically less well sampled than the land region, and provide wind information for times when it is difficult to obtain aircraft observations as a result of severe weather conditions, such as tropical cyclones (Langland et al. 1999). Many studies have been carried out using both techniques across several NWP centers that have evaluated AMV data (Tomassini et al. 1999; Wang et al. 2006; Zapotocny et al. 2008; Kieu et al. 2012; Le Marshall et al. 2013; Gelaro et al. 2010; Jung et al. 2013; Kim et al. 2017). These studies confirm the importance of the data in a variety of situations, particularly around areas lacking conventional wind data and in the forecasting of tropical cyclones.
Several operational NWP centers have started or are planning to use the weather products from the next-generation geostationary orbit satellite fleet {e.g., Himawari-8, Geostationary Korea Multipurpose Satellite-2A [GEO-KOMPSAT-2A (GK-2A)], GOES-16, Fengyun-4 (FY-4), and Meteosat Third Generation (MTG) series}; the quality of the products from geostationary satellites is expected to improve. The Himawari-8 (HIMA-8) was launched first by the Japan Meteorological Agency (JMA) among the next-generation geostationary orbit satellites. The HIMA-8 data have replaced Multifunctional Transport Satellite-2 (MTSAT-2) data since 0200 UTC 7 July 2015, and have been used in operational NWP centers worldwide. The new imager sensor, the Advanced Himawari Imager (AHI), is installed on HIMA-8 (Hidehiko et al. 2015). Compared with 5 [1 visible (VIS); 1 water vapor (WV); 3 infrared (IR)] observation channels of the MTSAT-2 imager sensor, the number of observation channels of AHI increased to 16 (3 VIS; 3 WV; 10 IR). The AMVs in HIMA-8 are derived from 5 (1 VIS; 3 WV; 1 IR) observation channels, which has two more WV channels compared with 3 (1 VIS; 1 WV; 1 IR) observation channels in MTSAT-2. The spatial resolution of HIMA-8 was improved to 0.5–1 km for VIS and to 1–2 km for IR from 1 km for VIS and 4 km for IR in the MTSAT-2; the temporal resolution observing the full disk is 10 min, compared to approximately 30 min for the MTSAT-2 and GOES-15. Because of the spatially and temporally higher resolution of the data of HIMA-8 and the advanced derivation algorithm for AMVs in HIMA-8 compared with MTSAT-2, the number of HIMA-8 AMVs is greater than that of MTSAT-2 AMVs (Lean et al. 2016). Yamashita (2016) evaluated the effect of replacing MTSAT-2 AMVs with HIMA-8 AMVs for 2015 summer forecasts with the JMA Global Spectral Model (GSM) four-dimensional variational (4DVAR) DA system and showed that the forecast error of the 500-hPa geopotential height decreases during the whole experimental period but that the assimilation of HIMA-8 AMVs does not affect the track forecast error of Typhoon Noul (1506) until the 18-h forecast and that of Typhoon Dolphin (1507) for all forecast times. Ma et al. (2017) analyzed the impact of the assimilation of HIMA-8 AMVs and clear-sky radiance using the Global Data Assimilation System (GDAS) and the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP) and showed a neutral to slight positive impact on analysis and forecast accuracy. To maximize the effect of assimilating HIMA-8 AMVs on the regional forecasts, additional studies on the observation impact of AMVs are needed using various NWP models and high-impact weather cases. In addition, the effect of assimilating AMVs of several satellites simultaneously produced by Himawari and the next-generation geostationary orbit satellites over East Asia (e.g., GK-2A in Korea and the FY series in China) needs to be diagnosed for better use of those AMV data in the forecasts over the region. The results of this study would be a pathway to use a more appropriate observation system for the forecasts over East Asia.
In this study, OSEs with the WRF regional model and three-dimensional variational (3DVAR) DA system are performed to quantitatively diagnose the fully nonlinear effect of assimilating HIMA-8 AMV data on forecasts over East Asia, and the observation impacts are evaluated using various measures (i.e., energy-based nonlinear forecast error, RMSE of geopotential height forecast, and RMSE of forecasts with respect to upper-air radiosonde observations) for the period from 0000 UTC 1 August to 1800 UTC 30 September 2015. By comparing the effect of assimilating HIMA-8 and MTSAT-2 AMVs on forecasts over East Asia, the characteristics of the AMVs are analyzed and a way to use the AMVs to improve the forecast accuracy over East Asia is suggested. Section 2 presents the method of calculating the observation impact, model, DA, observations, and experimental settings. The results are presented in section 3, and section 4 consists of a summary and a discussion.
2. Methodology
a. Observation impact by forecast error reduction
If the analysis of each experiment is used as the true state, the FER in Eq. (5) implies the impact of the observations because the difference of two forecast errors (i.e., 
b. Model and data assimilation
The WRF-ARW, version 3.8, system (Skamarock et al. 2008) developed by the National Center for Atmospheric Research (NCAR) was used. The model domain comprises 430 × 295 horizontal grid points with 18-km resolution, centered on 28°N and 132°E (Fig. 1), and has 61 vertical levels with an upper limit of 10 hPa. The parameterization schemes used are the Kain–Fritsch scheme (Kain 2004) for cumulus parameterization; the WRF single-moment 6-class scheme (Hong and Lim 2006) for microphysics parameterization; the Dudhia scheme (Dudhia 1989) and the Rapid Radiative Transfer Model (RRTM) scheme (Mlawer et al. 1997) for shortwave and longwave radiation parameterization, respectively; the Yonsei University (YSU) scheme (Hong et al. 2006) for planetary boundary layer parameterization; and the Noah land surface model (Chen and Dudhia 2001) for land surface parameterization. The NCEP Final Analysis (NCEP FNL) with 1° × 1° horizontal resolution produced from the GDAS is used for boundary conditions.
Experimental domain and the distribution of SOUND observations used. The locations of the SOUND observations used for both assimilation and verification (open circles and filled triangles, respectively) in section 3b(1) and those used for assimilation (open circles) and independent verification (filled triangles) in section 3b(2).
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
The 3DVAR of the WRF data assimilation system (WRFDA), version 3.8 (Barker et al. 2004, 2012), was used for assimilating observations. To use the background error covariance (BEC) that reflects the characteristics of the atmosphere during the experimental period, the BEC was produced using the differences between 12- and 24-h forecasts for the period from 1 August to 30 September 2015, based on the National Meteorological Center (NMC) method (Parrish and Derber 1992). The ±3-h assimilation window was used, but the ±1.5-h assimilation window was applied only for AMVs. Since the MTSAT-2 AMVs are within the ±1.5-h assimilation window at every analysis time, the HIMA-8 AMVs within the same window are used for fair comparison.
c. Observations
The observations used include conventional observations assimilated into the NCEP GDAS in Prepared Binary Universal Form for the Representation of Meteorological Data (PREPBUFR) format and Advanced Microwave Sounding Unit-A (AMSU-A) radiance data that are not included in PREPBUFR (Table 1). To assimilate the AMSU-A radiance data in WRFDA, a 90-km thinning resolution and variational bias correction were applied. The MTSAT-2 AMVs are already included in the conventional observation data. The hourly HIMA-8 AMVs are distributed in BUFR format by the JMA through the Global Telecommunications System (GTS). To minimize the spatial correlations between high-density AMVs, a 20-km thinning resolution was applied for both HIMA-8 and MTSAT-2 AMVs (Jung et al. 2013; Lin et al. 2015; Cordoba et al. 2017). The 20-km thinning was applied to the AMVs such that there is only one wind every 20 km but it could be from different satellites. The low-level (below 700 hPa) AMVs over land were not used following the blacklist of JMA (Yamashita 2016).
Details of the observations used: U, V, T, P, Q, TB, and TPW denote zonal wind, meridional wind, temperature, surface pressure, specific humidity, brightness temperature, and total precipitable water, respectively.
The quality index (QI) of MTSAT-2 AMVs was provided by PREPBUFR and that of HIMA-8 AMVs was provided by JMA. The forecast-independent QI was used for both MTSAT-2 and HIMA-8 AMVs. The average ratio of HIMA-8 AMVs (MTSAT-2 AMVs) with QI greater than 94 to those with QI greater than 70 is 58.5% (63.6%). Lean et al. (2016) showed that the root-mean-square vector difference (RMSVD) of HIMA-8 AMVs is smaller than that of MTSAT-2 AMVs. Figure 2 shows the RMSVD values of HIMA-8 AMVs for the forecast-independent and forecast-dependent QI in the model domain of this study for the period from 0000 UTC 1 September to 1800 UTC 30 September 2015. The 6-h forecasts from the WRF analysis–forecast system were used to calculate the RMSVD at 0000, 0600, 1200, and 1800 UTC. Lean et al. (2016) showed that the RMSVD of HIMA-8 AMVs does not show a trend for improvement, as the forecast-independent QI value increases for VIS and IR channels using the ECMWF model. Similar to Lean et al. (2016), the RMSVD of HIMA-8 AMVs tends to increase (decrease) as the forecast-independent (dependent) QI value increases using the WRF model in the East Asia model domain (Fig. 2). The lower thresholds for the forecast-independent QI and the forecast-dependent QI are 70 and 75, respectively (Fig. 2), which implies that using AMV data above these thresholds means using all the data.
The relationship between the RMSVD and (a) the forecast-independent QI and (b) the forecast-dependent QI of HIMA-8 AMVs.
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
The observation error covariance for MTSAT-2 AMVs used in this study is approximately 5 m s−1. Shimoji (2017) mentioned that the RMSVD for AMVs from the infrared channel of HIMA-8 is approximately 3.86–5.77 m s−1. Moreover, Météo-France uses approximately 3.6 m s−1 as the observation error covariance for HIMA-8 AMVs (https://nwpsaf.eu/monitoring/amv/amvusage/mfmodel.html). Considering these references, the wind observation error of HIMA-8 AMVs is approximately assigned as 3.8 m s−1.
Figure 3 shows the vertical distribution of time-averaged Northern Hemisphere AMVs that are produced from two satellites and used in 3DVAR for the experimental period. There are no AMVs above 100 hPa for both HIMA-8 and MTSAT-2. The HIMA-8 AMVs cover the layers lacking in upper-air wind observations in the middle and lower troposphere and increase the upper-air wind observation data. The number of AMVs in the midlayer increases by adding WV observation channels to extract AMVs in HIMA-8. The number of MTSAT-2 AMVs is greater than that of HIMA-8 AMVs, only for the layer from 800 to 900 hPa.
The average number of AMV data in the Northern Hemisphere included in the observation window (±1.5 h) at the analysis time (0000, 0600, 1200, and 1800 UTC) with pressure levels of 100 hPa for 2 months (0000 UTC 1 Aug 2015–1800 UTC 30 Sep 2015): (a) QI ≥ 70 and (b) QI ≥ 94.
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
Figure 4 shows horizontal and vertical distributions of AMVs at 0000 UTC 20 August 2015. The HIMA-8 AMVs (QI ≥ 70) exist at all three layers in the experimental domain and have the smallest region lacking in upper-air wind observations (Fig. 4a), whereas the MTSAT-2 AMVs (QI ≥ 70) exist mainly in the upper layer and have some regions lacking in upper-air wind observations (Fig. 4c). Compared with the number of MTSAT-2 AMVs (QI ≥ 70), there are 2.4 times more HIMA-8 AMVs (QI ≥ 70) in the upper layer (100–400 hPa), 11 times more in the midlayer (400–700 hPa), and 3.1 times more in the lower layer (700–1000 hPa) at 0000 UTC 20 August 2015. In addition, the number of HIMA-8 AMVs (QI ≥ 94; Fig. 4b) is 1.6 times greater in the lower and upper layers and 4.8 times greater in the midlayer than that of the MTSAT-2 AMVs (QI ≥ 70). The difference in the number of AMVs with respect to altitude apparently appears around the Typhoons Goni (1515) and Atsani (1516) near 20°N, and the region lacking in upper-air wind observations around typhoons decreases by replacing MTSAT-2 AMVs with HIMA-8 AMVs (Fig. 4). During the experimental period, compared with the number of MTSAT-2 AMVs (QI ≥ 70), there are 2.3 times more HIMA-8 AMVs (QI ≥ 70) in the upper layer (100–400 hPa), 8.1 times more in the midlayer (400–700 hPa), and 2 times more in the lower layer (700–1000 hPa) on average.
The horizontal and vertical distributions of AMV data (blue: 100–400 hPa, green: 400–700 hPa, red: 700–1000 hPa) at 0000 UTC 20 Aug 2015, for (a) HIMA-8 (QI ≥ 70), (b) HIMA-8 (QI ≥ 94), and (c) MTSAT-2 (QI ≥ 70).
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
d. Experimental setting
The analysis–forecast cycling experiments were performed using the WRF and WRFDA 3DVAR for the period from 0000 UTC August 1 to 1800 UTC 30 September 2015. Assimilating observations and producing analysis were performed every 6 h, and the model was run for 30 h starting with the analysis initialized at 0000, 0600, 1200, and 1800 UTC. The spinup time was the first week of the experimental period and was excluded from the verification of the results.
In all experiments, the observation systems commonly include the AMSU-A radiances and all conventional observations in PREPBUFR format, except for the MTSAT-2 AMVs. Table 2 represents the AMV observation selection for experiments. In Exp1 (i.e., control experiment), the MTSAT-2 AMVs are additionally assimilated with the commonly assimilated observations. In Exp2, the MTSAT-2 AMVs in Exp1 are replaced by the HIMA-8 AMVs (QI ≥ 70) to verify the effect of replacing the MTSAT-2 AMVs with the HIMA-8 AMVs on the forecast errors. The Exp3 is the same as Exp2 but only the HIMA-8 AMVs with a QI above 94 are assimilated to verify the impact of HIMA-8 AMVs depending on the QI. Here, the forecast-independent QI threshold above 94 is used because the number of HIMA-8 AMVs with a QI above 94 is approximately similar to that of MTSAT-2 AMVs with a QI above 70, on average for all layers during the experimental period. In addition, Exp4 and Exp5, in which MTSAT-2 AMVs are added to the observation systems of Exp2 and Exp3, respectively, are performed to verify the effect of assimilating the AMVs of two satellites simultaneously.
GeoAMV observation selection for experiments; QI denotes the quality index of AMV data. In addition to AMV data, all experiments commonly assimilate all conventional observations and AMSU-A radiances.
3. Results
a. Verification with respect to the analysis of each experiment
1) Forecast error
The time- and domain-averaged 24-h forecast errors along the analysis trajectory 
The average forecast error for the forecast integrated from the analysis and that integrated from the background for each experiment, and the OI based on the analysis of the experiment. The numbers in parentheses denote the standard deviation.
(a) The averaged total energy-norm forecast error (solid line: Exp1–3; dotted line: Exp4 and Exp5) based on the analysis of the experiment for the forecast integrated from the background (red) and that integrated from the analysis (blue) for each experiment. (b) OI based on the analysis of the experiment for each experiment in the energy norm (solid line: Exp1–3; dotted line: Exp4 and Exp5).
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
In each experiment, 
In terms of the total energy-norm forecast error in Exp2–5 compared with that of Exp1, 
The time series of the total energy-norm forecast error during the experimental period shows that the variability and magnitude of the forecast error (not shown) are mostly similar to the time- and domain-averaged forecast error shown in Table 3. The standard deviations of forecast errors in Exp2–4 (Exp5) are smaller (slightly greater) than that in Exp1 (Table 3). The smaller forecast errors in Exp2–4 than Exp1 and the comparable standard deviations of forecast errors in Exp1–4 imply that the forecast error ranges of Exp2–4 are smaller than those of Exp1, which confirms that the effect of assimilating HIMA-8 AMVs is consistent during the experimental period except Exp5.
2) Observation impact
The OIs in energy units for each experiment are shown in Table 3 and Fig. 5b. The OI is the largest in Exp1 (−271.50 × 105 J Kg−1), followed by Exp3 with −259.32 × 105 J Kg−1, which is 4.4% smaller than Exp1. The OIs in Exp2, Exp4, and Exp5 are 4.9%–7.2% smaller than the OI in Exp1. The OI differences among Exp1–5 are not statistically significant based on the RM ANOVA with 90% confidence. Although the OIs of specific variables show similar trends, the decreasing rate of the OIs varies depending on variables (e.g., approximately 5.5%–10.4% for U and V, 0.7%–2.7% for T). The 
3) RMSE of the 500-hPa geopotential height forecast
The time- and domain-averaged RMSE for the 500-hPa geopotential height forecast based on the analysis of each experiment as reference is shown in Fig. 6a. The RMSE of the 500-hPa geopotential height forecast is evaluated because it is associated with typhoon forecasts over East Asia (Kim et al. 2017).
(a) The average RMSE based on the analysis of the experiment for the 500-hPa geopotential height forecast of each experiment, and (b) time series of the average RMSE for the 500-hPa geopotential height forecast for Exp1–3.
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
In all experiments, the RMSE increases as the forecast time increases and the RMSEs of Exp2–5 are always smaller than that of Exp1. The RMSE differences between Exp1 and the other experiments are statistically significant with 99% confidence. The reduction rate of the RMSE is approximately 5.4% in Exp2, 7% in Exp4, 4% in Exp3, and 3.7% in Exp5, compared with that in Exp1. Thus, the RMSEs in Exp2 and Exp4 with HIMA-8 AMVs of QI ≥ 70 are smaller than those in Exp3 and Exp5 with HIMA-8 AMVs of QI ≥ 94, which reaffirms the results in section 3a(1).
Figure 6b represents the time series of the domain-averaged RMSE of the 500-hPa geopotential height forecast for the whole experimental period. The average value and fluctuation range of the RMSE is smaller at 0000 and 1200 UTC than at 0600 and 1800 UTC. The magnitude and variation of the RMSE is the largest for the period from 15 to 24 August, when the Typhoons Goni (1515) and Atsani (1516) are generated and dissipated simultaneously. There was no typhoon case in the domain from 26 August to 5 September, when the magnitude and variation of the RMSE are the smallest.
b. Verification with respect to observations
1) RMSE of the forecast with respect to SOUND observations used for the assimilation
The vertically averaged RMSEs of the 6- and 24-h forecasts in each experiment are analyzed for U, V, and T based on upper-air radiosonde (SOUND) observations assimilated in the model (Fig. 7). The locations of the SOUND observations used for both assimilation and verification are shown in Fig. 1. The RMSEs of Exp2–5 are smaller than those of Exp1, which is more noticeable in the 6-h forecast than in the 24-h forecast and in U and V than in T. As the forecast time increases, the RMSEs of the experiments increase with a similar magnitude because the impact of assimilating AMVs decreases along the forecast trajectory as the forecast time increases. The RMSE differences between the 24-h T forecasts are very small (i.e., on the order of 10−3) because T is indirectly improved by assimilating the AMVs. The RMSE of the 24-h forecast for U is larger than that of V.
The average RMSE for wind (U, V) and temperature (T) forecasts verified by SOUND observations (0000 and 1200 UTC) used in data assimilation (open circles) and not used in data assimilation (open triangles): (a),(c),(e) 6-h forecast and (b),(d),(f) 24-h forecast.
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
The RMSEs of the 6-h forecasts for U, V, and T in Exp2–5 are statistically significantly different compared with those in Exp1 at a 99% confidence level. The RMSEs of the 24-h forecasts for U in Exp2 and Exp4 and T in Exp2 are statistically significantly different at a 95% confidence level compared with those in Exp1, whereas those for V in Exp1–5 are not statistically significantly different at a 90% confidence level. The RMSEs of the 6-h forecasts for U between Exp2 and Exp3–5 and between Exp4 and Exp5, V between Exp2 and Exp3 (or Exp5) and between Exp4 and Exp5, and T between Exp2 and Exp4 (or Exp5) are statistically significantly different at a 95% confidence level. The RMSEs of the 24-h forecasts for U, V, and T between Exp2 and Exp5, those for U between Exp2 and Exp3, and those for V between Exp3 and Exp5 are statistically significantly different at a 95% confidence level.
The forecast accuracy of U and V in Exp2 is the largest; the RMSEs of the 6- and 24-h forecasts in Exp2 are reduced by 2.87% and 1.07% for U, 2.83% and 0.55% for V, and 1.40% and 0.51% for T, respectively, compared with those in Exp1. The RMSEs of 6- and 24-h forecasts in Exp3 is greater than those in Exp2, except the RMSEs of the 24-h forecasts for V, which are greater in Exp2 than in Exp3. The RMSEs of the 6- and 24-h forecasts for U and V in Exp4 are smaller than those in Exp5. This means that the forecast errors in Exp3 and Exp5 are generally greater than those in Exp2 and Exp4, respectively. Assimilating AMVs with a QI above 70 (Exp2) is more effective for the reduction of the RMSE than assimilating AMVs with a QI above 94 (Exp3), which is the same as the result based on the analysis of each experiment as referenced in section 3a(1). In addition, the RMSE of Exp2 is smaller than that of Exp4, which is also similar to the result in section 3a(1), which reaffirms that replacing MTSAT-2 AMVs with HIMA-8 AMVs affects the reduction of the forecast error more compared with assimilating the AMVs from both satellites together.
For the vertical profile of RMSEs of the 6-h forecasts in each experiment (Figs. 8a–c) based on SOUND observations used in DA, the RMSEs of Exp2–5 are smaller than the RMSE of Exp1, and the RMSEs of Exp2 and Exp4 are generally smaller than those of other experiments, which is noticeable above the midtroposphere. The smaller RMSEs in Exp2 and Exp4 compared to other experiments above the midtroposphere are related to the vertically varying AMV distributions shown in Figs. 3 and 4, which are associated with the spatially and temporally higher data numbers of HIMA-8 AMVs above the midtroposphere and the improved derivation algorithm for HIMA-8 AMVs.
The vertical profile of RMSE for wind (U, V) and temperature (T) 6-h forecasts verified by SOUND observations (0000 and 1200 UTC) (a)–(c) used in data assimilation and (d)–(f) not used in data assimilation.
Citation: Journal of Atmospheric and Oceanic Technology 35, 9; 10.1175/JTECH-D-17-0093.1
2) RMSE of the forecast with respect to SOUND observations that are not used in assimilation
To evaluate the impact of HIMA-8 AMVs based on SOUND observations that are not assimilated, the OSEs are performed without SOUND observations from eight observation sites and the experiments are named Exp1-1, Exp2-1, Exp3-1, Exp4-1, and Exp5-1. The SOUND observations that are not used in the DA are presented as triangles in Fig. 1, which are evenly distributed in the domain.
The vertically averaged RMSEs for U, V, and T based on the independent SOUND observations are marked as open triangles in Fig. 7. As the forecast time increases, the RMSEs of the experiments increase for U, V, and T and become relatively similar. The RMSEs based on the assimilated and not assimilated SOUND observations have similar trends, except the RMSEs for U that were reduced more in Exp3 and Exp5 than in Exp2 and Exp4. The RMSEs based on independent SOUND observations are generally greater than those based on assimilated SOUND observations, except the RMSEs of 6- and 24-h T forecasts. The smaller RMSEs of 6- and 24-h T forecasts based on independent SOUND observations may be caused by the indirect improvement of T when assimilating the AMVs. The RMSEs in Exp2 and Exp4 generally decrease compared with Exp1, with the smallest RMSE in Exp2. The RMSEs of the 6- and 24-h forecasts in Exp2 are reduced by 6.49% and 4.21% for U and 6.48% and 2.69% for V, respectively, compared with those of Exp1, which is a greater RMSE reduction compared with that based on the assimilated SOUND observations in section 3b(1). The RMSEs of the 6-h forecast for U, V, and T in Exp2–5 are statistically significantly different from those in Exp1 at a 95% confidence level, whereas the RMSEs of the 24-h forecast for U in Exp5 and V and T in Exp2–5 (except U in Exp2–4) are not statistically significantly different from those in Exp1 at a 90% confidence level. Unlike the vertical profile of RMSEs of the 6-h forecasts in each experiment based on SOUND observations used in DA (Figs. 8a–c), the RMSEs of Exp2 and Exp3 based on independent SOUND observations are smaller than those of other experiments, especially above the midtroposphere (Figs. 8d–f).
c. Additional experiments using the forecast-dependent QI
The RMSVD of HIMA-8 AMVs tends to decrease as the forecast-dependent QI value increases (Fig. 2b), which implies the agreement between the WRF Model and the JMA model that is used to deduce the forecast-dependent QI value. Thus, when the HIMA-8 AMVs with the high forecast-dependent QI (≥90) are included in the observation system, the energy-norm forecast error is further reduced compared to when all HIMA-8 AMVs (forecast-dependent QI ≥ 75 or forecast-independent QI ≥ 70) are assimilated (Exp2 or Exp4 when using the forecast-independent QI) (not shown). In addition, when the HIMA-8 AMVs with the high forecast-dependent QI (≥90) replace MTSAT-2 AMVs, the energy-norm forecast error is reduced the greatest (not shown). Here, the forecast-dependent QI threshold of 90 is used because the number of HIMA-8 AMVs with QI ≥ 90 is approximately similar to that of MTSAT-2 AMVs with QI ≥ 75.
In contrast, when all HIMA-8 AMVs with the forecast-dependent QI (≥75) are assimilated, the RMSEs of the 500-hPa geopotential height forecast and those of the forecast with respect to SOUND observations are greatly reduced (not shown), which is different from the results for the energy-norm forecast error. The differences in the results based on the energy-norm forecast error and the RMSE of the 500-hPa geopotential height forecast is caused by the differences in the vertical layers considered (total layers for total energy norm and 500 hPa for geopotential height) and different variables used to deduce the total energy and geopotential height. Compared with the energy-norm forecast error and associated OI, the RMSE of the geopotential height forecast tends to be smaller when assimilating more AMVs from two satellites than assimilating a small number of high forecast-dependent QI HIMA-8 AMVs. The energy-norm forecast error and OI are more closely associated with the quality of assimilated wind data (i.e., AMVs) in terms of the forecast-dependent QI, whereas the RMSE of the geopotential height seems to be associated with the number of AMVs assimilated because the geopotential height is indirectly constrained by AMVs.
Interestingly, the RMSEs of the 500-hPa geopotential height forecast and those of the forecast with respect to SOUND observations are greatly reduced when all HIMA-8 AMVs are assimilated, which does not require the classification of AMV data depending on the forecast-independent or forecast-dependent QIs.
4. Summary and discussion
In this study, the effect of HIMA-8 AMVs on the forecast error over East Asia is evaluated using OSEs with the WRF and WRFDA 3DVAR system. The experimental period is 1 August–30 September 2015, during which AMVs produced by both MTSAT-2 and HIMA-8 satellites exist. The HIMA-8 produces higher-quality and a larger number of AMVs than MTSAT-2. Especially in the midlayer (400–700 hPa), in terms of raw data, the number of the HIMA-8 AMVs is 8.1 times greater than that of the MTSAT-2 AMVs on average during the experimental period. The control experiment (i.e., Exp1) assimilates conventional observations, AMSU-A, and MTSAT-2 AMVs.
The impact of HIMA-8 AMVs is evaluated by calculating the energy-norm forecast error based on the analysis of each experiment. The reduction of the forecast error is greater in the experiment in which HIMA-8 AMVs are assimilated compared with the experiment in which MTSAT-2 AMVs are assimilated. When replacing MTSAT-2 AMVs with HIMA-8 AMVs rather than adding HIMA-8 AMVs to the existing MTSAT-2 AMVs, the 24- and 30-h energy-norm forecast errors are reduced by 1.02% and 1.97%, respectively, compared with those in the control experiment, with a statistical significance at the 99% confidence level in 
In terms of the RMSE of the 500-hPa geopotential height forecast based on the analysis of each experiment, the RMSE decreases more effectively when assimilating all AMVs from MTSAT-2 and HIMA-8, different from the energy-norm forecast error. Therefore, based on the analysis of each experiment as reference, the effect of HIMA-8 AMVs on the energy-norm forecast error and the RMSE of the 500-hPa geopotential height forecast varies depending the number of the assimilated AMVs. The energy-norm forecast error is more dependent on the direct assimilation of wind data, while the RMSE of the 500-hPa geopotential height is more dependent on the number of AMVs because the wind data indirectly constrain the geopotential height field. If either assimilated or not-assimilated SOUND observations are used as reference state, the RMSEs of the 6- and 24-h forecasts are smallest when more HIMA-8 AMVs (QI ≥ 70) are assimilated.
The high forecast-independent QI values do not seem to indicate high-quality data; thus, the selective assimilation of HIMA-8 AMVs based on the forecast-independent QI threshold is not beneficial in reducing the forecast error in any measurements. The forecast-dependent QI values could be used to select the HIMA-8 AMVs in reducing the energy-norm forecast error but not the RMSE of the 500-hPa geopotential height forecast or the RMSE with respect to the SOUND observations.
In terms of several verifications, replacing MTSAT-2 AMVs with HIMA-8 AMVs has a positive effect on the forecast improvement over East Asia. The effect of HIMA-8 AMVs varies depending on the reference state used to calculate the forecast error and RMSE. A larger number of HIMA-8 AMVs is more effective when the analysis of each experiment is used as reference for calculating the energy-norm forecast error and when the SOUND observations are used as reference. Assimilating both MTSAT-2 and HIMA-8 AMVs is more effective in reducing the RMSE of the 500-hPa geopotential height forecast when the analysis of each experiment is used as reference. Thus, the assimilation of AMVs from several satellites (e.g., Himawari, GK-2A, FY, etc.) simultaneously over East Asia may have benefits to constrain the 500-hPa geopotential height over East Asia.
The results of this study provide guidance for the use of HIMA-8 AMVs to improve NWP over East Asia. The effect of AMVs produced by the next-generation geostationary orbit satellite [Geostationary Korea Multipurpose Satellite-2A (GK-2A)] in Korea can also be roughly diagnosed before its launch because GK-2A has an observation area similar to that of HIMA-8. Designing the optimal observation system including GK-2A and HIMA-8 AMVs to improve NWP over East Asia is potential future work to be addressed in a further study.
Acknowledgments
The authors appreciate three reviewers for their valuable comments. This study was supported by the Korea Meteorological Administration Research and Development Program under Grant KMIPA 2015-5200 and the Korea Polar Research Institute (KOPRI; PN17081). The authors appreciate the Japan Meteorological Agency and the National Meteorological Satellite Center of the Korea Meteorological Administration for providing Himawari-8 AMV data and information for this study.
REFERENCES
Baker, N. L., and R. Daley, 2000: Observation and background adjoint sensitivity in the adaptive observation-targeting problem. Quart. J. Roy. Meteor. Soc., 126, 1431–1454, https://doi.org/10.1002/qj.49712656511.
Barker, D., W. Huang, Y.-R. Guo, A. J. Bourgeois, and Q. N. Xiao, 2004: A three-dimensional variational data assimilation system for use with MM5: Implementation and initial results. Mon. Wea. Rev., 132, 897–914, https://doi.org/10.1175/1520-0493(2004)132<0897:ATVDAS>2.0.CO;2.
Barker, D., and Coauthors, 2012: The Weather Research and Forecasting model’s Community Variational/Ensemble Data Assimilation System: WRFDA. Bull. Amer. Meteor. Soc., 93, 831–843, https://doi.org/10.1175/BAMS-D-11-00167.1.
Buizza, R., C. Cardinali, G. Kelly, and J.-N. Thepaut, 2007: The value of observations. II: The value of observations located in singular-vector-based target areas. Quart. J. Roy. Meteor. Soc., 133, 1817–1832, https://doi.org/10.1002/qj.149.
Cardinali, C., 2009: Monitoring the observation impact on the short-range forecast. Quart. J. Roy. Meteor. Soc., 135, 239–250, https://doi.org/10.1002/qj.366.
Cardinali, C., L. Isaksen, and E. Anderson, 2003: Use and impact of automated aircraft data in a global 4DVAR data assimilation system. Mon. Wea. Rev., 131, 1865–1877, https://doi.org/10.1175//2569.1.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modeling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–585, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Cordoba, M., S. L. Dance, G. A. Kelly, N. K. Nichols, and J. A. Waller, 2017: Diagnosing atmospheric motion vector observation errors for an operational high-resolution data assimilation system. Quart. J. Roy. Meteor. Soc., 143, 333–341, https://doi.org/10.1002/qj.2925.
Dudhia, J., 1989: Numerical study of convection observed during the Winter Monsoon Experiment using a mesoscale two-dimensional model. J. Atmos. Sci., 46, 3077–3107, https://doi.org/10.1175/1520-0469(1989)046<3077:NSOCOD>2.0.CO;2.
Fourrie, N., A. Doerenbecher, T. Bergot, and A. Joly, 2002: Adjoint sensitivity of the forecast to TOVS observations. Quart. J. Roy. Meteor. Soc., 128, 2759–2777, https://doi.org/10.1256/qj.01.167.
Gelaro, R., and Y. Zhu, 2009: Examination of observation impacts derived from observing system experiments (OSEs) and adjoint models. Tellus, 61A, 179–193, https://doi.org/10.1111/j.1600-0870.2008.00388.x.
Gelaro, R., R. H. Langlang, S. Pellerin, and R. Todling, 2010: The THORPEX observation impact intercomparison experiment. Mon. Wea. Rev., 138, 4009–4025, https://doi.org/10.1175/2010MWR3393.1.
Gueorguieva, R., and J. H. Krystal, 2004: More over ANOVA: Progress in analyzing repeated-measures data and its reflection in papers published in the Archives of General Psychiatry. AMAArch. Gen. Psychiatry, 61, 310–317, https://doi.org/10.1001/archpsyc.61.3.310.
Hidehiko M., T. Masaya, and K. Yuki, 2015: VIS and IR bands of Himawari-8/AHI compatible with those of MTSAT-2/Imager. JMA Meteorological Satellite Center Tech. Note 60, 18 pp.
Hong, S.-Y., and J.-O. Lim, 2006: The WRF single-moment 6-class microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129–151.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with and explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
Jung, B.-J., H. M. Kim, Y.-H. Kim, E.-H. Jeon, and K.-H. Kim, 2010: Observation system experiments for Typhoon Jangmi (200815) observed during T-PARC. Asia-Pac. J. Atmos. Sci., 46, 305–316, https://doi.org/10.1007/s13143-010-1007-y.
Jung, B.-J., H. M. Kim, F. Zhang, and C.-C. Wu, 2012: Effect of targeted dropsonde observations and best track data on the track forecasts of Typhoon Sinlaku (2008) using an ensemble Kalman filter. Tellus, 64B, 14984, https://doi.org/10.3402/tellusa.v64i0.14984.
Jung, B.-J., H. M. Kim, T. Auligne, X. Zhang, X. Zhang, and X.-Y. Huang, 2013: Adjoint-derived observation impact using WRF in the western North Pacific. Mon. Wea. Rev., 141, 4080–4097, https://doi.org/10.1175/MWR-D-12-00197.1.
Kain, J. S., 2004: The Kain–Fritsch convective parameterization: An update. J. Appl. Meteor., 43, 170–181, https://doi.org/10.1175/1520-0450(2004)043<0170:TKCPAU>2.0.CO;2.
Kelly, G., J.-N. Thepaut, R. Buizza, and C. Cardinali, 2007: The value of observations. I: Data denial experiments for the Atlantic and the Pacific. Quart. J. Roy. Meteor. Soc., 133, 1803–1815, https://doi.org/10.1002/qj.150.
Kieu, C. Q., N. M. Truong, H. T. Mai, and T. Ngo-Duc, 2012: Sensitivity of the track and intensity forecasts of Typhoon Megi (2010) to satellite-derived atmospheric motion vectors with the ensemble Kalman filter. J. Atmos. Oceanic Technol., 29, 1794–1810, https://doi.org/10.1175/JTECH-D-12-00020.1.
Kim, H. M., and B.-J. Jung, 2006: Adjoint-based forecast sensitivities of Typhoon Rusa. Geophys. Res. Lett., 33, L21813, https://doi.org/10.1029/2006GL027289.
Kim, H. M., and B.-J. Jung, 2009a: Singular vector structure and evolution of a recurving tropical cyclone. Mon. Wea. Rev., 137, 505–524, https://doi.org/10.1175/2008MWR2643.1.
Kim, H. M., and B.-J. Jung, 2009b: Influence of moist physics and norms on singular vectors for a tropical cyclone. Mon. Wea. Rev., 137, 525–543, https://doi.org/10.1175/2008MWR2739.1.
Kim, H. M., M. Morgan, and R. E. Morss, 2004: Evolution of analysis error and adjoint-based sensitivities: Implications for adaptive observations. J. Atmos. Sci., 61, 795–812, https://doi.org/10.1175/1520-0469(2004)061<0795:EOAEAA>2.0.CO;2.
Kim, H. M., J. K. Kay, and B.-J. Jung, 2008: Application of adjoint-based forecast sensitivities to Asian dust transport events in Korea. Water Air Soil Pollut., 195, 335–343, https://doi.org/10.1007/s11270-008-9750-8.
Kim, H. M., J. K. Kay, E.-G. Yang, S. Kim, and M. Lee, 2013: Statistical adjoint sensitivity distributions for meteorological forecast errors of Asian dust transport events in Korea. Tellus, 65B, 20 554, https://doi.org/10.3402/tellusb.v65i0.20554.
Kim, M., H. M. Kim, J. Kim, S.-M. Kim, C. Velden, and B. Hoover, 2017: Effect of enhanced satellite-derived atmospheric motion vectors on numerical weather prediction in East Asia using an adjoint-based observation impact method. Wea. Forecasting, 32, 579–594, https://doi.org/10.1175/WAF-D-16-0061.1.
Kim, S.-M., and H. M. Kim, 2014: Sampling error of observation impact statistics. Tellus, 66A, 25435, https://doi.org/10.3402/tellusa.v66.25435.
Kim, S.-M., and H. M. Kim, 2017: Adjoint-based observation impact of Advanced Microwave Sounding Unit-A (AMSU-A) on the short-range forecast in East Asia (in Korean). Atmosphere, 27, 93–104, https://doi.org/10.14191/Atmos.2017.27.1.093.
Langland, R. H., and Coauthors, 1999: The North Pacific Experiment (NORPEX-98): Targeted observations for improved North American weather forecasts. Bull. Amer. Meteor. Soc., 80, 1363–1384, https://doi.org/10.1175/1520-0477(1999)080<1363:TNPENT>2.0.CO;2.
Lean, K., N. Bormann, and K. Salonen, 2016: Assessment of Himawari-8 AMV data in the ECMWF system. EUMETSAT/ECMWF Fellowship Programme Research Rep. 42, 25 pp.
Le Marshall, J., and Coauthors, 2013: The operational generation of continuous winds in the Australian region and their assimilation with 4DVAR. Wea. Forecasting, 28, 504–514, https://doi.org/10.1175/WAF-D-12-00018.1.
Lin, L.-F., A. M. Ebtehaj, R. L. Bras, A. N. Flores, and J. Wang, 2015: Dynamical precipitation downscaling for hydrologic applications using WRF 4D-Var data assimilation: Implications for GPM era. J. Hydrometeor., 16, 811–829, https://doi.org/10.1175/JHM-D-14-0042.1.
Ma, Z., E. S. Maddy, B. Zhang, T. Zhu, and S. A. Boukabara, 2017: Impact assessment of Himawari-8 AHI data assimilation in NCEP GDAS/GFS with GSI. J. Atmos. Oceanic Technol., 34, 797–815, https://doi.org/10.1175/JTECH-D-16-0136.1.
Martinet, P., L. Lavanant, N. Fourrie, F. Rabier, and A. Gambacorta, 2014: Evaluation of a revised IASI channel selection for cloudy retrievals with a focus on the Mediterranean basin. Quart. J. Roy. Meteor. Soc., 140, 1563–1577, https://doi.org/10.1002/qj.2239.
Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and S. A. Clough, 1997: Radiative transfer for inhomogeneous atmosphere: RRTM, a validated correlated-k model for the longwave. J. Geophys. Res., 102, 16 663–16 682, https://doi.org/10.1029/97JD00237.
Parrish, D. F., and J. C. Derber, 1992: The National Meteorological Center’s spectral statistical-interpolation analysis system. Mon. Wea. Rev., 120, 1747–1763, https://doi.org/10.1175/1520-0493(1992)120<1747:TNMCSS>2.0.CO;2.
Rabier, G., E. Klinker, P. Courtier, and A. Hollingsworth, 1996: Sensitivity of forecast errors to initial conditions. Quart. J. Roy. Meteor. Soc., 122, 121–150, https://doi.org/10.1002/qj.49712252906.
Schmetz, J., K. Holmlund, J. Hoffman, B. Strauss, B. Mason, V. Gaertner, A. Koch, and L. van de Berg, 1993: Operational cloud motion winds from Meteosat infrared images. J. Appl. Meteor., 32, 1206–1225, https://doi.org/10.1175/1520-0450(1993)032<1206:OCMWFM>2.0.CO;2.
Shimoji K., 2017: Introduction to the Himawari-8 Atmospheric Motion Vector algorithm. JMA Meteorological Satellite Center Tech. Note 62, 5 pp.
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., http://dx.doi.org/10.5065/D68S4MVH.
Tomassini, M., G. Kelly, and R. Saunders, 1999: Use and impact of satellite atmospheric motion winds on ECMWF analyses and forecasts. Mon. Wea. Rev., 127, 971–986, https://doi.org/10.1175/1520-0493(1999)127<0971:UAIOSA>2.0.CO;2.
Tremolet, Y., 2007: Model-error estimation in 4D-Var. Quart. J. Roy. Meteor. Soc., 133, 1267–1280, https://doi.org/10.1002/qj.94.
Velden, C., C. M. Hayden, S. J. Nieman, W. P. Menzel, S. Wanzong, and J. S. Goerss, 1997: Upper-tropospheric winds derived from geostationary satellite water vapor observations. Bull. Amer. Meteor. Soc., 78, 173–195, https://doi.org/10.1175/1520-0477(1997)078<0173:UTWDFG>2.0.CO;2.
Velden, C., and Coauthors, 2005: Recent innovations in deriving tropospheric winds from meteorological satellites. Bull. Amer. Meteor. Soc., 86, 205–223, https://doi.org/10.1175/BAMS-86-2-205.
Wang, D., X. Liang, and Y. Duan, 2006: Impact of four-dimensional variational data assimilation of atmospheric motion vectors on tropical cyclone track forecasts. Wea. Forecasting, 21, 663–669, https://doi.org/10.1175/WAF940.1.
Wu, T.-C., H. Liu, S. J. Majumdar, C. S. Velden, and J. L. Anderson, 2014: Influence of assimilating satellite-derived atmospheric motion vector observations on numerical analyses and forecasts of tropical cyclone track and intensity. Mon. Wea. Rev., 142, 49–71, https://doi.org/10.1175/MWR-D-13-00023.1.
Wu, T.-C., C. S. Velden, S. J. Majumdar, H. Liu, and J. L. Anderson, 2015: Understanding the influence of assimilating subsets of enhanced atmospheric motion vectors on numerical analyses and forecasts of tropical cyclone track and intensity with an ensemble Kalman filter. Mon. Wea. Rev., 143, 2506–2531, https://doi.org/10.1175/MWR-D-14-00220.1.
Xiao, Q., and Coauthors, 2008: Application of an adiabatic WRF adjoint to the investigation of the May 2004 McMurdo, Antarctica, severe wind event. Mon. Wea. Rev., 136, 3696–3713, https://doi.org/10.1175/2008MWR2235.1.
Yamashita, K., 2016: Assimilation of Himawari-8 atmospheric motion vectors into the numerical weather prediction systems of Japan Meteorological Agency. Proc. 13th Int. Winds Workshop, Monterey, CA, International Winds Working Group, http://cimss.ssec.wisc.edu/iwwg/iww13/proceedings_iww13/papers/session4/IWW13_Session4_3_Yamashita_final.pdf.
Zapotocny, T. H., J. A. Jung, J. Le Marshall, and R. E. Treadon, 2008: A two-season impact study of four satellite data types and rawinsonde data in the NCEP global data assimilation system. Wea. Forecasting, 23, 80–100, https://doi.org/10.1175/2007WAF2007010.1.
