1. Introduction
A wide spectrum of satellite data have been used in operational and research data assimilation systems for numerical weather predictions (NWPs) because of their significantly favorable impacts on maintaining and improving the accuracy of the predictions. However, use of cloud- and precipitation-affected data is still limited. Regarding radiance data of infrared and microwave imagers and sounders, for example, many NWP centers mainly assimilate only clear-sky radiances. This is attributed to the great complexity, non-Gaussian characteristics, and highly nonlinear response of cloud and precipitation processes, which makes it difficult to handle those data in models and data assimilation systems. Recently, significant effort has been made to assimilate radiance data affected by clouds and precipitation with advancements of data assimilation systems and NWP models (McNally 2009; Bauer et al. 2010; Geer et al. 2010; Martinet et al. 2013; Okamoto 2013; Stengel et al. 2013). For precipitation radars on the ground, direct assimilation of the reflectivity factor (hereafter reflectivity, or Ze) in addition to Doppler winds has undergone extensive development, and this has been operationally implemented in several NWP centers (Aksoy et al. 2010; Ikuta and Honda 2011; Kawabata et al. 2011; Wattrelot et al. 2014). In contrast, there has been relatively less effort to assimilate space-based active sensors that measure signals backscattered by clouds or precipitation. Some studies have shown promising results on assimilating space-based precipitation radars and cloud radars (Benedetti et al. 2005; Janisková et al. 2012; Janisková 2015). However, the progress seems to be modest compared with studies on radiances and ground-based radars. This is probably because space-based active sensors are assumed to have much smaller impacts than passive infrared and microwave imagers and sounders because of the narrower observation coverage. Another reason is that there is no solid plan to realize those radars on a future operational observation basis.
However, space-based precipitation radars can obtain detailed vertically resolved information that is not readily available for passive instruments. In addition, space-based precipitation radars can observe areas beyond the range of ground-based radars, such as in areas over the ocean. These features make space-based radars complementary to space-based passive imagers and sounders and ground-based radars. Thus, space-based precipitation radars are expected to have positive impacts on certain cases, for instance, severe storms and tropical cyclones (TCs) over the ocean. Benedetti et al. (2005) showed that by assimilating the Precipitation Radar (PR) on board the Tropical Rainfall Measuring Mission (TRMM) satellite, they could obtain improved track predictions of several TCs in a global data assimilation system. They pointed out that, despite the limited impact on a global scale, even a small number of PR data had a comparable impact with microwave imagers on TRMM when PR sampled a meaningful portion of the storm such as its center.
Despite these successful results, Benedetti et al. (2005) admitted the underuse of vertical information from the active sensors because they assimilated total column water vapor (TCWV) transformed from radar reflectivity profiles. This was justified by their global data assimilation system that then had a relatively crude resolution (~40 km) and in which more emphasis was put on temperature and humidity analyses than cloud variables. Our interest here is on what impacts can be obtained from space-based precipitation radars in assimilation systems based on cloud-resolving models (CRMs) with higher resolution. In particular, we are interested in the situation when microwave imager radiances are already assimilated, and we would like to evaluate what additional impacts precipitation radars give us for severe meteorological situations such as TCs.
To this end, we have developed an assimilation technique for the reflectivity of space-based precipitation radars by using a regional CRM and data assimilation system that can explicitly handle cloud variables. The observations we mainly targeted were the Dual-Frequency Precipitation Radar (DPR) on board the Global Precipitation Measurement (GPM) Core Observatory. The DPR observability was enhanced relative to the PR with respect to better sensitivity, double frequencies, and higher vertical resolution. We selected a case where the GPM Core Observatory sampled the center of a rapidly intensifying TC in the west Pacific Ocean and assimilated vertical Ze profiles from DPR. We also compared impacts of the DPR and the GPM Microwave Imager (GMI) on board the GPM Core Observatory by assimilating them both separately and in combination. This paper is organized as follows. In section 2, observation, CRM, assimilation system and observation operators are briefly explained. Section 3 presents the comparison study between CRM simulation and DPR observation in the reflectivity space. Section 4 demonstrates results of assimilation of DPR and/or GMI. We conclude with summary and future perspectives in section 5.
2. Methods
a. Observations
The GPM Core Observatory is a follow-on satellite for the TRMM, which carried the world’s first space-based precipitation radar, and was launched on 28 February 2014. The GPM Core Observatory has GMI and DPR mounted on board. The DPR is the first spaceborne dual-frequency precipitation radar. It is expected to advance precipitation science by expanding the coverage of observations to higher latitudes than those obtained by the TRMM PR and by measuring snow and light rain via high-sensitivity observations. In addition, the DPR can provide drop size distribution (DSD) information based on the differential scattering properties of the two frequencies. The advanced measurement capability of the DPR will promote improved model microphysics and the analysis of hydrometeor and related dynamical variables with data assimilation. The DPR consists of two radars at the Ku band (13.6 GHz) and Ka band (35.55 GHz), which are named KuPR and KaPR, respectively. The horizontal resolutions of KuPR and KaPR are both about 5.2 km at the ground level when the satellite altitude is 407 km. There are three scanning modes with different geometries and vertical resolutions; these include “KuNS,” “KaMS,” and “KaHS,” where “NS,” “MS,” and “HS” represent normal scans, matched scans, and high sensitivity scans, respectively. For the KuNS mode, the KuPR beam scans 49 angle bins with an entire swath width of 245 km. The range resolution is 250 m, although the data are sampled with 125-m intervals. For the KaMS mode, KaPR scans the same angle bins and range bins as the inner 25 KuPR angle bins with a 125-km swath width. In contrast, the KaHS mode scans the interlaced scan area with a range resolution of 500 m and a sampling range interval of 250 m, and it detects weaker precipitation signals than KaMS.
In this study, we did not use KaMS mode data because of its narrower swath than the KuNS mode and larger noise than the KaHS mode. Range bins with main lobe clutter were removed because they are not able to be simulated by the observation operator. Moreover, because sidelobe clutter contamination is often found in KuPR data (Kubota et al. 2016), range bins where a procedure to reduce sidelobe clutter were applied were removed in this study. The dataset we employed was the DPR level 2 (2ADPR) version V03b, which was released on 2 September 2014. This dataset undergoes radiometric corrections and contains quality information, reflectivity factors, precipitation data, and DSD data. We used attenuation corrected reflectivity factors in both model evaluation and assimilation in this study. Additional details about GPM, DPR, and the associated processing steps can be found in Hou et al. (2014) and Kubota et al. (2014).
The GMI is a conical-scanning microwave imager that has improved capabilities compared to the predecessor TRMM Microwave Imager (TMI). With a rotating antenna, a cone-shaped scan is made with a swath width of 904 km at the ground level, which is much wider than DPR. The GMI has 10 vertically and horizontally polarized channels at 10.65, 18.70, 36.50, 89.00, and 166.0 GHz, and three vertically polarized channels at 23.8, 183.31 ± 3, and 183.31 ± 7 GHz. Hereafter, each channel is called, for example, 19V for the vertically polarized channel at 18.70 GHz after its frequency and polarization. The horizontal resolution or instantaneous field of view (FOV) size varies by 19.4 (32.2) km in the along-scan (cross-scan) direction for 10V to 4.4 (7.2) km for 183V. In assimilation preprocessing, we produced “superobbed” GMI radiances by averaging the original footprints within 5 × 5 model grids (25 km × 25 km) to make consistent footprint sizes at different channels.
b. Model
The CRM used in this study is the Japan Meteorological Agency’s nonhydrostatic model (JMA-NHM; Saito et al. 2006). The JMA-NHM has been used as an operational mesoscale weather prediction system since September 2004. The JMA-NHM used in this study has a horizontal resolution of 5 km and 50 vertical layers up to 21.8 km. It employs a Kain–Fritsch (KF) convection scheme (Kain and Fritsch 1993) and a three-ice bulk microphysics scheme (Ikawa and Saito 1991) based on the work of Lin et al. (1983). Among the several options in the cloud microphysics scheme, we chose to predict the mass mixing ratio of liquid clouds Qc, rain Qr, ice clouds Qci, snow Qs, and graupel Qg, and number density of ice clouds Nci, snow Ns, and graupel Ng. The DSDs were assumed to follow an inverse exponential function for rain, snow, and graupel, and a monodisperse function for liquid and ice clouds. These settings are the same as the operational ones except for the two-moment bulk scheme for snow and graupel.
c. Assimilation scheme
The data assimilation scheme was an ensemble variational (EnVA) method with preprocessing of displacement error correction (DEC; Aonashi and Eito 2011). The EnVA method seeks an optimum analysis state that minimizes a cost function defined in the ensemble forecast error subspace. We defined the control variables as the zonal, meridional, and vertical winds (U, V, and W), the potential temperature PT, the ratio of the total water content (sum of Qi, Qc, and the humidity mixing ratio) to the saturation specific humidity (RHW), and the sum of the flux of rain, snow, and graupel (Pr). The DEC shifts the first-guess field to minimize the horizontal displacement error defined by a misfit of the spatial pattern between observations and first-guess brightness temperature TB values at 19V. Aonashi et al. (2015, manuscript submitted to Mon. Wea. Rev.) improved EnVA to reduce sampling errors especially for precipitation-related variables by developing a neighboring ensemble (NE) approach and scale-dependent separation of control variables. The NE approach employed 5 × 5 grids surrounding a grid that were analyzed based on spectral localization (Buehner 2012) to virtually increase ensemble members. As for the scale-dependent variable separation, two groups of control variables were defined that consisted of large-scale variables (U, V, Ps, and RHW) and small-scale variables (W, Pr, and anomalies from spatial averages for the large-scale variables). Aonashi et al. (2015, manuscript submitted to Mon. Wea. Rev.) showed that the new EnVA successfully suppressed spurious forecast error correlations.
d. Observation operators
As for the microwave radiances, we adopted the four-stream, plane-parallel radiative transfer model that was developed by Liu (2004, hereafter called LiuRTM). The LiuRTM approximates the single scattering properties of nonspherical ice particles. The LiuRTM was implemented at each model grid point, and the computed radiances were averaged within 5 × 5 model grid boxes to match the superobbed GMI radiances.
3. Model evaluation
We compared model simulations and DPR KuNS and KaHS observations in Ze space to understand the characteristics of the JMA-NHM, Joint-Simulator, and DPR observations. This investigation helped to develop preprocessing of assimilating Ze. Among the several comparison cases we made, one typical result shown in this paper relates to Typhoon Halong. Halong developed in the Mariana Islands on 29 July 2014, moved westward to the Philippines, and rapidly intensified on 1 August 2014. Halong then moved northward and made a landfall in Japan on 6 August 2014. The GPM Core Observatory passed over Halong’s center region around 14°N, 140°E at 1135 UTC 31 July 2014. We ran 12-h forecasts from the operational mesoscale analysis at 0000 UTC 31 July 2014. Before the comparison, we removed observations flagged as bad quality (i.e., by using “FLG%flagEcho” in the 2ADPR dataset) and contaminated by ground clutter (“FLG%qualityData”). We also excluded observations or simulations with Ze values smaller than 14 dBZ. This threshold was chosen through a visual examination to make balance between minimizing Ze noise and maximizing the number of used data, and based on the nominal minimum detectable level of 18 dBZ for KuPR and 12 dBZ for KaPB (Hou et al. 2014). Because a recent study showed the minimum detectable echoes were estimated to be between 12 and 14 dBZ for KuNS and 12 dBZ for KaHS (Toyoshima et al. 2015), probably we can set lower thresholds (e.g., 12 dBZ), especially for KaHS. The comparison was made at observation locations after implementing a bilinear horizontal interpolation of the JMA-NHM at the model layers and computing the radar echo simulation according to Eq. (1) with the Joint-Simulator. It is noted that no vertical interpolation was performed in this comparison study to avoid blurring the vertical structure.
Figure 1 shows horizontal cross sections at 2.5-km altitude and vertical cross sections at the nadir angle bin for KuNS Ze. Removal of observations affected by sidelobe clutter is responsible for the two blank lines along the satellite path in Fig. 1a. Overall, the JMA-NHM reproduces observed Ze around the eyewall well. However, it seems that there is a lack of spread in the modest echo area (Ze = 30–35 dBZ). The vertical cross section in Fig. 1 shows that the simulated strong rain echo expands beyond 10 km around 15.5°N, although the observed echo is capped with a melting layer at 5 km. Figure 2 shows the contoured frequency by altitude diagrams (CFADs), which represent the normalized probability distribution of Ze at different altitudes, for the observed and simulated Ze of KuNS. The melting layer was identified at around a 5-km altitude in the CFAD from observed Ze in Fig. 2a. One notable feature was the overestimation of simulated echo above the melting layer up to 12 km, which made it difficult to distinguish between ice and liquid regions. This problem was already reported earlier by Eito and Aonashi (2009) who compared ground-based precipitation radar and JMA-NHM simulations. They attributed the overestimation to the larger size of simulated snow particles. Kotsuki et al. (2014) also showed that the echo top of the DPR Ze in simulations from the global nonhydrostatic CRM and Joint-Simulator was systematically higher than that of observations, thus suggesting that there was an overestimation of snow and graupel in the model. We further investigated the cause of this overestimation by comparing ice particle diameters from the JMA-NHM and KuNS observations.
KuNS reflectivity Ze (a),(b) observed and (c),(d) simulated for Typhoon Halong at 1200 UTC 31 Jul 2014. (a),(c) The horizontal cross section at an altitude of 2.5 km and (b),(d) the vertical cross section at the nadir angle bin are plotted. The Ze simulation was performed at the ground point of observations and JMA-NHM’s vertical levels; this produced gaps in the vertical cross section corresponding to the model resolution.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Contoured frequency by altitude diagrams (CFAD) for (a) the observed KuNS Ze and (b) the simulated KuNS Ze from the samples shown in Fig. 1. The bin size of Ze and altitude are 2 dBZ and 125 m, respectively. The black lines represent frequencies of 0.01, 0.05, 0.1, and 0.2.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Figure 3 shows the accumulated hydrometeor number density as a function of particle size diameter D and altitude for observations and simulations. This diagram was made by counting particles in a diameter bin based on the DSD form for each hydrometeor. The DSD parameters were determined from the mixing ratio and number density at every grid point for each hydrometeor for the simulations, and from the mass-weighted mean diameter and number density at every radar bin given in the 2ADPR dataset for the observations (Seto et al. 2013). The observation diagram displays the most frequent occurrence at log10(D) ~ −0.3 or D ~ 0.8 mm while the simulation diagram shows it at the smallest diameter. This difference reflects different DSD forms of a gamma distribution for the observations and inverse exponential and monodisperse distributions for the JMA-NHM data. The outer contour at the logarithm of the number density at 0.1 delineates a gradual decline above 5 km for the observations, while it stays constant from 5 to 12 km and rapidly decreases above 12 km for the simulations. These trends correspond well to the CFAD in Fig. 2. We separated the accumulated number density into five hydrometeor species for simulations. Figure 4 shows that snow accounts for the largest portion between 5 and 12 km, especially for hydrometeors with large diameter of D > 10 mm. These results suggest that the overestimation of simulated Ze above the melting layer, especially between 5 and 12 km, can be explained by the JMA-NHM excessively generating larger snow particles, which is consistent with the findings of Eito and Aonashi (2009). This overestimation trend became slightly worse when a one-moment bulk microphysics scheme for snow and graupel was applied (not shown).
Accumulated hydrometeor number density (N, m−3 mm−1) as a function of the particle (D, mm) size diameter and altitude for (a) observations and (b) simulations. The number density and diameter are displayed in the common logarithmic scale. Black lines are plotted where the logarithm of density is 0.1 and 2.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
(a) The same as Fig. 3b, and the accumulated number density for (b) liquid clouds, (c) rain, (d) ice clouds, (e) snow, and (f) graupel.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
The CFAD in Fig. 2 also shows that the highest frequency was around 32 dBZ below 5 km for the observations, while it decreased toward the ground for the simulations. We speculate that this was associated with the evaporation process of stratiform rain in the JMA-NHM in which a one-moment bulk microphysics scheme was adopted for rain hydrometeor. The evaporation process decreases mass of rain particles, and then reduces effective radius in the one-moment scheme. In reality, however, smaller particles are more likely to be evaporated and more large particles remain at a low altitude. Because this decreasing trend could not always be seen in other TC cases, however, more careful investigation will be necessary, for example, by examining the relationship between the reducing radius and downdraft, temperature and humidity, before drawing conclusions.
The CFAD for KaHS in Fig. 5 shows that the overall features were similar to that of KuNS (excessive scattering above the melting layer and decreasing scattering toward the ground in the simulation). The smaller Ze below the melting layer can be probably generated by evaporation process in the model. An additional possible cause is the fact that Joint-Simulator does not include the multiscattering effect. Battaglia et al. (2015) suggested that a single-scattering echo can be more strongly attenuated than a multiple-scattering echo and that this effect was more obvious for the Ka band than the Ku band. Although observations affected by ground clutter were supposed to have been already removed based on the clutter identification flags in the 2ADPR dataset, there still remained anomalously strong Ze below 2 km, thus implying some deficiency in the KaPR clutter identification algorithm. Another interesting difference from KuNS CFAD was the relatively clear distinction between ice and liquid regions. This is probably associated with the fact that scattering processes in the Ka band slightly deflects away from the Mie scattering regime that is strongly sensitive to particle size. Therefore, the JMA-NHM’s snow particle size bias was less evident in KaHS than in KuNS.
As in Fig. 2, but for KaHS and the altitude bin size of 250 m.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
4. Experimental assimilation
a. Assimilation system for DPR reflectivity
We extended EnVA to assimilate reflectivity profiles from space-based precipitation radars by incorporating the Joint-Simulator and developing preprocessing dedicated to reflectivity data. EnVA minimizes a cost function through iterative calculations in a linear approximation (inner loop) and by updating the background state and observation operators (outer loop). We implemented five outer-loop updates and, on average, three to four inner-loop minimizations in the individual outer loop.
Quality control (QC) procedures can exclude not only erroneous observations but also those that are not well reproduced by the models or observation operators. Several QC procedures were developed based on the findings in section 3. Observations in and above the melting layers were discarded because the serious bias of the JMA-NHM and the inability to simulate brightband echo by the JMA-NHM and Joint-Simulator. Observations contaminated by ground clutter were also removed. As the clutter flag in the 2ADPR dataset occasionally missed the identification of clutter contamination, we built a look-up table containing the highest altitude of the clutter-contaminated range bin as a function of the view angle. We excluded bins where there was no precipitation both in observations and ensemble mean first-guess (FG) values. We set the minimum Ze to 14 dBZ for both KuNS and KaHS and considered the presence of precipitation when observed or simulated Ze was over this level. If precipitation was present in observations but not in FG simulations, we assigned the minimum Ze (14 dBZ) to the FG and vice versa. The range bins with isolated precipitation, which was defined as the absence of precipitation in consecutive range bins, were rejected. Finally, we removed cases with large differences between observed Ze and ensemble mean FG Ze values. Note that the QC procedures were applied in each outer loop; hence, some data could be rescued or excluded as the background state was updated.
The DPR Ze observations that passed all of the QC procedures were averaged (or superobbed) in two horizontal and vertical bins. The superobbed Ze reduced the random noise and helped EnVA to effectively minimize the cost functions. Moreover, we believe the reduced spatial resolution in the superob (about 10 km in contrast to the original 5-km resolution) was more agreeable to the effective model representative scale.
Figure 6 shows an example of observed Ze before the QC procedures and superobbed Ze that passed the QC procedures for KuNS Ze. The available data were significantly reduced with the QC procedures, in this case, from 2 658 600 to 133 176 data points (the survival ratio was 5.0%), and then, the data were further decreased by the superobbing procedure to 27 849 data points (1.1%). A similar reduction ratio was found for KaHS; that is, 661 461, 40 438, and 7540 data points before QC procedures, after QC procedures, and after superobbing, respectively (6.1% and 1.1%).
(a),(b) Observed Ze before QC procedures and (c),(d) superobbed Ze after QC procedures for KuNS around Typhoon Halong at 1200 UTC 31 Jul 2014. (a),(c) Bins at 2.5-km altitude and (b),(d) bins at the nadir angle are plotted. The Ze values smaller than 14 dBZ were set to the assumed minimum value of 14 dBZ.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Observation errors were empirically set to 4 dBZ for KuNS and 3 dBZ for KaHS. These figures were larger than what Benedetti et al. (2005) used for PR reflectivities (1.1 dBZ). This could be partially justified by omitting observation error correlations in neighboring bins, but such an approach may need reevaluation in the future. In this study, we did not implement a bias correction procedure because we removed data suffering from serious bias in the ice region.
b. Assimilation experiment setup
We performed assimilation experiments to assess the impacts of DPR reflectivities for the Typhoon Halong case, as already discussed in section 3. Many NWP centers wrongly predicted a northward track instead of westward propagation, and they also failed to predict the rapid intensification. The GMI radiances and/or DPR reflectivities were assimilated at 1200 UTC 31 July 2014 when the GPM Core Observatory satellite overpassed the center region of Halong, Vietnam. We implemented JMA-NHM and EnVA with the same horizontal and vertical resolution (5 km and 50 layers) in 401 × 401 grid points. Figure 7 shows the experiment domain, the track of Halong every 6 h and KuNS and KaHS observation pixels at the assimilation time. We confirmed in several sensitivity tests that this domain was large enough for boundary conditions not to affect analysis and subsequent 48-h forecast. The number of ensemble members was 52, and initial perturbations were created from JMA’s weekly global ensemble forecast system (JMA 2013). We ran 12-h ensemble forecasts, which were used as FG and to construct the flow-dependent background error covariance in EnVA. We found that 12 h was sufficient to generate reasonable hydrometeors and their error covariance by validating, for example, the disappearance of model spinup, in several different cases.
Domain of assimilation experiment (gray shading, 2000 km × 2000 km). Track of Halong and minimum pressures from the best track are plotted every 6 h with colored circles. The square symbol indicates when the assimilation was implemented at 1200 UTC 31 Jul 2014. KuNS and KaHS pixels at the assimilation time were overlaid with green and blue dots, respectively.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
We carried out seven assimilation experiments with different data configurations, and these are summarized in Table 1. The first group of experiments (“Kuonly,” “Kaonly,” and “GMIonly”) assimilated the single instrument KuNS Ze, KaHS Ze, and GMI radiance data, respectively. The second group (“GMI + Ku” and “GMI + Ka”) assimilated both GMI radiance and Ze data from KuNS or KaHS. The third group (“GMI + KuKa”) assimilated all of these three types of observations. In the GMI-related experiments, only clear-sky GMI radiances were included in the first outer loop and rainy GMI radiances were assimilated in the second and later outer loops. Assimilated channels were 10V, 19V, 24V, 37V and 89V, and their observation errors were assigned to 2.236, 2.236, 2.236, 5.0, and 10.0 K for clear-sky condition, and 3.162, 3.162, 3.162, 10.0, and 20.0 K for rainy conditions, respectively. In the Kuonly and Kaonly experiments, Ze was assimilated in all of the five outer loops. In the GMI + Ku, GMI + Ka, and GMI + KuKa experiments, Ze was assimilated in the third to fifth outer loop. This was intended to better assimilate Ze after the radiance assimilation updated the background state, in which clouds and rain should be better represented than by the FG. Note that we did not run cycle experiments in this study indicating that all of these experiments used the same FG field. Conventional data were provided with all of these experiments, but only 47 bogus wind data points (TC bogus; JMA 2013) were assimilated. As a reference experiment, we ran a “CNTL” experiment that assimilated only conventional data. The DEC was applied to all of these experiments including CNTL.
Observation data (“O”) configuration used in the assimilation experiments.
c. Assimilation results: Analysis
An example of the reflectivity assimilation results is presented in Fig. 8, and the results show the ensemble mean of the FG and analysis (AN) of KuNS Ze at 2.5 km and the nadir (26th) angle bin for the Kuonly experiment. Hereafter, FG indicates the ensemble mean of the FG. Overall, the FG echo was less structured and broader than the observations in Fig. 6. Assimilating the KuNS Ze sharpened the Ze structure by getting closer to the observations shown in Fig. 6. Figures 9 and 10 compare the assimilation results among the Kuonly, GMIonly, and GMI + Ku experiments. Figure 9 shows FG and AN increments (AN minus FG) for the rain mixing ratio and liquid cloud mixing ratio, vertical wind, and zonal wind. Assimilating KuPR Ze alone resulted in a detailed adjustment for rain (Fig. 9b), while use of GMI alone mostly increased rain (Fig. 9c). As for clouds, KuPR Ze produced invisibly small changes in Fig. 9f, while GMI TB significantly increased clouds in Fig. 9g. This can be explained by the observation sensitivity formulated in the observation operators and the cross-variable correlations in the background error covariance in EnVA. There is a very large (small) dependence of Ze on rain (clouds) at DPR frequencies, but TB has large sensitivity to both rain and clouds although it varies with the frequency. Second, in the background error covariance of EnVA, rain was only marginally correlated with analysis variables of clouds (Fig. 9f) and potential temperature (Fig. 9j). In contrast, there was substantial correlation between rain and vertical wind, and small correlation between rain and horizontal winds. These correlations produced obvious (small) changes in vertical wind (horizontal winds) by assimilating Ze, shown in Fig. 9n (Fig. 9r), leading to the change in structure and location of Halong. Finally, fine-structured analysis increments in the Kuonly experiment stemmed from the small horizontal lengths of the background error correlations of rain and the relatively high resolution of KuNS superobs. The GMI + Ku experiment produced intermediate corrections between the Kuonly and GMIonly experiments (Figs. 9d,h,l,p,t), as we had expected.
(a),(b) Ensemble mean of first-guess (FG) Ze and (c),(d) analyzed Ze of KuNS for the Kuonly experiment at (a),(c) 2.5-km altitude and (b),(d) the nadir bin.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
(a) First guess (FG), and (b)–(d) analysis (AN) increment (AN minus FG) for the rain mixing ratio (g kg−1) in the (b) Kuonly, (c) GMIonly, and (d) GMI + Ku experiments. (e)–(h) As in (a)–(d), but for the liquid cloud mixing ratio (g kg−1); (i)–(l) as in (a)–(d), but for potential temperature (K); (m)–(p) as in (a)–(d), but for vertical wind (m s−1); and (q)–(t) as in (a)–(d), but for zonal wind (m s−1). KuNS superobs are plotted as gray dots to facilitate the positional correspondence between observations and increments.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
(a) FG minus observation, and (b)–(d) AN minus observation of KuNS Ze (dBZ) at 2.5-km altitude for the (b) Kuonly, (c) GMIonly, and (d) GMI + Ku experiments. (e)–(h) As in (a)–(d), but for the GMI brightness temperature (TB, K) at 19V.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
The correction by the assimilation was verified based on the agreement of AN with observations in Fig. 10. Figure 10a shows positive FG differences from observed Ze in the center of Halong and negative differences on some portions of the rainbands, which are indicative of overestimations and underestimations of FG rain in the respective area. Figure 10e shows negative FG differences from observed TB at 19V in most of the observed area, thus indicating insufficient hydrometeors and humidity in the FG. Assimilating KuNS Ze obviously brought AN closer to Ze observations except for in a detached area in the north of the satellite path (Fig. 10b). However, the AN difference in TB at 19V was scarcely changed from the FG difference except for in the center region (Fig. 10f). Assimilating only TB corrected the wide area and various variables, leading to smaller AN differences in both Ze and TB (Figs. 10c and 10g). However, the AN difference in Ze was obviously positive around the center of Halong (Fig. 10c), which was probably due to the excessive increase of rain. Figures 10d and 10h show that the GMI + Ku experiment resulted in the best balanced analysis, which was sufficiently close to both Ze and TB, and that the experiment seemed to avoid the damage from excessive correction by TB assimilation.
The root-mean-square (RMS) and average (bias) of FG and AN differences from TB and Ze observations were computed for the seven experiments. Figure 11 shows that assimilating GMI reduced the RMS of TB, while assimilating Ze, either in the Kuonly or Kaonly experiment, had a negligible impact on the RMS of TB. The GMIonly experiment showed the largest reduction in RMS but generated positive bias at 37V, thus suggesting an excessive increase in clouds. Figures 12 and 13 show the RMS and bias for KuNS and KaHS calculated from samples that passed all of the QC procedures except for the QC for rejecting the ice region. Note that samples to the north of 18.5°N, which showed almost the same values in terms of the FG and AN in all the experiments, were removed to clarify the comparison among the experiments. The most notable finding was that, even if Ze was assimilated in limited range bins below the melting layer around a 5-km altitude, both the RMS and bias were significantly reduced far beyond the layer up to 14 km. Above the melting layer, FG overestimated Ze, as was already seen in the CFAD in Fig. 2, and this resulted in a large positive bias. This bias was significantly reduced by assimilating Ze due to the vertical correlation in the background error covariance. Figure 14 shows that the vertical error correlation of precipitation analysis variable (Pr) is large in rainy scene although it varies dependent on situation.
(left) Average (or bias) and (right) root-mean-square (RMS) of AN minus observed TB (K) for five channels of GMI in seven assimilation experiments. The bias and RMS of FG minus observed TB are also plotted with black lines as a reference.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
As in Fig. 11, but for KuNS Ze (dBZ) from 0.5 to 14 km.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
As in Fig. 11, but for KaHS.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Vertical error correlation of background error covariance for precipitation analysis variable at a location with (a) strong rain and (b) weak rain.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Figure 12 also shows that the GMIonly experiment excessively increased rain and slightly worsened the positive bias below 5 km. We speculate that this excessive increase was associated with the model’s overestimation of large snow as described in section 3. The overestimation yielded excessive TB depression via the scattering effect, promoting the data assimilation to increase rain (and clouds) to raise the TB. Interestingly, however, the positive bias was reduced above 9 km and it became smaller above 12 km than that in the Kuonly or Kaonly experiments. This probably resulted from the reduction of frozen hydrometeors through assimilating the TB at 89V, which is sensitive to ice at high altitudes. Although the Kuonly and Kaonly experiments still had positive impacts at such a high altitude with respect to the reduction of both bias and RMS from CNTL, the degree of improvement was smaller than that when assimilating GMI. This indicates that the effect of assimilating PR weakens at this high altitude and that, instead, the high-frequency channel of GMI has a greater impact on the hydrometeor analysis. A negative bias was noticed around 5 km, which suggests that the QC procedures may have failed to remove some samples affected by melting particles. A similar feature was observed for the fitting to KaHS Ze in Fig. 13.
For the final discussion of this section, we focus on the impacts of additional radar on the AN difference with respect to Ze. Compared to the statistics of KuNS Ze in Fig. 12, adding KaHS in the GMI + Ku experiment hardly changed the RMS and bias (cf. the GMI + Ku and GMI + KuKa experiments). In contrast, adding KuNS in GMI + Ka led to clear reductions in the RMS and bias (GMI + Ka and GMI + KuKa). The statistics of KaHS Ze in Fig. 13 show similar features whereby there was almost no impact of adding KaHS to GMI + Ku and better agreement was achieved by adding KuNS to GMI + Ka. The impact of additional radar seen in the KuNS statistics in Fig. 12 can be explained by the proposition that added observations produce a tighter fit to the observations (e.g., additional KuNS makes the analysis closer to KuNS observations). However, this hypothesis was not valid for the additional radar impact seen in the KaHS statistics in Fig. 13 (e.g., additional KaHS did NOT make the analysis in better agreement with KaHS observations). Another possible explanation is that KaHS was not fully exploited in our assimilation system and had difficulty adding new information to the analysis where KuNS was already assimilated. This was probably caused by our conservative data usage and the smaller coverage of KaPR compared to KuPR. Although KaPR is believed to give us the better information on frozen hydrometeors than KuPR, the QC procedure excluding Ze data in the ice region resulted in underuse of the advantages of KaPR.
d. Assimilation results: Forecast
In this section, we present the results of 48-h forecast runs from the analysis for the seven experiments and discuss the mechanisms that produced different forecasts for Halong. Figure 15 shows the time sequence of minimum pressure for Typhoon Halong during the seven experiments and the best track dataset given by JMA. Halong rapidly intensified after 0600 UTC 1 August 2014, but none of the experiments correctly forecasted this change. This result suggests predicting the rapid intensification of Halong requires improvement of initial conditions and/or CRM (e.g., spatial resolution and physical process).
The minimum sea level pressure (hPa) of 48-h forecasts for Typhoon Halong in the seven assimilation experiments and the best track dataset. Filled circles are plotted with 1-h intervals for the experiments and with 6-h intervals for the best track.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
The center position of Halong for the 48-h forecasts from the seven assimilation experiments and the best track are plotted in Fig. 16. Although several NWP centers made erroneous forecasts of the typhoon moving northward in their global operational deterministic and ensemble prediction systems (not shown), all of our experiments correctly forecast Halong moving westward except for during the late forecast hours. To closely compare the track forecasts, track errors are plotted in Fig. 17. The track error was defined by the distances of the center positions between forecasts and the best track. Findings from Figs. 16 and 17 can be summarized in the following four points.
The GMIonly experiment produced the worst forecast before 12-h forecast, which was caused by the initial condition where the TC center was analyzed at different locations from the other experiments and the best track. Note that the smallest error in CNTL at the initial time was brought about by DEC preprocessing.
GMI-related experiments (GMIonly, GMI + Ku, GMI + Ka, and GMI + KuKa), in contrast, forecasted smaller position errors than the CNTL and PR-only experiments at times after the 12-h forecast. This better forecast corresponded to smaller northward propagation or farther westward movement as shown in Fig. 16.
Although adding Ze made a small difference in the results when compared to CNTL [cf. CNTL and Kuonly (or Kaonly)] or GMIonly [e.g., cf. GMIonly and GMI + Ku (or GMI + Ka)], the combined use of Ze and TB produced forecasts comparable to or better than CNTL and the other experiments, on average, for the whole forecast range.
Difficulty was encountered in identifying additional impacts from KaHS when KuNS was already assimilated.
Tracks of the center positions of Halong from 48-h forecasts for the seven assimilation experiments and the best track dataset. Positions in the 12-, 24-, and 36-h forecasts are plotted with squares and others with circles. The bold line in the best track corresponds to the 48-h forecast period.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
Position errors (km) of the center of Halong as a function of the forecast hour for the seven assimilation experiments. Position errors were defined as displacement from the best track.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
These results probably can be explained as follows. For the first result, the GMIonly experiment increased clouds and rain as well as humidity and upwind over a wide area, which modified the TC structure and enhanced circulation. This relatively large adjustment resulted in displacement of the center position.
For the second result, Halong maintained the stronger circulation and generated greater amounts of precipitation until the 24- or 36-h forecast in the GMI-related experiments compared to the other experiments (Fig. 18). The enhanced precipitation was more obvious to the west and south of the TC center, as shown in Fig. 18. This was validated with an hourly analyzed rainfall dataset called the Global Satellite Mapping of Precipitation MVK version 6 (GSMaP; Kubota et al. 2007; Aonashi et al. 2009; Ushio et al. 2009). GSMaP is a blended passive microwave and infrared satellite product with a grid resolution of 0.1° and a temporal resolution of an hour. Figure 19a shows that the precipitation band intensified to the west and south of the TC center at 14.7°N, 137.7°E from best tracks. Furthermore, Figs. 19b–e show that GMI-related experiments (Figs. 19d,e) were more agreeable to GSMaP (greater number in hit category) than CNTL and Kuonly experiments (Figs. 19b,c). This could have generated vorticity to the west and south of the center through asymmetric diabatic heating and promoted Halong to move westward (Wu and Wang 2001). However, this circulation enhancement started to weaken around the 24-h forecast (not shown), and it was not enough to cause the rapid intensification.
(a) 1-h accumulated precipitation (mm h−1) at 24-h forecast for the CNTL experiment, and 1-h precipitation difference from CNTL at 24-h forecast for the (b) Kuonly, (c) Kaonly, (d) GMIonly, (e) GMI + Ku, and (f) GMI + KuKa experiments. The difference was taken at relative locations from the TC center for each experiment.
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
(a) Hourly rainfall (mm h−1) from GSMaP with 0.1° grid at 1200 UTC 1 Aug 2014. (b)–(e) Category map at 24-h forecast for (b) CNTL, (c) Kuonly, (d) GMIonly, and (e) GMI + KuKa experiments verified against GSMaP. Red, blue, green, and white pixels suggest hit (rainfall over 5 mm h−1 was forecasted and observed in GSMaP), miss (not forecasted but observed), false alarm (forecasted but not observed), and correct negative events (neither forecasted nor observed), respectively. (f) As in (e), but for assimilated ice-scattering Ze. Number of hit pixels and threat score are printed on the top of (b)–(f).
Citation: Monthly Weather Review 144, 6; 10.1175/MWR-D-15-0399.1
With respect to the third result, as we already discussed in the previous section, Ze assimilation resulted in a delicate correction in the rain and updraft primarily in the TC center region and this produced the best balanced analysis by suppressing excessive increases in clouds and rain given by the GMI. The effect of this fine adjustment was hardly sustained for the long forecast range, and this led to an indiscernible difference among the GMI-related experiments after 18-h forecasts. The effects, however, were sufficient enough to prevent the dislocation of the GMIonly experiment at the initial condition.
The fourth result was expected from the small impact of additional KaHS that we already discussed along with the analysis statistics. Because Sawada and Iwasaki (2007) suggested the importance of cold rain processes in TC development, assimilating Ze in ice regions is likely to bring further improvements to TC forecasts.
5. Summary and discussion
To improve mesoscale analyses and forecasts, we have assimilated reflectivity factor (Ze) profiles of KuPR and KaPR for DPR from the GPM Core Observatory, in addition to rain-affected radiances of GMI. We employed JMA’s nonhydrostatic model (JMA-NHM) and the ensemble-based variational (EnVA) method to handle complex precipitation processes and nonlinearity. The comparisons of DPR observations and JMA-NHM simulations with respect to Ze showed that the model overall reproduced the observations well, but significant overestimates of Ze were produced as a result of ice-particle scattering mainly due to excessive snow with large diameter particles. Based on this finding and other characteristics of the model and observations, we developed QC procedures that, for example, removed data in and above the melting layers and data that were contaminated by ground clutter. These conservative QC procedures limited the usage of DPR, but we assumed that this was a reasonable approach to take at the initial development stage described in this study.
We assessed the impacts of assimilating KuPR, KaPR, and GMI both separately and in combination. Although assimilating GMI radiance had large impacts due to its wide observation area and sensitivity to various control variables, it seemed to overcorrect the cloud and rain mixing ratios. In contrast, assimilating DPR Ze made a fine and delicate correction of the rain and updraft due to the high spatial resolution, narrow coverage, and limited sensitivity to other variables in the observation operators and background error covariance in EnVA. Despite the limited usage of DPR, the impact extended to as high as 14 km as a result of the background vertical error correlation. The combined use of GMI and KuPR (or KaPR) resulted in the best balanced analysis and forecast with respect to the agreement of observations and position errors of TC forecasts. The importance of the synergetic use of the microwave radiance and radar Ze is one of the significant findings of this study.
One advantage of DPR Ze over microwave radiance is the ability to measure vertically resolved precipitation. This advantage was exploited in the QC procedure to discriminate liquid particle regions from solid and melting regions that our model poorly represented. Since microwave radiance measures vertically integrated information, its simulation can be harmed by the model deficiencies in frozen hydrometeors. The overestimation of clouds and rain in the GMIonly experiment may occur as a counteraction against TB depression due to the exaggerated snow in the model. To further recognize the advantages of the detailed vertical resolution of DPR, it would be worthwhile to study cases such as large frontal systems where the vertical structure of precipitation plays an important role.
Further investigations are under consideration to make better use of DPR data in model evaluation and data assimilation. The excessive snow with large diameter particles in JMA-NHM was found in a previous study using a ground-based radar and space-based microwave passive imager (Eito and Aonashi 2009). They suggested that this problem was caused by deficiency in cloud microphysics and showed that the excessive snow content was alleviated by adjusting the snowfall speeds and threshold of conversion from snow to graupel. Another possibility is the assumption of inverse exponential size distribution that has a long tail in extremely large diameter bins. A global continuous dataset provided by DPR will be helpful to understand and improve the simulation of cloud microphysical processes.
As for data assimilation development, relaxing QC procedures and assimilating DPR Ze data above the melting layer are expected to have additional positive impacts on forecasts. These key approaches should be explored for better exploiting KaPR too. This will require appropriate handling of the overestimation of large snow particles or strong Ze biases in the model, for example, by applying bias correction or tuning cloud microphysical processes. As a trial, we added KuNS and KaHS Ze above the melting layer on the top of GMI + KuKa experiment without any special treatment such as bias correction. This new experiment naturally showed better fitting of AN to Ze observation above the layer (not shown). Unexpectedly forecast performance was comparable to (not degraded from) other GMI-related experiments with respect to position errors and precipitation distribution (Fig. 19f). From our experience of operational radiance assimilation development (Okamoto et al. 2005) and other previous studies (e.g., Dee 2005), however, cycled assimilation using a biased model resulted in imbalanced and biased analysis and forecast degradation including significant spinup and drift to the model climate. Thus, assimilating Ze above the melting layer affected by model bias involves more comprehensive evaluation based on cycled assimilation experiments. In addition, bias below the melting layer was not negligible, especially for KaHS Ze in Fig. 13. Thus, development of a bias correction procedure dependent on the altitude and, probably, the angle bin may be necessary. Adjusting observation errors dependent on a representative scale (Janisková 2015) and precipitation amount can also be an important target. Finally, many other meteorological cases need to be tested to validate the findings and presumptions in this study.
Acknowledgments
Thanks are given to the work of the JMA-NHM and the Joint-Simulator development group. The authors acknowledge Akihiro Hashimoto, Tempei Hashino, Munehiko Yamaguchi, Masahiro Sawada, Ahoro Adachi, Hiroshi Yamauchi, Yasutaka Ikuta, Naofumi Yoshida, and Hideaki Hase for their valuable comments and discussion. This study was partly supported by the Seventh Precipitation Measurement Mission (PMM) Japanese Research Announcement of the Japan Aerospace Exploration Agency (JAXA).
REFERENCES
Aksoy, A., D. C. David, and C. Snyder, 2010: A multicase comparative assessment of the ensemble Kalman filter for assimilation of radar observations. Part II: Short-range ensemble forecasts. Mon. Wea. Rev., 138, 1273–1292, doi:10.1175/2009MWR3086.1.
Aonashi, K., and H. Eito, 2011: Displaced ensemble variational assimilation method to incorporate microwave imager brightness temperatures into a cloud-resolving model. J. Meteor. Soc. Japan, 89, 175–194, doi:10.2151/jmsj.2011-301.
Aonashi, K., and Coauthors, 2009: GSMaP passive, microwave precipitation retrieval algorithm: Algorithm description and validation. J. Meteor. Soc. Japan, 87A, 119–136, doi:10.2151/jmsj.87A.119.
Battaglia, A., S. Tanelli, K. Mroz, and F. Tridon, 2015: Multiple scattering in observations of the GPM dual-frequency precipitation radar: Evidence and impact on retrievals. J. Geophys. Res. Atmos., 120, 4090–4101, doi:10.1002/2014JD022866.
Bauer, P., A. J. Geer, P. Lopez, and D. Salmond, 2010: Direct 4D-Var assimilation of all-sky radiances. Part I: Implementation. Quart. J. Roy. Meteor. Soc., 136, 1868–1885, doi:10.1002/qj.659.
Benedetti, A., P. Lopez, P. Bauer, and E. Moreau, 2005: Experimental use of TRMM precipitation radar observations in 1D+4D-Var assimilation. Quart. J. Roy. Meteor. Soc., 131, 2473–2495, doi:10.1256/qj.04.89.
Buehner, M., 2012: Evaluation of a spatial/spectral covariance localization approach for atmospheric data assimilation. Mon. Wea. Rev., 140, 617–636, doi:10.1175/MWR-D-10-05052.1.
Dee, D. P., 2005: Bias and data assimilation. Quart. J. Roy. Meteor. Soc., 131, 3323–3343, doi:10.1256/qj.05.137.
Eito, H., and K. Aonashi, 2009: Verification of hydrometeor properties simulated by a cloud-resolving model using a passive microwave satellite and ground-based radar observations for a rainfall system associated with the Baiu front. J. Meteor. Soc. Japan, 87A, 425–446, doi:10.2151/jmsj.87A.425.
Geer, A. J., P. Bauer, and P. Lopez, 2010: Direct 4D-Var assimilation of all-sky radiances. Part II: Assessment. Quart. J. Roy. Meteor. Soc., 136, 1886–1905, doi:10.1002/qj.681.
Hashino, T., M. Satoh, Y. Hagihara, T. Kubota, T. Matsui, T. Nasuno, and H. Okamoto, 2013: Evaluating cloud microphysics from NICAM against CloudSat and CALIPSO. J. Geophys. Res. Atmos., 118, 7273–7292, doi:10.1002/jgrd.50564.
Hou, A. H., and Coauthors, 2014: The Global Precipitation Measurement Mission. Bull. Amer. Meteor. Soc., 95, 701–722, doi:10.1175/BAMS-D-13-00164.1.
Ikawa, M., and K. Saito, 1991: Description of a non-hydrostatic model developed at the Forecast Research Department of the MRI. MRI Tech. Rep. 28, 238 pp.
Ikuta, Y., and Y. Honda, 2011: Development of 1D+4DVAR data assimilation of radar reflectivity in JNoVA. CAS/JSC WGNE Res. Activ. Atmos. Oceanic Modell., 41, 01.09–01.10.
Janisková, M., 2015: Assimilation of cloud information from space-borne radar and lidar: Experimental study using a 1D+4D-Var technique. Quart. J. Roy. Meteor. Soc., 141, 2708–2725, doi:10.1002/qj.2558.
Janisková, M., P. Lopez, and P. Bauer, 2012: Experimental 1D+4D-Var assimilation of CloudSat observations. Quart. J. Roy. Meteor. Soc., 138, 1196–1220, doi:10.1002/qj.988.
JMA, 2013: Outline of the operational numerical weather prediction at the Japan Meteorological Agency. Appendix to WMO technical progress report on the global data processing and forecast system (GDPFS) and numerical weather prediction (NWP), 9–40.
Kain, J., and J. Fritsch, 1993: Convective parameterization for mesoscale models: The Kain-Fritsch scheme. The Representation of Cumulus Convection in Numerical Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 165–170.
Kawabata, T., T. Kuroda, H. Seko, and K. Saito, 2011: A cloud-resolving 4DVAR assimilation experiment for a local heavy rainfall event in the Tokyo metropolitan area. Mon. Wea. Rev., 139, 1911–1931, doi:10.1175/2011MWR3428.1.
Kotsuki, S., K. Terasaki, and T. Miyoshi, 2014: GPM/DPR precipitation compared with a 3.5-km-resolution NICAM simulation. SOLA, 10, 204–209, doi:10.2151/sola.2014-043.
Kubota, T., and Coauthors, 2007: Global precipitation map using satelliteborne microwave radiometers by the GSMaP project: Production and validation. IEEE Trans. Geosci. Remote Sens., 45, 2259–2275, doi:10.1109/TGRS.2007.895337.
Kubota, T., and Coauthors, 2014: Evaluation of precipitation estimates by at-launch codes of GPM/DPR algorithms using synthetic data from TRMM/PR observations. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., 7, 3931–3944, doi:10.1109/JSTARS.2014.2320960.
Kubota, T., T. Iguchi, M. Kojima, L. Liao, T. Masaki, H. Hanado, and R. Oki, 2016: A statistical method for reducing sidelobe clutter for the Ku-band precipitation radar on board the GPM Core Observatory. J. Atmos. Oceanic Technol., doi:10.1175/JTECH-D-15-0202.1, in press.
Lin, Y. H., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22, 1065–1092, doi:10.1175/1520-0450(1983)022<1065:BPOTSF>2.0.CO;2.
Liu, G., 2004: Approximation of single scattering properties of ice and snow particles for high microwave frequencies. J. Atmos. Sci., 61, 2441–2456, doi:10.1175/1520-0469(2004)061<2441:AOSSPO>2.0.CO;2.
Martinet, P., N. Fourrié, V. Guidard, F. Rabier, T. Montmerle, and P. Brunel, 2013: Towards the use of microphysical variables for the assimilation of cloud-affected infrared radiances. Quart. J. Roy. Meteor. Soc., 139, 1402–1416, doi:10.1002/qj.2046.
Masunaga, H., and C. D. Kummerow, 2005: Combined radar and radiometer analysis of precipitation profiles for a parametric retrieval algorithm. J. Atmos. Oceanic Technol., 22, 909–929, doi:10.1175/JTECH1751.1.
McNally, A. P., 2009: The direct assimilation of cloud-affected satellite infrared radiances in the ECMWF 4D-Var. Quart. J. Roy. Meteor. Soc., 135, 1214–1229, doi:10.1002/qj.426.
Okamoto, K., 2013: Assimilation of overcast cloudy infrared radiances of the geostationary MTSAT-1R imager. Quart. J. Roy. Meteor. Soc., 139, 715–730, doi:10.1002/qj.1994.
Okamoto, K., M. Kazumori, and H. Owada, 2005: The assimilation of ATOVS radiances in the JMA global analysis system. J. Meteor. Soc. Japan, 83, 201–217, doi:10.2151/jmsj.83.201.
Saito, K., and Coauthors, 2006: The operational JMA nonhydrostatic model. Mon. Wea. Rev., 134, 1266–1298, doi:10.1175/MWR3120.1.
Sawada, M., and T. Iwasaki, 2007: Impacts of ice phase processes on tropical cyclone development. J. Meteor. Soc. Japan, 85, 479–494, doi:10.2151/jmsj.85.479.
Seto, S., T. Iguchi, and T. Oki, 2013: The basic performance of a precipitation retrieval algorithm for the Global Precipitation Measurement Mission’s single/dual-frequency radar measurements. IEEE Trans. Geosci. Remote Sens., 51, 5239–5251, doi:10.1109/TGRS.2012.2231686.
Stengel, M., M. Lindskog, P. Undén, and N. Gustafsson, 2013: The impact of cloud-affected IR radiances on forecast accuracy of a limited-area NWP model. Quart. J. Roy. Meteor. Soc., 139, 2081–2096, doi:10.1002/qj.2102.
Toyoshima, K., H. Masunaga, and F. A. Furuzawa, 2015: Early evaluation of Ku- and Ka-band sensitivities for the Global Precipitation Measurement (GPM) Dual-frequency Precipitation Radar (DPR). SOLA, 11, 14–17, doi:10.2151/sola.2015-004.
Ushio, T., and Coauthors, 2009: A Kalman filter approach to the Global Satellite Mapping of Precipitation (GSMaP) from combined passive microwave and infrared radiometric data. J. Meteor. Soc. Japan, 87A, 137–151, doi:10.2151/jmsj.87A.137.
Wattrelot, E., O. Caumont, and J.-F. Mahfouf, 2014: Operational implementation of the 1D+3D-Var assimilation method of radar reflectivity data in the AROME model. Mon. Wea. Rev., 142, 1852–1873, doi:10.1175/MWR-D-13-00230.1.
Wu, L., and B. Wang, 2001: Effects of convective heating on movement and vertical coupling of tropical cyclones: A numerical study. J. Atmos. Sci., 58, 3639–3649, doi:10.1175/1520-0469(2001)058<3639:EOCHOM>2.0.CO;2.
