1. Introduction
Numerical weather prediction (NWP) technologies can reduce the damage to human lives and social resources caused by heavy rainfalls; their successes have however been confined to heavy rainfalls induced by strong forcings, such as large-scale low-pressure systems, fronts, and orography. Operational NWP systems have a limited capacity to forecast small-scale heavy rainfalls (10–50 km) with weak forcings owing to their coarse resolution, parameterization of cumulus convection, and the limitations of data assimilation systems at that scale. In addition, the stochastic nature of strongly convective systems limits their intrinsic predictability (Zhang et al. 2006).
Several studies have shown that a high-resolution simulation using explicit cloud microphysics rather than parameterized convection is necessary to predict a small-scale heavy rainfall event. Kato and Aranami (2005) showed that a horizontal grid spacing of 1.5 km is fine enough to predict an intense rainband measuring 50 km × 200 km, but a spacing of 5 km is not. Done et al. (2004) noted the importance of explicit cloud microphysics in high-resolution simulations with a horizontal grid spacing of 4 km. Zhang et al. (2006) showed that the error in mesoscale forecasts grows rapidly with time and arises from moist convection, and after comparing horizontal grid spacings of 30 and 3.3 km they suggested that accurate initial conditions contribute to improving the predictability of mesoscale forecasts. Today, the Met Office and the Japan Meteorological Agency (JMA) operate nonhydrostatic models without parameterized convection at horizontal grid spacings of 1.5 (Tang et al. 2013) and 2 km (Hara et al. 2013), respectively. However, computational limitations compel both agencies to use initial conditions supplied by three-dimensional variational data assimilation systems (3D-Var).
For research purposes, most studies have used initial conditions that are downscaled from operational NWPs. Because heavy rainfalls of meso-β–γ scale (10–200 km) have large temporal variation, it is preferable to rely on four-dimensional advanced data assimilation systems with small horizontal grid spacing and explicit cloud microphysics. Furthermore, high-density observation networks are needed to support these techniques.
Mechanisms of meso-β–γ-scale heavy rainfalls remain obscure given the lack of successful NWPs and finescale observation systems. Radar networks are powerful tools for investigating convective systems, but they usually are effective only where water or ice particles are present. Observations of environmental fields thus remain insufficient. As the sizes of meso-β–γ cumulonimbus systems are small (~100 km2), instruments for conventional wind, temperature, and humidity observations are too widely spaced to capture these environmental fields.
Environmental wind fields can be observed using a Doppler radar that observes clear-air echoes caused by reflections from airborne insects, although the technique is limited to relatively calm atmospheric conditions in the lower atmosphere (e.g., Kusunoki 2002). For instance, Kawabata et al. (2007) reproduced the entire life cycle of an isolated cumulonimbus by assimilating clear-air echoes as environmental observations. Rennie et al. (2011) reported that assimilating clear-air echoes in 3D-Var had small but positive impacts for three convective cases.
Another available technique for this purpose is Doppler wind lidar (DWL), which derives information on air motion from the Doppler shift in backscattered signals from aerosols and/or molecules. Most ground-based DWLs rely on aerosols, which limits their observational range. Resolution along the radial direction, as determined by the pulse width, is very high (typically 50–100 m) in comparison to other observational instruments. Whereas airborne and space-based DWLs are powerful tools for observing winds over wide areas (e.g., Ishii et al. 2006; Stoffelen et al. 2005; Baker et al. 2014), ground-based three-dimensional scanning DWLs can provide local observations with high temporal and spatial resolution (e.g., Banta et al. 1996) and have been used operationally to detect significant wind shear (e.g., Shun and Chan 2008). For instance, in 2007 the JMA installed a DWL at the Haneda Airport for monitoring clear-air turbulence that uses laser light at eye safe wavelengths. The principal disadvantages of DWL are that its low signal-to-noise ratio limits its observation range and that strong absorption and multiple scattering by water particles limit observations in rainfall regions. As Doppler radar provides wind observations only inside cumulonimbi, Doppler radar and lidar complement each other to provide comprehensive data (e.g., Bluestein et al. 2014). This study aimed 1) to evaluate the assimilation of DWL data in a forecast of a meso-γ convective system and 2) to gain insight into the convective system’s mechanism.
Assimilating observations from ground-based horizontal scanning and vertical fixed DWLs is similar to assimilating wind profiler observations. Zhang and Pu (2011) conducted assimilation experiments with the Weather Research and Forecasting (WRF) Model (Skamarock et al. 2005) using its four-dimensional variational data assimilation (4D-Var) system (Huang et al. 2009) and a large number of vertical wind profiles provided by a DWL. They reported that assimilating these wind profiles had positive impacts on squall line forecasting.
Airborne DWLs have been used in some field campaigns. For instance, Weissmann et al. (2012) showed that assimilating airborne DWL observations from The Observing System Research and Predictability Experiment (THORPEX) Pacific Asian Regional Campaign project into a global model had positive impacts on a forecast of typhoon and atmospheric conditions. The first space-based DWL is planned to be deployed on the Atmospheric Dynamics Mission Aeolus instrument (ADM-Aeolus) by the European Space Agency (Baker et al. 2014).
Three-dimensional scanning DWLs have so far been used in studies of the atmospheric boundary layer (ABL). Drechsel et al. (2009, 2010) retrieved three-dimensional wind fields using dual-Doppler lidars with the Multiple-Doppler Synthesis and Continuity Adjustment Technique (MUSCAT) method. Newsom and Banta (2004) developed a 4D-Var system with a simple forward model based on an incompressible dry, shallow flow model using the Boussinesq approximation and its adjoint and assimilated DWL observations. As their purpose was to retrieve three-dimensional winds and other meteorological elements in the ABL, the forward model was very simple and had a narrow calculation domain (3 km × 3 km × 0.8 km). Our study is the first to apply assimilation of 3D scanning DWL data to a mesoscale convective system (MCS) with an advanced numerical model and other observations.
We used an isolated convective system on a meso-γ scale as a case study. This system initiated in northern Tokyo, Japan, around 1200 JST (Japan Standard Time, UTC + 9 h) 5 July 2010 and traveled eastward, producing heavy rainfall from 1500 to 2100 JST. The heavy rainfall caused 819 houses to become inundated as small rivers in Tokyo overflowed. A total rainfall amount of 126 mm was recorded at the Itabashi observation site; therefore, this case has become known as the “Itabashi heavy rainfall.” It was characterized by a lack of orographic forcing or a large-scale front (see section 2). Yamada (2012) investigated the cumulonimbi that produced the rainfall using a Doppler radar network and reported that they apparently formed without solid ice particles as the cloud-top height was below the freezing level of 6 km above ground level (AGL). Warm rain clouds produce heavy rainfalls, and they have been studied mainly in tropical regions (e.g., Takahashi 1981). In mid- and high-latitude areas (north of 35°N), warm rain clouds usually produce weak rain, and heavy rains are produced by clouds containing water and ice particles. Therefore, it is worthwhile to study the mechanisms of this warm midlatitude heavy rainfall event in detail.
Yamada (2012) mapped the airflow field of the Itabashi rainfall using radar reflectivity, but did not report the related environmental fields (e.g., wind field, water vapor, and temperature). Because of the small horizontal scale of the Itabashi rainfall event, a detailed investigation of its mechanism requires a high-resolution simulation that is initialized with a high-resolution advanced data assimilation system using a high-density observation network.
Kawabata et al. (2011) assimilated radar reflectivity data for forecasting a line-shaped rainband, but the forecasting period was as short as 30 min. This limitation reflects the nonlinearity of the rainband’s environmental field, which was difficult to predict environmental fields of the rainband on the basis of observations limited to inside the cumulonimbus. The authors concluded that assimilating environmental fields is crucial for forecasting MCSs. Kawabata et al. (2013) developed an assimilation method that used slant total delay observations from the global positioning system (GPS) to provide environmental information such as temperature and humidity information, and they showed that this assimilation appeared to be effective in forecasting a rainband. In this study, we focused on environmental wind observations.
This paper is organized as follows: An overview of the Itabashi heavy rainfall event is given in section 2. The assimilation system and experimental design are described in section 3. The assimilation and forecast results, plus impact tests for each observation type and observational errors, are presented in section 4. Section 5 presents discussion and conclusions.
2. Overview of the Itabashi heavy rainfall event
The surface weather map from the time of the Itabashi rainfall event (Fig. 1) shows that there were several low-pressure systems around Japan, and a baiu front existed south of the Japanese islands. A short wave in the baiu front just south of the Kanto Plain (partially overlapping the experimental domain in Fig. 1) moved eastward while the baiu front moved southward until evening (not shown). However, these large-scale systems did not directly affect rainfall on the Kanto Plain. Upper sounding data observed at Tateno (Fig. 2) show that the atmosphere below 700 hPa was very humid and that wind directions below 850 hPa were southerly to south-southeasterly. In addition, a weak inversion layer was visible around 700 hPa. Convective available potential energy (CAPE) and the level of neutral buoyancy (LNB) were approximately 71 J kg−1 and 630 hPa, respectively. It would have been difficult to initiate deep convection under such atmospheric conditions.
(left) Surface weather map at 0900 JST 5 Jul 2010 and the (right) experimental domain and observation stations and land names appearing in this study. Red crisscrosses in the experimental domain show wind observation stations used in the verification (see section 4a). Grayscale indicates surface elevation.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
Upper sounding data at 0900 JST 5 Jul 2010 at Tateno.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
The MCS that produced the Itabashi rainfall event, MCS A (Fig. 3), initiated west of Tokyo around noon, then traveled slowly eastward. It is clear that the MCS was not maintained by orographic effects because the MCS traveled over flat land (Fig. 1). Around 1600 JST, the MCS approached the DWL instrument installed by the National Institute of Information and Communications Technology (NICT; Fig. 3). At this time the MCS was becoming enhanced, with maximum radar reflectivity exceeding 30 dBZ. The MCS measured approximately 40–50 km from east to west and 20 km from north to south. The MCS moved east and produced heavy rainfall that reached its greatest total amount of 125 mm at Itabashi (Fig. 4) within 1.5 h. This rainfall caused flash flooding of the channelized and heavily sewage-fed Shakujii River, which rose 6.6 m within an hour.
Radar reflectivity observed by the Haneda Airport radar (gray shaded) and Doppler radial velocity observed by the NICT lidar (color; location is shown by a black circle). The locations of the Haneda radar and NICT lidar are shown in Fig. 1; A and B denote MCSs.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
Time series (JST; 5 Jul 2010) of 10-min rainfall amounts (left axis) and the water level of the Shakujii River (right axis) observed at Itabashi.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
Our study focused on the enhancement stage of the Itabashi event, when the NICT DWL station effectively captured the inflow wind field into MCS A. From the DWL radial velocity observation, a southeasterly inflow wind direction can be estimated. Another MCS (B in Fig. 3) south of DWL is visible in Fig. 3, but it was much weaker than MCS A.
3. Observations and assimilation system
a. Observations
The DWL used in this study (Ishii et al. 2010) is a coherent differential absorption and wind lidar deployed at the NICT observation site (Fig. 1) that observes CO2 concentration and radial velocity using a 2-μm Tm,Ho:YLF laser with high output power (2.4 W). The laser has an operating wavelength of 2.05 μm and is strongly backscattered by aerosols. The elevation angle was fixed at 4°, the horizontal resolution along the beam was 76 m, the azimuthal resolution was 2°, and the rotation speed was 1 min−1 (Iwai et al. 2013). Observation range was approximately 10–15 km depending on atmospheric conditions. DWL observations were assimilated after each rotation.
We also assimilated radial velocity and radar reflectivity observations from Doppler radar instruments at the Haneda and Narita Airports, as well as GPS precipitable water vapor (GPS-PWV) data supplied by the Geospatial Information Authority of Japan and analyzed following Shoji (2013). The locations of these observations are shown in Fig. 1. Radial velocities and reflectivity were assimilated for each elevation layer every 1 min, and GPS-PWV values were assimilated every 5 min.
Evaluations of the assimilation method were conducted with two sets of JMA observations. One is rainfall intensity (millimeters per hour) distribution data, which are derived by the JMA radar network and corrected with JMA rain gauge observations. The other is surface wind observations by the JMA automated surface observation stations.
b. Assimilation method of DWL
Because the observation resolution of the DWL data was very high (76 m), we adopted a superobservation method similar to the assimilation method for radial velocity and radar reflectivity observations by Doppler radar, following previous studies (Kawabata et al. 2007, 2011). In these studies, observations of extremely high resolution (e.g., representing turbulence) were smoothed and a “super observation” was produced representing the average wind in a 2-km grid box. In general, this procedure is suitable for data assimilation schemes because short waves generated by high-resolution observations may disturb model dynamics. The DWL distributions and reflectivity observations in Fig. 3 are examples of such super observations. High-reflectivity areas, such as those labeled A and B in Fig. 3, are discussed in this paper.
Next, a model equivalent of radial velocity corresponding to the observations was created in the model by using zonal and meridional wind vectors at the eight nearest grid points of the model. This step took into consideration the broadening of the laser beams and the curvature of Earth. Only horizontal winds were used in the model because the elevation angle of the lidar was low, and vertical winds outside of the cumulonimbus were assumed to be weak for super observations. The small instrumental error in the DWL makes possible a high accuracy of 0.12 m s−1 and no relevant bias for radial velocities, both of which were verified with in situ observations by Iwai et al. (2013). DWL wind observations are more accurate than radial winds observed by Doppler radars, but because the narrowness of the DWL beam led to large representative errors, we assigned DWL observations the same error that we used for Doppler radar observations (1 m s−1) in our previous studies. As this error value is only a crude estimate; we performed two sensitivity experiments with different assigned errors (described in section 4c).
The principal difference between DWL and radial velocity measurements by Doppler radar arises from differences in the size of the beams, as lidar beams only broaden by a few decimeters. Thus, only the eight grid points neighboring an observation point are used in DWL assimilation, whereas a much wider set of grid points is used in Doppler radar assimilation. The original assimilation method for radial velocity derived from Doppler radar was developed by Seko et al. (2004). GPS-PWV observations were assimilated with consideration of the difference between actual and model topography (Kawabata et al. 2007). Radar reflectivity observations were directly assimilated using the Z–Qr relationship (Sun and Crook 1997; Kawabata et al. 2011).
c. Assimilation system
In this study, we used the nonhydrostatic four-dimensional variational data assimilation system (NHM-4DVAR), which is based on the JMA nonhydrostatic model (JMANHM) and is designed to reproduce and predict MCSs at cloud-resolving scale. We adopted the full model of JMANHM (Saito et al. 2006, 2007; Saito 2012) as the forward model, which includes three ice bulk cloud microphysics and excludes cumulus convection parameterization. In the first version of NHM-4DVAR (Kawabata et al. 2007), the adjoint model considered only dry dynamics and advection of water vapor. Kawabata et al. (2011) added a warm rain process to the adjoint model for assimilating radar reflectivity data. The horizontal resolution of NHM-4DVAR is 2 km. The control variables are the three wind components, potential temperature, surface pressure, nonhydrostatic pressure, total water (water vapor and cloud water), the relative mixing ratio of rainwater, and the pseudorelative humidity calculated using the saturation mixing ratio of water vapor that is defined in the background field (only for lateral boundary conditions).
d. Experimental design
To investigate the effects of DWL assimilation, we conducted three assimilation experiments using data denial. In the first experiment, a forecast was initiated with the first-guess field (hereafter NODA); that is, NODA was a downscaled forecast without assimilation. In the second experiment, Doppler radial velocity and reflectivity observations by Doppler radar and GPS-PWV data were assimilated (hereafter CTL). In the third experiment, the DWL observations were assimilated in addition to the observations used in CTL (hereafter LDR).
For further investigation of the impact of the DWL assimilation, we conducted five additional data denial experiments: 1) only radar data (reflectivity and radial velocity) were assimilated; 2) only GPS-PWV data were assimilated; 3) only DWL data were assimilated; 4) DWL and radar data were assimilated; and 5) DWL and GPS-PWV data were assimilated. In addition, we conducted tests of the impact of observational errors in DWL data using two different error values of 1.5 and 2.0 m s−1 (LDR15 and LDR20, respectively).
In the assimilation domain (Fig. 1), the horizontal grid spacing was 2 km, and the assimilation window was set from 1600 to 1630 JST 5 July 2010. After the assimilations, 1.5-h forecasts were started at 1600 JST. Forecasts were conducted using the JMANHM with cloud microphysics and 2-km horizontal grid spacing, excluding a cumulus parameterization scheme. In this paper, simulations of 1600–1630 JST are called “assimilations” and subsequently calculated simulations of 1630–1730 JST are called “forecasts.”
First-guess and lateral boundary conditions were generated by a downscaling method. Dynamical downscaling was conducted by the JMANHM with a horizontal grid spacing of 2 km from 1200 to 1800 JST. The 3-hourly JMA operational mesoscale analyses (MANAL) with 5-km horizontal grid spacing were used as initial and boundary conditions. Only three ice cloud microphysics were used as a precipitation process; that is, no convection parameterization was used.
4. Assimilation and forecast results
a. Assimilation impact on rainfall forecasts
Two MCSs, A and B, are present in the rainfall intensity distribution from the JMA operational radar network, which is a composite of multiple radars and is corrected with JMA surface rain gauges (Fig. 5a). MCS A produced the Itabashi heavy rainfall, which was about 40 km across and reached a maximum rainfall intensity of 57.6 mm h−1.
Distribution of surface wind and rainfall intensity (mm h−1) derived by the (a) JMA operational radar network and 1-h accumulated rainfall amounts in forecasts using (b) NODA, (c) CTL, and (d) LDR at 1730 JST.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
The forecast of the NODA experiment (Fig. 5b) included a small but high-intensity MCS (A′) with a maximum rainfall intensity of 56 mm h−1, located west of the observed MCS A. In addition, MCS B was smaller than and south of the observed MCS B.
In the CTL experiment (Fig. 5c), MCS A had a location and scale close to those of the observed MCS, but its intensity was low (<10 mm h−1). The location and scale of MCS B were well predicted, but its intensity was very low (<1 mm h−1). The spurious MCS A′ in NODA was weakened in CTL by assimilating null data for reflectivity observations, as was done by Kawabata et al. (2011).
The LDR experiment (Fig. 5d) predicted MCS A fairly well. The maximum rainfall intensity of 56 mm h−1, the horizontal scale of approximately 40 km, and the location were very close to the observations, but two peaks of maximum intensity appeared. MCS B was also predicted, and its scale and location were close to the observations, although its maximum intensity was relatively low. It can be concluded that because MCS A was well predicted only in the forecast of LDR, the DWL assimilation has an impact on the rainfall forecast.
To assess improvements of forecast skill, we calculated threat scores (Table 1) at grid points spaced 2 km apart using JMA operational radar observations. The threat score for a 1.0-mm threshold of precipitation in CTL (0.03) was slightly better than in NODA (0.01), but the threat score in LDR (0.24) showed a distinct improvement. Threat scores for a 10.0-mm threshold in NODA and CTL (both 0.00) show that they have no skill in forecasting heavy rainfall, whereas the threat score in LDR (0.11) indicates meaningful skill.
Threat scores of NODA, CTL, and LDR verified with distribution of 1-h rainfall amount (Fig. 5a). Numbers of 1.0 and 10.0 mm represent threshold of the verification.
To illustrate the factors contributing to improved rainfall forecasts, Fig. 6 shows the difference (LDR minus NODA) in precipitable water vapor at the beginning of the assimilation window. The difference was due mainly to the GPS-PWV assimilation. Because MCSs A′ and B in NODA (Fig. 5) were located to the east and south of the observations, precipitable water vapor in these areas decreased after assimilating GPS-PWV data. Moreover, precipitable water vapor increased in MCSs A and B in LDR. These changes improved the locations of MCSs A and B in LDR and may have contributed to the increasing rainfall amounts of MCSs A and B in LDR. The difference increased with time (not shown), and the difference in rainfall amounts between the two experiments became large.
Difference (LDR minus NODA) in precipitable water vapor (shaded) and the mixing ratio of rainwater in LDR at a height of 20 m (contours, interval of 0.5 g kg−1) at 1600 JST.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
The impact of assimilating the DWL observations on horizontal winds, as indicated by the difference between LDR and CTL, is shown in Fig. 7. Wind speeds increased southeast of MCS A and south of MCS B. These areas correspond to inflow (windward) areas of the MCSs. Difference vectors of wind direction to the southeast of MCS A indicate a northeast direction, indicating that the assimilation changed the direction of the inflow wind toward the south. Increments of wind fields were distributed over a wider area than the DWL observation range (~20-km radius). The predicted winds were verified with the JMA automated surface observations in the area of inflow winds to MCS A (see right panel of Fig. 1), and the root-mean-square error (RMSE) was used as the verification score. RMSEs in CTL were 1.41 m s−1 at 1600 JST and 1.13 m s−1 at 1630 JST, and the corresponding RMSEs in LDR were 1.42 m s−1 and 1.09 m s−1, respectively. Because the RMSE at the end of the assimilation window (1630 JST) in LDR is slightly better than that in CTL, we conclude that the 4D-Var assimilation of the DWL data yielded improvements in both the wind field and the rainfall forecast.
Difference (LDR minus CTL) in wind (vectors) and wind speed (color) at a height of 225 m at 1600 JST. Contours show the mixing ratio of rainwater (interval of 0.5 g kg−1) in LDR.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
The difference between LDR and CTL in water vapor flux is shown in Fig. 8. A positive flux area southeast of MCS A is consistent with the area of modified horizontal wind. This enhancement in water vapor flowing into MCS A accounts for the heavy rainfall event. In addition, the difference in water vapor flux south of MCS B, generated by the improvement in the modified wind field, appears to improve the forecasting. Because these two differences are located windward of the MCS, these modifications may enhance the provision of water vapor to MCS A over the open sea. We conclude that assimilating the DWL observations improved the rainfall forecast through improvements in the wind field first and then in water vapor flux.
Difference (LDR minus CTL) in water vapor flux (color). Vectors and contours show horizontal wind and the mixing ratio of rainwater (interval of 0.5 g kg−1), respectively, in LDR.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
b. Sensitivity experiments for individual data types
To further examine the impact of the individual observation types, we conducted experiments that assimilated them selectively (Fig. 9). Experiment LDR (Fig. 9a) clearly shows that assimilating all kinds of data (Doppler velocity and reflectivity by Doppler radar, GPS-PWV, and DWL data) suppressed the spurious MCS A′ and reinforced the intensities of MCSs A and B over the results of the NODA experiment. When we assimilated only radar data (Fig. 9b), the spurious MCS A′ was suppressed and a part of MCS A was promoted. As mentioned in Kawabata et al. (2011), the assimilation method for selected null reflectivity data easily suppressed a spurious MCS, whereas creating heavy rainfall was difficult because of the nonlinearity. Considering these points, the result of this experiment is reasonable. However, adding DWL observations to radar did not improve the forecast (Fig. 9c). The number of DWL observations was three orders of magnitude smaller than the number of radar data (102 and 105, respectively) and thus they had much smaller weight. When only DWL observations were assimilated, a clear influence on the northern part of MCS A was apparent (Fig. 9d), perhaps because southerly winds increased but water vapor did not such that MCS A fully strengthened. When we assimilated only GPS-PWV data, only the southern part of MCS A appeared (Fig. 9e). Combining DWL and GPS-PWV data appeared to smoothly merge their separate assimilation effects (Fig. 9f), likely because both observations had comparable numbers (both on the order of 102). Notably, the three of these five cases without radar data did not suppress the spurious MCS A′. It is interesting that for each individual observation type the assimilation effects on MCS A are weak, but in combination their effect is large (Fig. 9a).
Difference of rainfall amounts with respect to NODA in forecast results of LDR assimilating (a) all data types, (b) only radar data, (c) only radar and DWL data, (d) only DWL data, (e) only GPS-PWV data, and (f) only DWL and GPS-PWV data.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
c. Sensitivity experiments for different assigned errors for DWL data
In general, the magnitude of the observation error is important for data assimilation. We assigned a value of 1.0 m s−1 to the observational error of DWL data in this study, given our previous experience and the instrumental and representativeness errors of the DWL data. To evaluate this choice in greater detail, we investigated the effects of different error magnitudes. First, we found that the standard deviation between the DWL data and the first guess was 1.9 m s−1 during the 30-min assimilation window. We then conducted two runs of experiment LDR using error magnitudes of 1.5 and 2.0 m s−1 (hereafter LDR15 and LDR20, respectively).
The pattern of rainfall distribution in LDR15 and LDR were almost the same, and the maximum intensity was only slightly decreased from 56.5 in LDR to 52.5 mm h−1 in LDR15 (not shown). However, LDR20 showed a large impact. Although the rainfall pattern was very close to those in LDR and LDR15, and the maximum intensity fell by half to 27.7 mm h−1 (Fig. 10a; cf. Fig. 5d), the assimilation of DWL data with a large error of 2.0 m s−1 still changed the wind field surrounding the observation sites and upstream from MCS A (cf. Fig. 7); however, the magnitude of the difference was definitely smaller. This small difference in the wind field led to a small difference in water vapor flux (not shown). Therefore, we conclude that doubling the observational error has a large impact on the assimilation and the forecast.
(a) Distribution of surface wind and rainfall intensity (mm h−1) derived by LDR20 and (b) difference (LDR20 minus CTL) in wind (vectors) and wind speed (color) at a height of 225 m at 1600 JST.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
As the standard deviation of 1.9 m s−1 was calculated with about 100 observations within only 30 min, we need more data to determine its statistical reliability. In addition, the rainfall intensity in LDR20 was stronger and the pattern of rainfall distribution was much closer to the observation than in our other experiments. This shows the importance of assimilating all available types of data simultaneously.
d. Mechanisms of MCS A
In this section, we discuss details of the structures of the predicted cumulonimbi in experiment LDR. Although MCS B dissipated at 1700 JST, MCS A was still intense at that time (Fig. 11a) with cloud tops reaching 6 km AGL, close to but still below the freezing level (Fig. 11c). Water particles in the cores of these predicted cumulonimbi (line L–M in Fig. 11a) were distributed as liquid water particles almost up to the freezing level (Fig. 11c), and ice water contents were very low even above the freezing level (Fig. 11b). We conclude that these predicted cumulonimbi were formed only with warm rain particles.
(a) Horizontal distribution of the mixing ratio of rainwater at 1700 JST at a height of 225 m and (b) vertical cross sections along line L–M of the mixing ratio of ice water (cloud ice, snow, graupel, and hail) and temperature and (c) the mixing ratio of liquid water (cloud and rainwater) and temperature.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
The cloud-top limit of 6 km AGL reflects the existence of a stable layer, as shown along line O–P in Fig. 11a, which is aligned in the wind direction of the sea breeze from Tokyo Bay (Fig. 12a). That is, because the value of dθ/dz was high at 5–6 km AGL (the inversion layer), convection was inhibited above this level. The upper sounding from that morning at Tateno (Fig. 2) shows two weak inversion layers, which may have been produced by the lowering of upper layers at ~700 and ~450 hPa, similar to the forecast result. The JMA operational mesoscale analysis confirms that a stable layer of this type was distributed above and south of the Kanto Plain, together with a weak trough of 552–551 hPa at 5 km AGL (not shown).
Vertical cross sections of (a) dθ/dz (shades), (b) the mixing ratio of water vapor (shades), winds projected on this plane (vectors), and the mixing ratio of rainwater (contours) along the line O–P in Fig. 11a at 1700 JST.
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
Furthermore, generous water vapor inflow is important for heavy rainfall, even when only the warm rain process is operative. Very humid air flowed strongly into the predicted cumulonimbus, with winds of >10 m s−1 below 1.5 km AGL and a mixing ratio of water vapor of 16–18 g kg−1 (Fig. 12b). This strong input of humid air probably was effective in generating water particles and inducing the heavy rainfall event. In addition, the change of wind directions at low levels due to assimilation may have enhanced the efficiency of simulated rainfall production.
In a 3D image of the predicted MCS and its surrounding airflow at 1700 JST (Fig. 13), cloud tops are limited to below 6 km AGL, a south-southeasterly wind of 8–10 m s−1 blows below 2 km AGL, and westerly winds are present above the clouds, as they were in the upper sounding observations (Fig. 2). The resulting wind shear is what built up the inversion layer at 6 km AGL. These characteristic features of the MCS contrast with those of another MCS in Tokyo in 2005, reported by Kawabata et al. (2011), which was primarily the result of deep convection (for the corresponding 3D image, see Fig. 3 of Saito 2012). The characteristic low cloud-top height of the MCS in this study is consistent with the radar data analysis of Yamada (2012).
The 3D image of the predicted MCS in this study along with the mixing ratio of water substances (blue isosurface; >1.5 g kg−1) and cloud water (gray isosurface; >3 g kg−1).
Citation: Monthly Weather Review 142, 12; 10.1175/MWR-D-13-00362.1
5. Discussion and conclusions
In this study, we investigated the effect of assimilating Doppler wind lidar (DWL) observations by applying a cloud-resolving nonhydrostatic 4D-Var assimilation system to forecasting the Itabashi heavy rainfall event of 5 July 2010. We examined the individual impacts of assimilating Doppler radar and GPS water vapor observations, and we evaluated the sensitivity of these impacts to the observational error assigned to DWL data. The primary aim of our study was to evaluate the effectiveness of DWL assimilation in heavy rainfall forecasting. A secondary aim was to investigate the mechanisms involved in the development of the MCS.
The superobservation method was adopted for the DWL data assimilation to reduce the effects of turbulence. Observational error was set to 1 m s−1, identical to that of the Doppler radar radial velocity assimilation in our previous studies. The assimilation methods for Doppler radar and lidar data are very similar except for considerations of their beam broadening.
Three assimilation experiments with a 2-km horizontal grid spacing and a 30-min assimilation window were conducted: the first with no observations (NODA), the second assimilating radar reflectivity and radial velocities from Doppler radar plus GPS-PWV (CTL), and the third adding DWL radial velocities to the second experiment (LDR). After the assimilations, 1.5-h forecasts were carried out. Five more experiments tested different combinations of assimilated data, and two more experiments tested different values for observational errors.
An intense MCS was predicted in NODA, but its location was displaced compared with the observations; in CTL it was weaker than in the observations; and LDR succeeded in capturing its intensity, location, and horizontal scale. The location and scale of another weak MCS were well predicted in CTL and LDR. The threat scores of the forecast skill showed that CTL and LDR improved forecasts of moderate rainfall, but only LDR had meaningful skill in a forecast of heavy rainfall. We conclude that the assimilation of DWL observations improved the forecasting of the heavy rainfall event. Assimilating DWL improved the wind field forecast and led in turn to an improvement in the water vapor flux forecast, specifically by providing the intense MCS with easy access to water vapor from the sea.
Simulation results from assimilating radar data but not DWL data showed that a spurious MCS in NODA was successfully suppressed and that the location and scale of the two MCSs were well simulated, but not their intensities. Generally speaking, direct assimilation of cloud microphysical variables improves forecast skills related to the location and intensity of MCSs. However, assimilation of radar reflectivity data is problematic because of nonlinearity in the cloud microphysics (Kawabata et al. 2011; Wang et al. 2012). Therefore, it is necessary for intensifying MCSs to perform multiple cycles of forecasts and analysis. On the other hand, assimilating null observation of radar reflectivity, developed by Kawabata et al. (2011), was effective in suppressing spurious convection, because assimilating null observation is relatively linear. These factors appear to be the reasons why RMSE was improved in experiment CTL but the intensity of the MCS was still weak.
Through impact tests for each observation type, we found that assimilating radar data contributed to suppressing a spurious MCS, and GPS-PWV and DWL data contributed to creating an intense MCS. The effect of each observation type was weak individually, but their combination in experiment LDR yielded a notable improvement.
Our sensitivity experiments using different assigned error levels for DWL data showed that even when the error was doubled the results of experiment LDR were superior to those of other experiments. We conclude that simultaneous assimilation of radar, DWL, and GPS-PWV data is a robust technique for improved forecasting of intense MCSs.
We used our simulation to explore the mechanisms of the intense MCS. Only liquid water particles and no ice water particles were present (Yamada 2012). The MCS was, therefore, formed by warm rain clouds, as the cloud tops were kept below 6 km AGL by an inversion layer. A stable layer of this type was present over the Kanto Plain in our simulation. It also showed how very moist air (>18 g kg−1 water vapor) in combination with strong onshore winds (>10 m s−1) caused the heavy rainfall of 5 July 2010.
Fujibe (2002) has shown that isolated thunderstorms over the Kanto Plain tend to arise from a meso-β-scale convergence between the southerly sea breeze from Tokyo Bay and the northeasterly sea breeze from the Pacific Ocean. Two lidar instruments deployed by the JMA at Haneda Airport are very useful for observing the southerly sea breeze. The limits imposed by their observational range can be overcome by the 4D-Var technique to expand the effect of assimilating DWL data to a wider area, including the offshore areas of this study, through background error covariance and flow dependency of 4D-Var. Therefore, the combination of DWL data and 4D-Var can be a powerful tool for operational rainfall forecasts if DWLs are sited where they can capture the wind fields affecting hazardous weather conditions.
This study demonstrates that data from a single DWL, despite its limited observation range, can improve a heavy rainfall forecast and that a minor change in wind direction can result in large changes in water vapor transportation. These results illustrate the importance of environmental field observations of MCSs.
Our study suggests some topics for future research. Four additional DWLs operate in the Kanto Plain, and we plan to use their data in other heavy rainfall cases. Because three of these DWLs are operated by the JMA for routine observations at airports, assimilating their observations is expected to improve operational NWP scores. GPS slant total delay observations are effective for representing environmental fields of temperature and humidity and should be assimilated together with DWLs. A final issue is estimating accurate observational errors for DWL data.
Acknowledgments
We thank Dr. Tadashi Tsuyuki at the Meteorological Research Institute (MRI) for his useful advice. We thank Mr. Hiroshige Tsuguti at the MRI for productive discussions and Mr. Akifumi Nishi at the University of Tsukuba for drawing 3D images of the MCS. We are grateful to the anonymous reviewers and the editor for their useful suggestions. The Geospatial Information Authority of Japan kindly provided GPS observations. This study was partly supported by the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) through a Grant-in-Aid for Scientific Research (21244074), “Study of advanced data assimilation and cloud-resolving ensemble technique for the prediction of local heavy rainfall.”
REFERENCES
Baker, W. E., and Coauthors, 2014: Lidar-measured wind profiles—The missing link in the global observing system. Bull. Amer. Meteor. Soc.,95, 543–564, doi:10.1175/BAMS-D-12-00164.1.
Banta, R. M., L. D. Olivier, P. H. Gudiksen, and R. Lange, 1996: Implications of small-scale flow features to modeling dispersion over complex terrain. J. Appl. Meteor., 35, 330–342, doi:10.1175/1520-0450(1996)035<0330:IOSSFF>2.0.CO;2.
Bluestein, H. B., J. B. Houser, M. M. French, and J. C. Snyder, 2014: Observations of the boundary layer near tornadoes and in supercells using a mobile, collocated, pulsed Doppler lidar and radar. J. Atmos. Oceanic Technol., 31, 302–325, doi:10.1175/JTECH-D-13-00112.1.
Done, J., C. A. Davis, and M. Weisman, 2004: The next-generation of NWP: Explicit forecasts of convection using the Weather Research and Forecasting (WRF) model. Atmos. Sci. Lett., 5, 110–117, doi:10.1002/asl.72.
Drechsel, S., M. Chong, G. J. Mayr, M. Weissmann, R. Calhoun, and A. Dörnbrack, 2009: Three-dimensional wind retrieval: Application of MUSCAT to dual-Doppler lidar. J. Atmos. Oceanic Technol., 26, 635–646, doi:10.1175/2008JTECHA1115.1.
Drechsel, S., G. J. Mayr, M. Chong, and F. K. Chow, 2010: Volume scanning strategies for 3D wind retrieval from dual-Doppler lidar measurements. J. Atmos. Oceanic Technol., 27, 1881–1892, doi:10.1175/2010JTECHA1495.1.
Fujibe, T., 2002: Surface wind patterns preceding short-time heavy rainfall in Tokyo in the afternoon of midsummer days (in Japanese). Tenki, 49, 393–405.
Hara, T., and Coauthors, 2013: The operational convection-permitting regional model at JMA. Res. Act. Atmos. Oceanic Modell.,43, 5.5–5.6. [Available online at http://www.wcrp-climate.org/WGNE/BlueBook/2013/individual-articles/05_Hara_Tabito_wgne_2013_lfm.pdf.]
Huang, X.-Y., and Coauthors, 2009: Four-dimensional variational data assimilation for WRF: Formulation and preliminary results. Mon. Wea. Rev., 137, 299–314, doi:10.1175/2008MWR2577.1.
Ishii, S., and Coauthors, 2006: Development of 2-micron airborne coherent Doppler lidar at NICT. Lidar Remote Sensing for Environmental Monitoring VII, U. N. Singh, T. Itabe, and D. N. Rao, Eds., International Society for Optical Engineering (SPIE Proceedings, Vol. 6409), 64090J, doi:10.1117/12.693526.
Ishii, S., and Coauthors, 2010: Coherent 2 μm differential absorption and wind lidar with conductively cooled laser and two-axis scanning device. Appl. Opt., 49, 1809–1817, doi:10.1364/AO.49.001809.
Iwai, H., S. Ishii, R. Oda, K. Mizutani, S. Sekizawa, and Y. Murayama, 2013: Performance and technique of coherent 2-μm differential absorption and wind lidar for wind measurement. J. Atmos. Oceanic Technol., 30, 429–449, doi:10.1175/JTECH-D-12-00111.1.
Kato, T., and K. Aranami, 2005: Formation factors of 2004 Niigata-Fukushima and Fukui heavy rainfalls and problems in the predictions using a cloud-resolving model. SOLA, 1, 1–4, doi:10.2151/sola.2005-001.
Kawabata, T., H. Seko, K. Saito, T. Kuroda, K. Tamiya, T. Tsuyuki, Y. Honda, and Y. Wakazuki, 2007: An assimilation and forecasting experiment of the Nerima heavy rainfall with a cloud-resolving nonhydrostatic 4-dimensional variational data assimilation system. J. Meteor. Soc. Japan, 85, 255–276, doi:10.2151/jmsj.85.255.
Kawabata, T., T. Kuroda, H. Seko, and K. Saito, 2011: A cloud-resolving 4D-Var assimilation experiment for a local heavy rainfall event in the Tokyo metropolitan area. Mon. Wea. Rev., 139, 1911–1931, doi:10.1175/2011MWR3428.1.
Kawabata, T., Y. Shoji, H. Seko, and K. Saito, 2013: A numerical study on a mesoscale convective system over a subtropical island with 4D-Var assimilation of GPS slant total delays. J. Meteor. Soc. Japan, 91, 705–721, doi:10.2151/jmsj.2013-510.
Kusunoki, K., 2002: A preliminary survey of clear-air echo appearances over the Kanto Plain in Japan from July to December 1997. J. Atmos. Oceanic Technol., 19, 1063–1072, doi:10.1175/1520-0426(2002)019<1063:APSOCA>2.0.CO;2.
Newsom, R. K., and R. M. Banta, 2004: Assimilating coherent Doppler lidar measurements into a model of the atmospheric boundary layer. Part I: Algorithm development and sensitivity to measurement error. J. Atmos. Oceanic Technol., 21, 1328–1345, doi:10.1175/1520-0426(2004)021<1328:ACDLMI>2.0.CO;2.
Rennie, S., J. S. L. Dance, A. J. Illingworth, S. P. Ballard, and D. Simonin, 2011: 3D-Var assimilation of insect-derived Doppler radar radial winds in convective cases using a high-resolution model. Mon. Wea. Rev., 139, 1148–1163, doi:10.1175/2010MWR3482.1.
Saito, K., 2012: The JMA nonhydrostatic model and its application to operation and research. Atmospheric Model Applications, I. Yucel, Ed., Intech, 85–110, doi:10.5772/35368.
Saito, K., and Coauthors, 2006: The operational JMA nonhydrostatic mesoscale model. Mon. Wea. Rev., 134, 1266–1298, doi:10.1175/MWR3120.1.
Saito, K., J. Ishida, K. Aranami, T. Hara, T. Segawa, M. Narita, and Y. Honda, 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan, 85B, 271–304, doi:10.2151/jmsj.85B.271.
Seko, H., T. Kawabata, T. Tsuyuki, H. Nakamura, K. Koizumi, and T. Iwabuchi, 2004: Impacts of GPS-derived water vapor and radial wind measured by Doppler radar on numerical prediction of precipitation. J. Meteor. Soc. Japan, 82, 473–489, doi:10.2151/jmsj.2004.473.
Shoji, Y., 2013: Retrieval of water vapor anisotropy using the Japanese nationwide GPS array and its potential for prediction of convective precipitation. J. Meteor. Soc. Japan, 91, 43–62, doi:10.2151/jmsj.2013-103.
Shun, C. M., and P. W. Chan, 2008: Applications of an infrared Doppler lidar in detection of wind shear. J. Atmos. Oceanic Technol., 25, 637–655, doi:10.1175/2007JTECHA1057.1.
Skamarock, W.C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. Barker, W. Wang, and J. G. Powers, 2005: A description of the Advanced Research WRF version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp. [Available online at www2.mmm.ucar.edu/wrf/users/docs/arw_v2.pdf.]
Stoffelen, A., and Coauthors, 2005: The atmospheric dynamics mission for global wind measurement. Bull. Amer. Meteor. Soc., 86, 73–87, doi:10.1175/BAMS-86-1-73.
Sun, J., and A. Crook, 1997: Dynamical and microphysical retrieval from Doppler radar observations using a cloud model and its adjoint. Part I: Model development and simulated data experiments. J. Atmos. Sci., 54, 1642–1661, doi:10.1175/1520-0469(1997)054<1642:DAMRFD>2.0.CO;2.
Takahashi, T., 1981: Warm rain study in Hawaii—Rain initiation. J. Atmos. Sci., 38, 347–369, doi:10.1175/1520-0469(1981)038<0347:WRSIHI>2.0.CO;2.
Tang, Y., H. W. Lean, and J. Bornemann, 2013: The benefits of the Met Office variable resolution NWP model for forecasting convection. Meteor. Appl., 20, 417–426, doi:10.1002/met.1300.
Wang, H., T. Auligne, and H. Morrision, 2012: Impact of microphysics scheme complexity on the propagation of initial perturbations. Mon. Wea. Rev., 140, 2287–2296, doi:10.1175/MWR-D-12-00005.1.
Weissmann, M., R. H. Langland, C. Cardinali, P. M. Pauley, and S. Rahm, 2012: Influence of airborne Doppler wind lidar profiles near Typhoon Sinlaku on ECMWF and NOGAPS forecasts. Quart. J. Roy. Meteor. Soc., 138, 118–130, doi:10.1002/qj.896.
Yamada, Y., 2012: A unique cumulonimbus producing a localized heavy rainfall in Tokyo metropolitan during TOMACS. Proc. 16th Int. Conf. on Clouds and Precipitation, Leipzig, Germany, ICCP.
Zhang, F., A. M. Odins, and J. Nielsen-Gammon, 2006: Mesoscale predictability of an extreme warm-season precipitation event. Wea. Forecasting, 21, 149–166, doi:10.1175/WAF909.1.
Zhang, L., and Z. Pu, 2011: Four-dimensional assimilation of multitime wind profiles over a single station and numerical simulation of a mesoscale convective system observed during IHOP_2002. Mon. Wea. Rev., 139, 3369–3388, doi:10.1175/2011MWR3569.1.
