Satellite Radiance Assimilation in the JMA Operational Mesoscale 4DVAR System

MARCH 2014 K A Z U M O R I 1361

MASAHIRO KAZUMORI Japan Meteorological Agency, Tokyo, Japan

(Manuscript received 27 April 2013, in ﬁnal form 1 October 2013)

ABSTRACT

The direct radiance assimilation scheme used in the Japan Meteorological Agency (JMA) global analysis system is applied to the JMA mesoscale four-dimensional variational data assimilation (4DVAR) system with two modifications. First, the data-thinning distance is shortened, and, second, the atmospheric profiles are extrapolated from the mesoscale model top to the radiative transfer model top using the U.S. Standard At- mosphere lapse rate. Although the variational bias correction method is widely used in many numerical weather prediction centers for global radiance assimilations, a radiance bias correction method for regional models has not been established because of difficulties in estimating the biases within limited regions and times. This paper examined the use of the bias correction coefficients estimated in the global system for the mesoscale system when the radiance data were introduced instead of the retrievals. It was found that the profile extrapolation was necessary to reduce the biases. Moreover, the use of common bias coefficients enables the use of the radiance data in the same way as the global system. The radiance data assimilation experiments in the mesoscale system demonstrated considerable improvements to the tropospheric geopotential height forecasts and precipitation forecasts. The improvements resulted from the introduction of radiance data from multiple satellites into data-sparse regions and times. However, the major effect of the radiance assimilation on the precipitation forecasts was limited to weak precipitation areas over oceans; the effects on deep convective areas and over land were relatively small.

1. Introduction For the simulation of the radiance, fast radiative transfer models have been developed and utilized by the opera- Satellite radiance data are currently assimilated in tional NWP centers. The two most commonly used fast many operational numerical weather prediction (NWP) radiative transfer models are the Radiative Transfer for centers primarily using the variational data assimilation the Television Infrared Observation Satellite (TIROS) approach (Derber and Wu 1998; Andersson et al. 1994). Operational Vertical Sounder (RTTOV; Saunders et al. The radiance data are one of the most important obser- 1999) and the Community Radiative Transfer Model vations for global forecast performance, especially over (CRTM; Weng 2007). In practice, biases between the areas where conventional observations are limited (e.g., observed and simulated radiances are present and should the Southern Hemisphere). Theoretically, direct radiance be properly corrected. Eyre (1992) proposed a bias- assimilation is superior to retrieval assimilation because correction method that can estimate the bias from two complicated error contamination in various retrieval components. The first component is dependent on the processes can be avoided. Furthermore, the radiance as- scan angle; the second component depends on the air similation allows for the early introduction of the radi- mass. Moreover, the airmass-dependent bias is expressed ance data into the operational systems. Moreover, its as a linear function of several predictors. The coefficients rapid, real-time processing without the retrieval steps is of the predictors are assumed to be channel dependent an advantage for operational data usage. and can be determined offline (Harris and Kelly 2001) or To use the radiance data in the variational data as- adaptively updated within a minimization process in the similation, a forward model for transforming analysis variational assimilation system (VarBC; Dee 2005). variables into radiance and its adjoint model are required. Although the VarBC method is widely used in global data assimilation systems, a bias correction method for Corresponding author address: Masahiro Kazumori, 1-3-4 Otemachi, regional data assimilation has not been established Chiyoda-ku, Tokyo 100-8122, Japan. because of difficulties in estimating the biases within E-mail: [email protected] limited regions and times. The global radiance dataset

DOI: 10.1175/MWR-D-13-00135.1

Ó 2014 American Meteorological Society Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1362 MONTHLY WEATHER REVIEW VOLUME 142 contains observational information on various meteo- the fine structures and rapid changes of mesoscale phe- rological conditions. However, meteorological phenomena in the initial NWP fields. nomena in a regional model are not sufficient to capture In the JMA mesoscale NWP system [sections 3.7 and general trends in the radiance bias within its limited 4.5 of JMA (2007)], satellite data have been retrieved domain. Polar-orbiting satellite coverage is nonuniform from several sources, including temperature profiles in the limited area and highly variable. Therefore, adap- from the operational polar-orbiting meteorological sat- tively estimated bias coefficients using only the regional ellites’ temperature sounders [the National Oceanic and model information and limited sampled radiance data are Atmospheric Administration (NOAA) and Meteoro- not as robust as those used in the global systems. More- logical Operational Satellite (MetOp) Advanced TIROS over, the model-top height generally restricts the usage of Operational Vertical Sounders (ATOVSs)] and the total high peaking channels. Regional models often focus on precipitable water and rain rate from research and de- weather events in the troposphere, and the model-top velopment satellite microwave imagers [the Defense height is set lower than in global models to save compu- Meteorological Satellite Program (DMSP) Special Sen- tational costs in the operational system. This limitation sor Microwave Imager (SSM/I), the Tropical Rainfall might cause additional bias in the simulated radiances. Measuring Mission (TRMM) Microwave Imager (TMI), Handling the radiance bias correction in the regional and the Aqua Advanced Microwave Scanning Radiom- data assimilation is challenging, and the best solution eter for Earth Observing System (AMSR-E)]. The in- remains an open topic. formation on temperature and moisture from the satellite Accurate precipitation forecasting is one of the most data has been indispensable in the production of realistic important tools in preventing natural disasters. The Japan initial fields and forecasts. The observational data used in Meteorological Agency (JMA) operates a global NWP the JMA mesoscale NWP system are outlined in section system and a mesoscale NWP system to provide guidance 3.2 of JMA (2007) and section 2.2 of JMA (2013). for issuing warnings and very short-range forecasts of The primary goal of the present study is to examine precipitation in Japan and its surrounding areas (JMA the impact of satellite radiance assimilation in the JMA 2013). Japan is an island country and suffers annually operational mesoscale NWP system. In the development from the heavy precipitation caused by typhoons and of the radiance assimilation scheme for the mesoscale seasonal fronts. The main source of the precipitation is system, the same radiative transfer model (RTTOV-9) moisture coming from the ocean. Therefore, the use of and the same preprocessor (Okamoto et al. 2005) were satellite observation data gathered from over the applied. However, as discussed above, a bias correction oceans is critical in the analysis of the JMA mesoscale scheme in the regional model has not been established, NWP system to produce correct and accurate initial and it is necessary to develop a method to handle the conditions. lower model-top height in the mesoscale model. To A dense, ground-based observation network is avail- overcome these difficulties for the mesoscale system, two able in and around Japan, including surface weather modifications were applied to the preprocessor. First, the stations, ground-based radars, upper-air soundings from data-thinning distance was shortened. The second modi- radiosondes and aircrafts, wind profilers, drifting buoys, fication was to extrapolate the atmospheric profile from ships, and ground-based GPS-integrated water vapor the mesoscale model top to the radiative transfer model data over the land. These in situ observations and top using the U.S. Standard Atmosphere, 1976 lapse rates. ground-based remote sensing data can capture the fine Moreover, estimated bias correction coefficients in the structure and rapid changes of mesoscale phenomena global radiance assimilation system were adopted for use over short time intervals in some areas. However, the in the mesoscale system. Before the radiance assimilation available ground-based data collection areas are limited scheme was operational, experiments comparing radi- to land-based systems and locations near the coast. ance assimilation and retrieval assimilation were con- Therefore, there are many temporal and spatial gaps in ducted using the JMA mesoscale NWP system. These the data coverage. experiments demonstrated the advantage of radiance In contrast, remote sensing observation platforms in assimilation in the operational mesoscale data assimila- space provide wide coverage of temperature, moisture, tion system. Several data-denial experiments were ap- and wind. Geostationary satellite data are available at plied to a typhoon event to investigate the effects of high-frequency time intervals. With regard to polar- individual radiance components on the mesoscale fore- orbiting satellites, although the temporal frequencies casts. The denial experiments revealed the impacts of the are not sufficient for use in operational regional models, radiance dataset in the mesoscale analysis and demon- data from multiple satellites can be used to fill gaps. strated the positive effects of the radiance data assimila- Frequent analysis updates enable the representation of tion in the typhoon case.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1363

Section 2 describes the JMA mesoscale analysis sys- the initial conditions. This final step is called the second tem, its preprocessor, and the modifications of the me- outer loop, and the high-resolution model in the outer soscale system. Section 3 introduces the experiments loops is called the outer model. The outer model is the that compare the radiance data assimilation and the same model used for the mesoscale forecast product (33-h retrieval data assimilation using the operational config- forecast). Two outer loops and one inner loop are used uration. Section 4 explains the denial experiments for for the JMA mesoscale 4DVAR system. The outer the typhoon case. Finally, a discussion and future pros- model has 50 vertical levels and the inner model for the pects are provided in section 5. minimization in the data assimilation has 40 vertical levels. The model-top height for both models is approximately 22 km (approximately 40 hPa). The lateral 2. The mesoscale analysis system boundary conditions for the analyses are obtained from the JMA global model forecast. The initial field as the a. Forecast model and analysis system first guess is taken from the previous mesoscale analysis. JMA introduced the world’s first regional four- A full description of the JMA NWP system is summa- dimensional variational data assimilation (4DVAR) rized in JMA (2013). system in 2002 (Ishikawa and Koizumi 2002), and b. Observational data in the JMA mesoscale the JMA mesoscale model was upgraded to a non- NWP system hydrostatic model (Saito et al. 2006) in 2004. A nonhydrostatic model-based 4DVAR system, the JMA In the JMA operational mesoscale NWP system, Nonhydrostatic Model-Based Variational Analysis various real-time observational data are available. In Data Assimilation scheme (JNoVA; Honda et al. 2005), terms of conventional data, reports from surface replaced the former hydrostatic 4DVAR scheme in 2009 weather stations, ships, and buoys (surface pressure); (Honda and Sawada 2009). The initial fields for the ground-based radar data (wind and retrieved relative forecasts are produced in JNoVA. The analyses are humidity); upper-air sounding data with radiosondes performed every 3 h. The JMA mesoscale model pro- (temperature, humidity, and wind); aircraft data (tem- duces short-range forecasts (15-h forecasts from 0000, 0600, perature and wind); wind profilers (wind); and ground- 1200, and 1800 UTC initializations and 33-h forecasts from based GPS integrated water vapor data are available. 0300, 0900, 1500, and 2100 UTC initializations). The model For typhoons over the western North Pacific, bogus ty- domain covers Japan and its surrounding areas (Fig. 1). phoon data, as a form of pseudo-observations are gen- Each analysis involves a 3-h assimilation time window. erated with a statistical method and assimilated for The cutoff time for receiving data is 50 min after the end a realistic tropical cyclone structure analysis. The data of the assimilation window. For the mesoscale analysis, are surface pressure and wind profiles from 1000 to a high-resolution (5-km horizontal resolution) forecast 300 hPa (JMA 2013). Additionally, atmospheric motion model is used for the assimilation window to obtain the vectors (wind data) from a geostationary satellite first-guess field and departures from the observations, (Multifunctional Transport Satellite, MTSAT) derived and the first guess is calculated. This process is called the by tracking successive cloud and moisture features are first outer loop, and the model is called the outer model. available. In terms of radiance data, polar-orbiting sat- Then, a minimization process to find an optimal field is ellite [i.e., Aqua AMSR-E, TRMM TMI, DMSP SSM/I executed on a low-resolution (15-km horizontal resolu- (and Special Sensor Microwave Imager/Sounder, tion) domain. This process is called the inner loop, and SSMIS), NOAA Advanced Microwave Sounding Unit the low-resolution model is called the inner model. For A (AMSU-A) and Microwave Humidity Sounder the minimization of a cost function in the low-resolution (MHS), and MetOp AMSU-A and MHS radiances] and space, a simplified nonlinear version of the mesoscale geostationary satellite [i.e., MTSAT Clear Sky Radiance, model is used to provide trajectories at every iteration CSR) data are available. CSR data from MTSAT-1R and instead of a tangent linear model of the mesoscale model -2 are routinely produced in the JMA Meteorological because of discontinuity and nonlinearity. An adjoint Satellite Center (MSC; Uesawa 2009, 39–48). The model of the mesoscale model provides gradient in- MTSAT CSR data are hourly clear-sky radiance data formation. The JMA analysis system uses an in- and are produced in real time. The data are superobbing cremental approach (Courtier et al. 1994). Estimated data of MTSAT clear-sky radiance data. The horizontal analysis increments in the low-resolution space are grid size of the superobbing is 60 km. The data include added to the high-resolution first guess. Moreover, the quality information, for example, the mean, maximum, high-resolution (5-km horizontal resolution) forecast and minimum values, as well as the standard deviation, model is used for the data assimilation window to obtain for the sampled clear-sky radiances in the superobbing.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1364 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 1. Data coverage of (a) AMSU-A radiance data and (b) ATOVS temperature retrievals over 3-hourly analysis times in the JMA mesoscale model domain on 1 Jul 2010. In (a) and (b), non-black-colored points indicate the data ingested into the analysis after the quality control. Small black-colored points indicate rejected data in the quality control.

Moreover, the data cover the hemisphere centered at Service (NESDIS) through the Global Telecommuni- 1408 east, including Japan and Australia. Because cation System (GTS) and the Internet, NOAA and MTSAT-1R was in operational mode during the exper- MetOp satellite data are directly received at the JMA iment periods in this study, MTSAT-1R CSR data were MSC. Moreover, data exchanged through the World used instead of the MTSAT-2 CSR data. Meteorological Organization (WMO) Regional ATOVS In addition to the global data delivered from the Na- Retransmission Service (RARS; WMO 2009) in the tional Environmental Satellite, Data, and Information Asia–Paciﬁc region are also available in real time. The

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1365 typical coverage of operationally available AMSU-A the mesoscale model-top height is approximately radiance and temperature data in the mesoscale model 40 hPa. The mesoscale model-top height is not suffi- domain is shown in Fig. 1. The mesoscale analysis uses ciently high to incorporate the higher peaking channels the same domain as the mesoscale model. Because the of the temperature-sounding instruments (AMSU-A) radiance data are level 1 data, all observed data are re- in the radiance calculations. Therefore, the atmospheric ported and can be applied using our own quality controls profiles from the mesoscale model are extrapolated (QCs). In contrast, the retrieved products are reported from the mesoscale model-top height using the U.S. only as clear-sky data. The QCs that are applied to the Standard Atmosphere, 1976 lapse rates to fill gaps in the products depend on the data providers (NESDIS and upper-air levels. Figure 2 depicts the relationship of MSC). Therefore, the data volume of the retrievals is the model-top height and the weighting function of the small, and there are many space and time gaps in the assimilating radiance data in the mesoscale analysis. coverage. Figure 3a compares the O 2 B departure with and without profile extrapolation for all AMSU-A chan- c. Preprocessor for the radiance data in the mesoscale nels. Figure 3b compares the O 2 B departure calcu- data assimilation system lated from the global model with the mesoscale model Preprocessing of radiance assimilation is one of the fields. The extrapolation of the input profiles sub- most important processes in the data assimilation sys- stantially reduced errors in the calculated brightness tem. Data selection (e.g., data thinning and bad data temperature, and the remaining biases of O 2 B de- rejection) is performed in this process. The radiance parture became approximately zero. It is clear that data under the clear-sky conditions are assimilated in the profile extrapolation and the bias correction are the current JMA global NWP system (JMA 2013; necessary for the use of sounding channels from the Okamoto et al. 2005). Cloud screening is based on the AMSU-A in the data assimilation. However, for cer- retrieved cloud liquid content data (Okamoto et al. tain higher-peaking channels (i.e., AMSU-A channel 9 2005; Kazumori 2009b). RTTOV-9 (Saunders 2008) is and above), degradation in the accuracy of the calcu- implemented in the JMA global data assimilation sys- lated radiance remains. The remaining degradation is tem (Kazumori 2009a) as an observation operator to caused by a lack of information in the upper tropo- calculate the radiance and the Jacobians in the varia- sphere and stratosphere from the input profile used in tional data assimilation. The same radiative transfer the radiative transfer calculations. These higher-peaking model (RTTOV-9) is introduced into the preprocessor channels were blacklisted in the mesoscale data assimi- and the analysis system itself as the radiative transfer lation system. However, these channels are used in the model for the radiance calculation in the mesoscale JMA global system. analysis. In the preprocessing step, innovation vectors The other modification is made with respect to the (e.g., observed brightness temperature minus simulated data-thinning distance. To reduce the horizontal corre- brightness temperature, O 2 B departure) are calculation of the observational errors and reduce the com- lated using atmospheric fields and surface properties putation time in the radiance calculation for the analysis, (surface temperature and wind) from the most recently data thinning is widely used in satellite radiance assim- produced mesoscale forecast fields using the high-resolution ilation. Because the horizontal resolution between the outer model in the mesoscale NWP system. global (approximately 20 km) and the mesoscale models Because the same radiative transfer model that is used (5 km) differs, the meteorological phenomena repre- in the JMA global system was selected for the pre- sented in the numerical model are also thought to be processor and for the 4DVAR analysis in the mesoscale different. The horizontal data-thinning distance in the system, QCs based on the models used for the radiance mesoscale system was set to 45 km to reflect its higher data in the JMA global analysis (JMA 2007; Okamoto resolution (5 km for the outer model and 15 km for the et al. 2005) can be applied. However, two modifications inner model). The data-thinning distance for the global were made to the QCs for radiance assimilation in the data assimilation system was set to 160–250 km, de- mesoscale system. pending on the sensor type. Because the horizontal The first change is in the treatment of the atmospheric resolution of the microwave radiometers varies between profiles for the radiative transfer calculation. The model- sensors and MTSAT CSR data have approximately 60-km top height is different between the global model and the horizontal resolution, the data-thinning distances were mesoscale model. The global model-top height is ap- empirically determined. MTSAT CSR data were not proximately 0.1 hPa, and the atmospheric profiles from thinned. Using denser data with short distance thinning the global model have 60 vertical levels to cover the requires substantial computational time in the pre- atmosphere from the surface to the stratopause, whereas processor and the analysis. Moreover, the observational

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1366 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 2. Weighting functions of (a) AMSU-A, (b) MHS, (c) AMSR-E, and (d) MTSAT. Only assimilated channels in the JMA NWP system are shown, except for AMSU-A channel 14. Solid curves indicate weighting functions used in both the global DA system and the mesoscale DA system. Broken lines in (a) indicate weighting functions used only in the global DA system. The straight lines in (a) and (d) indicate the mesoscale model-top height. information might be assimilated redundantly because radar data (Koizumi et al. 2005) for rainy areas, the ra- adjacent ﬁelds of view from the microwave sensors diance data from AMSU-A, MHS, and MTSAT are overlap in some areas. Identical observation errors were assimilated only under clear-sky conditions. assigned with the global analysis (Okamoto et al. 2005) Therefore, it is necessary to employ cloud screening to for the mesoscale analysis. The data assimilation system avoid the contamination of cloud- and rain-affected ra- may not use an optimal observation error setting. We diance data in the assimilation system. Methods identi- empirically reduced the data-thinning distance to 45 km. cal to those used in the global system were adopted for The reduced thinning distance did not show any nega- the mesoscale system. Cloud liquid water is derived tive effects. However, one of the important roles of the from observed radiance data. A retrieval algorithm data thinning is to avoid difﬁculty in explicitly treating (Kazumori et al. 2012) was applied to microwave images observation error correlations for time and space in to obtain cloud liquid water. For AMSU-A instruments, a data assimilation system. Future investigations re- the NESDIS cloud liquid water algorithm (Weng, et al. garding the data-thinning distance and the correlated 2002) was utilized. Based on the amount of cloud liquid observation errors when using high-frequency and high- water, cloud screenings were performed. The threshold 2 density radiance data (e.g., next generation’s geosta- of the cloud liquid water was set at 200 g m 2. In our tionary satellites) are necessary. previous paper [see Table 5 in Kazumori et al. (2008)], 2 a threshold of 100 g m 2 was used for the 37-GHz d. Cloud screening channel of the microwave imager (AMSR-E) for cloud Although the retrieved rain rate from microwave im- screening. Although this stringent threshold can select agers (Takeuchi and Kurino 1997; Sato et al. 2004) is as- purely clear conditions from the data, we slightly relaxed 2 similated with the rain rate derived from the ground-based the threshold from 100 to 200 g m 2 to include thin

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1367

FIG. 3. (a) Comparison of the AMSU-A brightness temperature O 2 B departure histogram between the proﬁle with extrapolation (solid line) and without extrapolation (dashed line) for 18 Sep 2009. (b) Comparison of the AMSU-A brightness temperature O 2 B departure histogram between the mesoscale model with proﬁle extrapolation (solid line) and the JMA global model (dashed line) for October 2011. The histograms indicate the frequency distribution of NOAA-18 AMSU-A O 2 B departure after QC and bias correction.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1368 MONTHLY WEATHER REVIEW VOLUME 142 cloudy data. The change in the threshold was acceptable The biases are estimated with a linear function using because the retrieved cloud liquid water is used as one of several predictors; the coefficients are updated to cap- the predictors in VarBC. The clear-sky microwave ra- ture seasonal and daily changes in the biases. However, diances are assimilated in the mesoscale data assimila- because of nonuniform coverage of the radiance data in tion system; the remaining cloud contamination effect is the limited areas of the model, the estimation of the adjusted in the bias correction. For the MHS, a scatter- radiance bias inside the mesoscale system is more diffi- ing index of 89/150 GHz (Bennartz et al. 2002) and cult to ascertain than in the global system. The coverage a gross error check of the O 2 B departure are used to and available data depend on the satellite orbits, and the remove the cloud and scattering effects caused by hy- data delivery conditions of the GTS and the Internet drometeors in the atmosphere. The CSR data from the vary in every analysis. Therefore, the estimated radiance MTSAT-1R and -2 produced in the JMA MSC are as- biases in the mesoscale analysis may be highly situation similated into the mesoscale system. We confirmed the dependent, based on the meteorological conditions and/ validity of our cloud detection algorithm using O 2 B or polar-orbiting satellite data coverage. This variability departure statistics. The distribution of the O 2 B de- may cause an overfitting in the bias correction, which partures is expected to follow a Gaussian distribution may cause the estimated bias correction coefficients to when cloud-affected data are rejected and biases are be unstable in a time series. Therefore, we use bias properly corrected. We confirmed the Gaussian distri- correction coefficients determined from the latest global bution for the O 2 B departures after QC and bias analysis in the JMA NWP system. The coefficients correction. produced with the global data are much more stable and It should be noted that, as discussed above, AMSU-A robust. The coefficients are updated every 6 h in the channels 4–8 are selected for the assimilation. However, global data assimilation cycle. The latest updated co- the surface-sensitive data are removed because the ac- efficients estimated in the global analysis are utilized for curacy of the simulated radiances is inadequate for as- the mesoscale system, and the same set of predictors in similation because of the insufficient accuracy of the the global analysis is also used in the mesoscale system. input surface temperature and land emissivity for the With respect to the AMSU-A, the integrated lapse rate, radiative transfer calculation. The effects of AMSU-A surface temperature, cloud liquid water, and satellite channels 1–3 and 15 in the JMA global analysis system zenith angle are used. The MHS uses the same pre- have not been fully investigated nor assimilated. dictors as the AMSU-A except that it does not use cloud Therefore, updates to the bias coefficients in the oper- liquid water. With respect to microwave imagers, total ational system are not available for those channels. precipitable water, surface temperature, surface wind These selection criteria are identical to those of the speed, and cloud liquid water are used. Finally, the sim- global analysis (Okamoto et al. 2005). With respect to ulated brightness temperature and sensor zenith angle microwave imagers (AMSR-E, SSMIS, and TMI), only are used for the MTSAT CSR. vertically polarized channels (19, 23, 37, and 89 GHz) The same bias correction coefficients can be used be- over the ocean were assimilated to obtain column water cause we implicitly assume that the difference between vapor information. To avoid the uncertainty of land the radiance biases in the global model and the mesoscale surface emissivity in the radiative transfer calculation, model in terms of the O 2 B departures are not large, and an infrared moisture channel (6.7 mm) from MTSAT the same radiative transfer model RTTOV-9 is used in CSR was selected for assimilation because the channel is the JMA global and mesoscale data assimilations. The less sensitive to the surface. For MHS channels 3–5, bias-corrected O 2 B departures shown in Fig. 3b suggest a water vapor absorption line at approximately 183 GHz that the use of the coefficients estimated in the global was also assimilated over the ocean. The weighting system for the selected channels is acceptable for the functions of the channels used in the mesoscale system mesoscale system. are shown in Fig. 2. e. Bias correction 3. Experiments with the JMA mesoscale 4DVAR system Radiance biases should be properly corrected before data assimilation because the variational data assimila- a. Experimental configuration tion method assumes no bias in either the observations or the forecast models. A variational bias correction Before the operational implementation of the radi- scheme is commonly used in the global data assimilation ance assimilation scheme, two assimilation experiments system (Dee 2005). In the JMA global analysis, VarBC were configured to examine the effects of radiance assim- schemes are used to correct radiance data bias (Sato 2006). ilation in the mesoscale system. The first run (Control run)

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1369

TABLE 1. The physical quantity related to the radiance data for each satellite/sensor that was used.

Satellite/sensor Control run Test run Microwave imager Aqua/AMSR-E Total precipitable water, rain rate Radiance, rain rate DMSP F-16/SSMIS N/A Radiance, rain rate DMSP F-17/SSM/I N/A Radiance, rain rate TRMM/TMI Total precipitable water, rain rate Radiance, rain rate DMSP F-16 SSM/I N/A Radiance NOAA-15/AMSU-A Temperature Radiance NOAA-16/AMSU-A N/A Radiance Microwave temperature sounder NOAA-18/AMSU-A Temperature Radiance MetOp/AMSU-A Temperature Radiance Aqua/AMSU-A N/A Radiance NOAA-15/AMSU-B N/A Radiance NOAA-16/AMSU-B N/A Radiance NOAA-17/AMSU-B N/A Radiance Microwave humidity sounder NOAA-18/MHS N/A Radiance NOAA-19/MHS N/A Radiance MetOp/MHS N/A Radiance NOAA-17/HIRS Temperature N/A IR sounder Aqua/AIRS N/A N/A MetOp/IASI N/A N/A IR imager MTSAT-1R/Imager N/A Radiance

N/A 5 Not available in the experiment. was set up identically to the operational configuration assimilation on typical meteorological phenomena and of the JMA mesoscale NWP system as of October 2010. hazardous weather events caused by heavy precipitation The Control run did not directly use any radiance data. in Japan. The periods from 16 to 26 July 2009 and from 1 The Control run assimilated conventional observa- to 5 July 2010 include heavy rainfall events caused by tions, including surface pressure data from meteoro- seasonal fronts, and the period from 9 to 12 August 2010 logical observation stations and ships; temperature, includes a typhoon. wind, and relative humidity data from radiosondes; and The data coverage of the temperature retrievals and aircraft; wind and rain-rate data from ground-based the raw radiances varies for different analysis times, as radars; satellite wind data (MTSAT atmospheric mo- shown in Fig. 1. The data from polar-orbiting satellites tion vector); and retrieved data [e.g., temperature vary temporally, and in most cases, the data are avail- profiles from ATOVS and total precipitable water va- able for two analysis times per day and cover some por (TPW) and rain rate from microwave imagers]. The fraction of the mesoscale model domain. The large data second run (Test run) assimilated radiances from volume is an advantage of using radiances in the anal- ATOVS (AMSU-A, MHS), AMSR-E, TMI, and SSM/ ysis. It should be noted that two runs were not clean I(SSMIS) on board polar-orbiting satellites and geosta- comparisons between the radiance assimilation and the tionary satellites (MTSAT CSR), in addition to the con- retrieval assimilation because the volume of data used ventional observational data used in the Control run. was different. However, this type of comparison is im- However, the retrievals (temperature profiles and TPW) portant from an operational standpoint. The configurations were removed in the Test run to avoid information from of the dataset used in both runs are summarized in Table 1. the same data source being used twice. Rain-rate retrievals b. Results from microwave imagers were used in both the Control and Test runs. DMSP F-16 and F-17 SSMIS microwave The accuracy of the analysis and the forecast were imager radiances and rain rates were newly added in the considerably improved with the radiance assimilation. Test run. The data used in both runs were delivered within Figure 4 shows the profiles of the RMSEs of the geo- the operational data receiving cutoff time. potential height forecast against radiosonde observa- We performed 3-hourly assimilation cycles for three tions. To show the difference in RMSEs between the selected periods and ran the mesoscale model to pro- Control run and the Test run, an improvement ratio is duce 33-h forecasts from 0300, 0900, 1500, and 2100 UTC defined as the relative value of the improvements for initialization fields every day. The three periods were each pressure level. The ratio is calculated with the selected to examine the performance of the radiance following equation:

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1370 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 4. RMSE of the geopotential height forecast for the mesoscale model against the radiosonde observations from 16 to 26 Jul 2009. The forecasts from the 0300, 0900, 1500, and 2100 UTC initializations were verified. The verified times were 0000 and 1200 UTC. The line type indicates the difference in forecast time (FT): solid, FT 5 3; dashed, FT 5 9; dotted, FT 5 15; fine dotted, FT 5 21; dotted–long dashed, FT 5 27; and dotted–short dashed line, FT 5 33. Shown are the (a) Control run, (b) the Test run, and (c) the improvement ratio of RMSE (%; see text for the definition). Plotted numbers on the right-hand side in (c) are the observation counts used in the verification. Statistical significances at 90% level using a t test are displayed with filled circles in (c).

RMSE of the Control run 2 RMSE of the Test run Improvement ratio (%) 5 3 100. RMSE of the Control run

In Fig. 4c, positive values indicate reductions in analyses within the JMA operational system’s job RMSEs in the Test run. In the Control run (Fig. 4a), the schedule framework. It is likely that the global fore- RMSE of the forecast time (FT) 5 3 (solid line) was casts from 0000 and 1200 UTC have better quality in larger than that of FT 5 9 (dashed line), whereas the the mesoscale model domain than others because the RMSE increases with the forecast time in the Test run radiosonde data around Japan are available for these (Fig. 4b). The veriﬁed forecasts were produced from 0900 global analysis times. In the Test run, the improvement and 2100 UTC analyses at FT 5 3andFT5 15 in the of the RMSE in the short-range (FT 5 3andFT5 15) forecast veriﬁcation, respectively. In these analyses, ra- forecasts in the lower troposphere from 1000 to 500 hPa diosonde data were not available in the assimilation time was remarkable. The results from the Test run are window. However, radiosonde data were available in the normal for the growth of NWP model errors. The assimilation time windows for other analyses (e.g., 0300 reason for the existing degradation in the short-range and 1500 UTC). Typically, radiosonde data are generated forecast accuracy in the Control run is thought to be at 0000 and 1200 UTC. The lateral boundary conditions related to unbalanced data coverage and the amount of (JMA global model forecast) produced from 0600 and conventional data in the 3-h cycle of the mesoscale 1800 UTC initializations are available for the 0900 and analysis.Theuseofsatelliteradiancedatainplaceof 2100 UTC analyses, respectively. Moreover, the global the retrieval data compensated for the lack of obser- model forecasts from 0000 and 1200 UTC initializations vation information in the assimilation cycle. The effects are available for the 0300 and 1500 UTC mesoscale on the upper troposphere and lower stratosphere were

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1371 small. One possible explanation for this lessened effect is In a heavy rain event caused by a seasonal front, that higher peaking radiance data were not used in the a significant difference in the analyzed TPW field was mesoscale system. Another possible reason is that the found. Figure 7 compares the analyzed TPW fields of the upper layers (near 40 hPa) in the mesoscale model are Test run and the Control run at 2100 UTC 2 July 2010. used as relaxation layers for the lateral boundaries, and The stationary front began over China and moved to the the analysis increments were suppressed vertically to western part of the Sea of Japan. In the Test run, the balance the lateral boundary condition (from the global analyzed TPW in the East China Sea was larger (ap- model) with the mesoscale model. proximately 5–10 mm) than that of the Control run. The As for moisture-related observations, TPW retrievals moisture flow from the southwest around Kyushu was from AMSR-E and TMI were assimilated in the Con- strengthened in the Test run; the TPW exceeded 70 mm. trol run, whereas AMSR-E and TMI radiance data Figure 8 shows 3-h accumulated rainfall forecasts for were directly assimilated in the Test run. The micro- 0300 UTC 3 July 2010 from various initialization times. wave imagers are thought to provide similar moisture The top panels in Fig. 8 indicate the rainfall forecasts information because most of the moisture is typically of the Control run and the middle panels indicate those concentrated in the lower part of the atmosphere, and of the Test run. The initial times are 2100, 1500, and the information from microwave imagers is considered 0900 UTC 2 July 2010, from the left side, corresponding in the total column amount (not the sounding in- to 6-, 12- and 18-h forecasts, respectively. The bottom formation). DMSP F-16 and F-17 SSMIS-retrieved panel in Fig. 8 indicates the rainfall distribution from TPW products were not used in the Control run be- radar observations, which are calibrated with rain cause the corresponding retrievals were not ready at gauges. A notable change was found in the 6-h rainfall the time of the experiment. SSMIS, MHS, and MTSAT forecast from the initial time of 2100 UTC 2 July 2010 (infrared water vapor channel) radiance data were (left panels in the top and middle). Strong rainfall 2 newly added in the Test run. These datasets provide [50 mm (3 h) 1] was predicted near the south of Kyusyu new lower- and upper-tropospheric moisture observa- and Tanegashima Island in the Test run, whereas only tion information. In the analysis, these data largely weak rainfall was predicted for those areas in the Con- increased the moisture content and strengthened the trol run. The predicted rainfall differences for other TPW contrast between moist and dry areas in a ty- forecast periods were less than the 6-h rainfall forecast phoon event, as shown in Fig. 5. In this case, the sub- difference. The improvement of the rainfall forecast was tropical high pressure system was located east of Japan outstanding for short-term forecasting, suggesting that (see Fig. 5e). The high pressure system was character- the TPW difference and the southwest moisture flow ized by 30–40 mm in the TPW range (green areas in over the ocean, shown in Fig. 7, enabled the rainfall Figs. 5a,c), whereas moisture flow along the high forecast to be more realistic. pressure boundary was described by 50–70 mm in the Figure 9 shows the bias scores and threat scores of the TPW range. The TPW analysis increment along the 3-h cumulative rainfall forecasts for the Control and coast of Japan was similar in the Test and Control runs. Test runs from three experiment periods. Each panel However, over the ocean, the Test run TPW analysis shows the scores at different thresholds within a 10-km increment was much larger and wider than that of the grid box. Generally, the results show that the Test run Control run. Because the ground-based GPS TPW and outperforms the Control run for both scores. the rain rate from ground-based radar were used in both the Test and Control runs, the runs produced 4. Data-denial experiment for Typhoon Roke a similar increment around Japan. The addition of the in 2011 new F-16 and F-17 SSMIS microwave imagers increased the increment over the ocean. Moreover, the a. Experimental configuration moisture increment became seamless and consistent from the land to the open ocean (see the dotted ellipses To investigate the individual radiance data’s contri- in Figs. 5b,d). As a result, the TPW contrast along the bution to the mesoscale analyses and forecasts, several boundary of the high pressure system was enhanced in data-denial experiments were performed. Typhoon Roke the Test run. in 2011 was selected for this case study. As shown in Figure 6 compares the MTSAT-observed IR image Fig. 10, Roke became a tropical storm near the south and the simulated image from the mesoscale analysis boundary of the mesoscale model domain at 0600 UTC 13 fields at 0900 UTC 9 August 2010. The separated cloud September 2011 and proceeded westward, approaching feature located south of Japan (see red arrow in Fig. 6c) Daito Islands on 16 September. The typhoon moved was well represented in the Test run. slowly in a counterclockwise direction, making a circular

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1372 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 5. Comparison of the analyzed TPW ﬁeld and the TPW analysis increment at 0000 UTC 9 Aug 2010, in the Test and Control runs. Shown on the left are the analyzed TPWs for the (a) Control and (c) Test runs. Shown on the right are the analysis increments for the (b) Control and (d) Test runs. The units are mm. (e) The surface weather chart at 0000 UTC 9 Aug 2010. Dotted lines in (b) and (d) highlight the areas where the difference in the TPW increment is large. The relaxation area for the lateral boundary (see the text) is included at the edge of the displayed area.

track west of Minami-Daito Island. The typhoon moved typhoon made landfall in Shizuoka at 0500 UTC 21 to the north toward the southeast of the Amami Islands September and moved to the northeast. After the land- and strengthened to a severe tropical storm with fall, the typhoon moved over the ocean to the southeast 2 a35ms 1 maximum wind speed at 1200 UTC 19 Sep- of Hokkaido and weakened around the Chishima Islands 2 tember. The wind increased to a maximum of 50 m s 1, on 22 September. The typhoon came into the mesoscale and the minimum central pressure was 940 hPa. The model domain from the south in the early stages of its

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1373

FIG. 6. Comparison of MTSAT-observed IR image (IR1, 10.3–11.3 mm) and the simulated image from the mesoscale analysis ﬁeld at 0900 UTC 9 Aug 2010. The simulated IR imagery from the (a) Test and (b) Control runs. (c) The observed IR image. The relaxation area for the lateral boundary (see the text) is included at the edge of the displayed area. The cloud discussed in the text is indicated with a red arrow in (c).

development. Because the available conventional and data and the retrieval of rain rate from the microwave ground-based observation data were sparse in southern imager were removed. Except for the difference in the Japan compared to the coastal areas of Japan, this case radiance data usage, the other components of the data is expected to clearly demonstrate the effects of the assimilation system and forecast model were identical in satellite radiance assimilation. all the experiments. The period of the data-denial experiment was set from b. Verification of analyzed fields against radiosonde 13 to 22 September 2011. The analyses were performed data every 3 h during the period and 33-h forecasts were made beginning at 0300, 0900, 1500, and 2100 UTC every The majority of the significant differences in the an- day. Table 2 summarizes the experimental configura- alyzed fields were found in the lower troposphere below tion. Here, the crisscross symbol (3) signifies that the 500 hPa. Figure 11 shows the RMSE of the 3-h forecast data were present. The control run has the same con- of the geopotential height against the radiosonde. The figuration as the JMA operational system as of June Control run RMSE was the best and that of experiment 2011. All operational observation data, including con- 4 (with no radiance data) was the worst. Additionally, ventional data and the radiance data, were assimilated. the degradation in experiment 3 (no MHS and no CSR) In experiment 1, the microwave imager (AMSR-E, F-16 was relatively small compared to the other cases. The and F-17 SSMIS, and TMI) data (radiance and retrieved large degradation of RMSEs for the geopotential height rain rate) were not assimilated. In experiment 2, the in the lower troposphere in experiments 1 and 2 in- microwave temperature sounder (AMSU-A) radiance dicates that the temperature and humidity information data were not assimilated. In experiment 3, the MHS from AMSU-A and the microwave imagers sub- radiance and the MTSAT CSR (water vapor channels) stantially contributed to the accuracy of the analysis and data were not assimilated. In experiment 4, all radiance the short-range forecasts.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1374 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 7. Comparison of the analyzed TPW ﬁeld for the (a) Test and (b) Control runs, and (c) their difference (Test 2 Control) at 2100 UTC 2 Jul 2010. The units are in mm. (d) Surface weather chart at 0000 UTC 3 Jul 2010. The black solid lines indicate the locations of the stationary front. The arrow in (d) shows the location of Kyusyu Island. The relaxation area for the lateral boundary (see the text) is included at the edge of the displayed area in (a)–(c). c. Impact on precipitation forecast TRMM satellite (TMI L2A12 version 7 product, available online from the Goddard Earth Sciences Data and In the 4DVAR analysis, precipitation is calculated in Information Services Center website: http://disc.sci.gsfc. the ﬁnal outer loop step with the high-resolution meso- nasa.gov/). The broad weak precipitation area observed in scale model (the horizontal resolution is 5 km). Figure 12 experiment 4 (Fig. 12d) did not exist in the rainbands in the shows comparisons of the distributions of the 3-h accu- TMI-observed rain distribution. This suggests the Control mulated precipitation at the analysis time (2100 UTC 15 run precipitation forecast (Fig. 12e) is similar to reality. September 2011). Strong rain areas were located over Moreover, the radiance assimilation can improve weak land, near the coastline of Kyushu Island, and around precipitation forecasts. the core of the typhoon. Additionally, the typhoon ex- The statistical precipitation forecast scores also support hibited several outer bands of rain around its center. the advantages of the radiance assimilation. Figure 13 Generally, the removal of the radiance data broadened presents the bias scores and threat scores for each of the 2 the weak rain [1–5 mm (3 h) 1] area around the typhoon runs against the analyzed radar precipitation. Figures 13a compared with the Control run, whereas the differences and 13b show the bias scores and the threat scores, re- 2 in the strong rain distribution at the core of the typhoon spectively, with the 5 mm (3 h) 1 threshold. Figures 13c 2 over the land and coastal areas were small. Figure 12g and 13d present identical scores but for the 40 mm (3 h) 1 shows the observed hourly rainfall intensity from the threshold. The results shown in Figs. 13a,c clearly show

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1375

FIG. 8. Comparison of the 3-h accumulated rainfall forecast for 0300 UTC 3 Jul 2010, from different initialization times. The rainfall forecasts of the (top) Control and (middle) Test runs. The initial times are 2100, 1500, and 0900 UTC 2 Jul 2010, from the left side, corresponding to 6-, 12-, and 18-h forecasts, respectively. (bottom) The rainfall distribution estimated from radar observations and rain gauges. The units are mm. The locations of Kyusyu and Tanegashima Islands are shown in the top-left panel.

that the mesoscale model has spinup issues in the pre- (no radiance) and the Control was small. Moreover, cipitation forecast for very short-range forecasts (0–18 h) differences between experiment 1–4 were also small in both threshold conditions. The radiance assimilation (not shown). Use of the radiance data had less effect on improves the precipitation spinup issues (i.e., increases the the typhoon track and intensity forecasts. Because the 2 precipitation) with weak precipitation [0–15 mm (3 h) 1] number of samples was not sufficient for statistical sig- in the short-range forecast. For the heavy rain cases (Figs. nificance, it is difficult to assign a general trend to the ef- 13c,d), the effects of the radiance assimilation were small. fect of the radiance assimilation on the typhoon forecast. The scores for Figs. 13a,b also suggest that the effect of the radiance assimilation persists for at least 24 h (approxi- 5. Discussion and future prospects mately) into the precipitation forecasts. The typhoon track and intensity forecasts were eval- In this study, a radiance assimilation scheme for uated (Fig. 14). The difference between experiment 4 a regional model was developed, and the impact of

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1376 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 9. (a) Bias scores and (b) threat scores of 3-h cumulative rainfall from the Test (solid line) and the Control (dotted line) runs from 16 to 26 Jul 2009. (c),(d) As in (a),(b), respectively, but from 1 to 5 Jul 2010. (e),(f) As in 2 (a),(b), respectively, but from 9 to 12 Aug 2010. The horizontal axis indicates the threshold [mm (3 h) 1] in the score calculations. The grid size of the score calculation is 10 km. satellite-radiance assimilation was investigated with the the radiative transfer model top. Moreover, the bias JMA mesoscale 4DVAR system. For the radiance as- coefficients estimated in the global system were adopted. similation in the regional model, a bias correction It is confirmed that the profile extrapolation is neces- method has not been established because of the diffi- sary to reduce the radiance biases from the comparison culties in estimating the radiance bias in the limited re- of O 2 B departures, allowing us to effectively use the gions and times. To overcome these difficulties, the radiance data in the mesoscale system in the same way radiance assimilation scheme used in the global system as in the global system. This result suggests that most of was applied in the mesoscale system. The data-thinning the radiance biases are related to satellite sensors and distance was shortened to better represent the high- the radiative transfer model origin. The dominant frac- resolution mesoscale model, and the atmospheric pro- tion of the O 2 B departure biases come from the satellite files were extrapolated from the mesoscale model top to instrument and the error in the radiative transfer model

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1377

FIG. 10. (a) The ‘‘best track’’ of Typhoon Roke provided by the Regional Specialized Meteorological Center (RSMC) Tokyo–Typhoon Center. The JMA mesoscale model domain is shown in gray. Roke’s center is plotted every 6 h with the following symbols: open circles, tropical depression; filled circles, tropical storm; squares, severe tropical storm; filled triangles, typhoon; and open triangles, extratropical cyclone. (b) Enlarged view of the dotted rectangle area in (a). The dates and the location names are shown. origin. The effect of using a different first-guess profile mid- and lower-tropospheric geopotential height fore- (the mesoscale model or the global model) in the ra- casts versus the retrieval assimilation. A reduction in the diative transfer calculation is relatively small because RMSE of the geopotential height forecast was con- the bias-corrected O 2 B departure in Fig. 3 shows less firmed with radiosonde observations. The reason for the bias. Therefore, the coefficients estimated in the global large forecast improvements in FT 5 3 and FT 5 15 is system represent a general trend in the radiance biases, believed to be related to the introduction of the radiance and their use in the mesoscale system is acceptable in data into the data-sparse analysis time. The use of ra- our initial implementation of the radiance assimila- diance data and the addition of the radiance data from tion scheme. However, the assumption that the pro- the new satellite filled the gap and complemented the files from the two models can be treated equally might lack of the observation information from the conven- not be valid for some situations. This is a topic of future tional observation network. AMSU-A radiance data research. Updating the coefficients using the past O 2 B assimilation and the use of radiance over land, where departure statistics inside the mesoscale analysis is one idea for the estimation of the biases in the mesoscale analysis system (daily or weekly updates to the coefficient). In the assimilation experiments, the direct use of the radiance data and the addition of new satellite radiance data demonstrated considerable positive effects on the

TABLE 2. The radiance data used in the data-denial experiments.

Microwave Microwave temperature Microwave imager data sounder humidity (radiance and (AMSU-A) sounder (MHS) Expt rain rate) radiance and CSR from MTSAT Control 33 3 1 33 2 33 3 33 FIG. 11. The RMSE of the geopotential height for the 3-h fore- 4 casts against the radiosonde data. The vertical axis is the pressure level (hPa) and the horizontal axis shows the RMSE (m). The 35data present. statistical period was from 13 to 22 Sep 2011.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1378 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 12. Analyzed 3-h accumulated rainfall in mm at 2100 UTC 15 Sep 2011 for (a) experiment 1, (b) experiment 2, (c) experiment 3, (d) experiment 4, (e) the Control run, and (f) radar observations. The units are 2 [mm (3 h) 1]. (f) TMI precipitation observations at 2000 UTC 15 Sep 2011 are also 2 shown. The units are mm h 1.

only channels sensitive to higher-level atmospheric of the radiances from various microwave imagers, hu- temperature are assimilated, are believed to have been midity sounders, and MTSAT CSR produced realistic the primary contributors to this improvement. Similar tropospheric moisture ﬁelds. unbalanced observation effects that depend on analysis The simulated MTSAT infrared image from the time were reported in Zapotocny et al. (2005). Assimilation analysis in the radiance assimilation showed similar

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1379

21 FIG. 13. Precipitation forecast score against analyzed radar precipitation. (a) Bias score with a 5 mm (3 h) 2 2 threshold, (b) threat score with a 5 mm (3 h) 1 threshold, (c) bias score with a 40 mm (3 h) 1 threshold, and (d) the 2 threat score with a 40 mm (3 h) 1 threshold. The grid size of the score calculation is 10 km. features to the real MTSAT image. This result suggests stage because much of the satellite radiance data were that the analyzed tropospheric moisture field and the assimilated into the analysis. The verification of the hydrometeors produced in the 4DVAR analysis in the Test geopotential height forecast with the radiosonde data run became more realistic than they were in the Con- implies that much of the satellite dataset is necessary to trol run. maintain the accuracy of the analysis and forecasts in the Together, these findings demonstrated that the satel- operational NWP system. lite data contribution to the analyses and forecast ac- Based on these findings, the radiance assimilation was curacy in the JMA mesoscale analysis was greater than introduced in the JMA mesoscale 4DVAR system on the contribution of the previous retrieval assimilation 13 December 2010. The radiance assimilation allowed

FIG. 14. The veriﬁcation of the typhoon track and intensity forecasts. (a) Positional error and (b) intensity error against the best track of Typhoon Roke provided by the RSMC Tokyo–Typhoon Center.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC 1380 MONTHLY WEATHER REVIEW VOLUME 142 for the early introduction of radiance data into the op- over land are expected to bring further improvements in erational system. After the implementation of the ra- precipitation forecasts. Finally, Kawabata et al. (2011) diance assimilation scheme in the JMA mesoscale reported that high-frequency ground-based radar re- system, DMSP F-18 SSMIS radiance data were in- flectivity assimilation can reproduce a heavy rain event. corporated into the system on 6 December 2011. The assimilation of high-frequency and high-density Furthermore, to clarify the contribution of the radi- radiance data from future geostationary satellites will be ance data (or sensor type) to the analysis and forecast, our next research topic. Moreover, issues related to the several data-denial experiments were performed in as- optimal data-thinning distance and the correlated ob- sociation with Typhoon Roke. One possible reason for servation errors must be investigated. the minor impact on the typhoon track forecasts is that the typhoon track forecast was strongly affected by Acknowledgments. The author would like to thank large-scale meteorological conditions (i.e., the location colleagues in the JMA numerical prediction division and of a high pressure system and upper-level jet). Most of the Meteorological Research Institute for helpful re- these conditions are determined by the lateral boundary marks and comments during the preparation of the conditions (Warner et al. 1997). manuscript. Moreover, four anonymous reviewers pro- Improving the analyzed fields with radiance assimi- vided detailed comments that notably improved the lation in the regional model domain may not directly quality of the paper. This work was supported by Global contribute to improving typhoon track forecasting for Change Observation Mission-Water 1 (GCOM-W1) re- short-range forecasting. In a recent study on microwave search with the Japan Aerospace Exploration Agency radiance assimilation using a limited-area ensemble data (JAXA) Earth Observation Research Center (EORC). assimilation system (Schwartz et al. 2012), improvements in precipitation forecasts caused by a tropical REFERENCES cyclone were reported. However, no significant positive impact on the track forecast was found. Andersson, E., J. Pailleux, J.-N. Thepaut, J. R. Eyre, A. P. Although the radiance assimilation brought im- McNally, G. A. Kelly, and P. Courtier, 1994: Use of cloud- cleared radiances in three/four-dimensional variational data provements in the forecast of the geopotential height assimilation. Quart. J. Roy. Meteor. Soc., 120, 627–653. and precipitation over the ocean, the major effects on Bennartz, R., A. Thoss, A. Dybbroe, and D. B. Michelson, 2002: precipitation forecasts were limited to the ocean and in Precipitation analysis using the Advanced Microwave Sounding light precipitation areas. Because the rain-rate data Unit in support of nowcasting applications. Meteor. Appl., 9, from ground-based radars were assimilated in the ex- 177–189. Courtier, P., J.-N. Thepaut, and A. Hollingsworth, 1994: A strategy periments, the precipitation over the land exhibited for operational implementation of 4D-Var, using an incremental similar features. The impact of the radiance assimilation approach. Quart. J. Roy. Meteor. Soc., 120, 1367–1387. was limited over the ocean because the surface-sensitive Dee, D. P., 2005: Bias and data assimilation. Quart. J. Roy. Meteor. radiances over the land and the rain-affected radiances Soc., 131, 3323–3343. have not yet been assimilated into the current JMA Derber, J. C., and W.-S. Wu, 1998: The use of TOVS cloud-cleared radiances in the NCEP SSI analysis system. Mon. Wea. Rev., systems. Cloud- and rain-affected radiance assimilation 126, 2287–2299. schemes for the JMA system are under development Eyre, J. R., 1992: A bias correction scheme for simulated TOVS (Kazumori et al. 2010). In the current JMA system, the brightness temperatures. ECMWF Tech. Memo. 186, 33 pp. radiance data strongly constrain the temperature and Harris, B. A., and G. A. Kelly, 2001: A satellite radiance bias moisture fields only in clear and nonprecipitation areas; correction scheme for data assimilation. Quart. J. Roy. Meteor. Soc., 127, 1453–1468. the contribution in cloudy areas, in rainy areas, and over Honda, Y., and K. Sawada, 2009: Upgrade of the Operational Me- land is relatively weak in the analysis. soscale 4D-Var System at the Japan Meteorological Agency. Zou et al. (2011) also reported that the assimilation of WGNE Blue Book: Research Activities in Atmospheric and Ocean the Geostationary Operational Environmental Satellites- Modelling, WMO/Working Group on Numerical Experimenta- 11 and -12 (GOES-11/12) radiances under preconvective tion, 1.11–1.12. [Available online at http://www.wcrp-climate.org/ WGNE/BlueBook/2009/documents/sections.html.] conditions improved quantitative precipitation forecasts ——, M. Nishijima, K. Koizumi, Y. Ohta, K. Tamiya, T. Kawabata, in a 3DVAR system with a limited-area model. The and T. Tsuyuki, 2005: A pre-operational variational data as- radiance assimilation of the clear-sky conditions is similation system for a non-hydrostatic model at the Japan a necessary step toward advanced cloud- and rain- Meteorological Agency: Formulation and preliminary results. 131, affected radiance assimilation in the JMA mesoscale Quart. J. Roy. Meteor. Soc., 3465–3475. Ishikawa, Y., and K. Koizumi, 2002: Meso-scale analysis. Outline system. of the operational numerical weather prediction at the Japan The assimilation of cloud- and rain-affected radiance Meteorological Agency. Appendix to WMO Technical Prog- near the typhoon core and of surface-sensitive channels ress Report on the Global Data-Processing and Forecasting

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC MARCH 2014 K A Z U M O R I 1381

System (GDPFS) and Numerical Weather Prediction (NWP), Okamoto, K., K. Kazumori, and H. Owada, 2005: The assimilation Japan Meteorological Agency, Tokyo, Japan, 26–31. of ATOVS radiances in the JMA global analysis system. JMA, 2007: Outline of the operational numerical weather pre- J. Meteor. Soc. Japan, 83, 201–217. diction at the Japan Meteorological Agency. Appendix to Saito, K., and Coauthors, 2006: The operational JMA non- WMO Technical Progress Report on the Global Data- hydrostatic mesoscale model. Mon. Wea. Rev., 134, 1266– Processing and Forecasting System (GDPFS) and Numerical 1298. Weather Prediction (NWP), Japan Meteorological Agency, Sato, Y., 2006: Introduction of variational bias correction tech- Tokyo, Japan. [Available online at http://www.jma.go.jp/jma/ nique into JMA global data assimilation system. WGNE Blue jma-eng/jma-center/nwp/outline-nwp/index.htm.] Book: Research Activities in Atmospheric and Ocean Model- ——, 2013: Outline of the operational numerical weather pre- ling, WMO/Working Group on Numerical Experimentation, diction at the Japan Meteorological Agency. Appendix 1–19. [Available online at http://www.wcrp-climate.org/ to WMO Technical Progress Report on The Global Data- WGNE/BlueBook/2006/documents/sections.html.] Processing and Forecasting System (GDPFS) and Numerical ——, Y. Takeuchi, and T. Tauchi, 2004: Use of TMI and SSM/I Weather Prediction (NWP) Research, Japan Meteorological data in the JMA operational meso analysis. WGNE Blue Agency, Tokyo, Japan. [Available online at http://www.jma. Book: Research Activities in Atmospheric and Ocean Model- go.jp/jma/jma-eng/jma-center/nwp/outline2013-nwp.] ling, WMO/Working Group on Numerical Experimentation, Kawabata, T., T. Kuroda, H. Seko, and K. Saito, 2011: A cloud- 1.27–1.28. [Available online at http://www.wcrp-climate.org/ resolving 4DVAR assimilation experiment for a local heavy WGNE/BlueBook/2006/documents/sections.html.] rainfall event in the Tokyo metropolitan area. Mon. Wea. Rev., Saunders, R. W., 2008: RTTOV-9 science and validation report. 139, 1911–1931. EUMETSAT, 74 pp. Kazumori, M., 2009a: Impact study of the RTTOV-9 fast radiative Saunders, R., M. Matricardi, and P. Brunel, 1999: An improved fast transfer model in the JMA Global 4D-Var data assimilation radiative transfer model for assimilation of satellite radiance system. WGNE Blue Book: Research Activities in Atmospheric observations. Quart. J. Roy. Meteor. Soc., 125, 1407–1425. and Ocean Modelling, WMO/Working Group on Numerical Schwartz, C. S., Z. Liu, Y. Chen, and Z. Hung, 2012: Impact of Experimentation, 1.21–1.22. [Available online at http://www. assimilating microwave radiances with a limited-area ensemble wcrp-climate.org/WGNE/BlueBook/2009/documents/sections. data assimilation system on forecasts of Typhoon Morakot. html.] Wea. Forecasting, 27, 424–437. ——, 2009b: The impacts of an improved quality control and Takeuchi, Y., and T. Kurino, 1997: Document of algorithm to de- ocean emissivity model for microwave radiance assimila- rive rain rate and precipitation with SSM/I and AMSR: Al- tion in the JMA global 4D-Var data assimilation system. gorithm description of PIs for SSM/I and ADEOS-II/AMSR, WGNE Blue Book: Research Activities in Atmospheric and Second AMSR Workshop, Tokyo, Japan, NASDA, 61.1–61.9. Ocean Modelling, WMO/Working Group on Numerical Uesawa, D., 2009: Clear sky radiance (CSR) product from Experimentation, 1.19–1.20. [Available online at http:// MTSAT-1R. Meteorological Satellite Center Tech. Note 52, www.wcrp-climate.org/WGNE/BlueBook/2009/documents/ 63 pp. sections.html.] Warner, T. T., R. A. Peterson, and R. E. Treadon, 1997: A tutorial ——, Q. Liu, R. Treadon, and J. C. Derber, 2008: Impact study of on lateral boundary conditions as a basic and potentially se- AMSR-E radiances in the NCEP Global Data Assimilation rious limitation to regional numerical weather prediction. System. Mon. Wea. Rev., 136, 541–559. Bull. Amer. Meteor. Soc., 78, 2599–2617. ——, T. Kadowaki, T. Komori, H. Nishihata, and K. Okamoto, Weng, F., 2007: Advances in radiative transfer modeling in support 2010: Operational status and recent developments on cloud of satellite data assimilation. J. Atmos. Sci., 64, 3799–3807. and precipitation assimilation at JMA. Proc. Workshop on ——, L. Zhao, R. R. Ferrao, G. Poe, X. Li, and N. C. Grody, 2002: Assimilating Satellite Observations of Clouds and Precipitation Advanced Microwave Sounding Unit cloud and precipitation into NWP Models, Reading, United Kingdom, ECMWF/ algorithms. Radio Sci., 38, 8068, doi:10.1029/2002RS002679. JCSDA, 16–19. [Available online at http://www.ecmwf.int/ WMO, 2009: Final report of the WMO RARS Implementation newsevents/meetings/workshops/2010/Satellite_observations/ Group and IGDDS Implementation Group joint third meet- presentations/index.html.] ing. 37 pp. [Available online at http://www.wmo.int/pages/ ——, T. Egawa, and K. Yoshimoto, 2012: A retrieval algorithm prog/sat/documents/RARS-IGDDS-IG-3_Final-Report.pdf.] of atmospheric water vapor and cloud liquid water for Zapotocny, T. H., W. P. Menzel, J. A. Jung, and J. P. Nelson III, AMSR-E. European J. Remote Sens., 45, 63–74, doi:10.5721/ 2005: A four-season impact study of rawinsonde, GOES, and EuJRS20124507. POES data in the Eta Data Assimilation System. Part I: The Koizumi, K., Y. Ishikawa, and T. Tsuyuki, 2005: Assimilation of total contribution. Wea. Forecasting, 20, 161–171. precipitation data to JMA mesoscale model with a four- Zou, X., Z. Qin, and F. Weng, 2011: Improved coastal precipitation dimensional variational method and its impact on pre- forecasts with direct assimilation of GOES-11/12 imager ra- cipitation forecasts. SOLA, 1, 45–48. diances. Mon. Wea. Rev., 139, 3711–3729.

Unauthenticated | Downloaded 10/11/21 10:05 AM UTC