Mesoscale Data Assimilation for a Local Severe Rainfall Event with the NHM–LETKF System

Masaru Kunii Forecast Research Department, Meteorological Research Institute, Tsukuba, Japan

Search for other papers by Masaru Kunii in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

This study seeks to improve forecasts of local severe weather events through data assimilation and ensemble forecasting approaches using the local ensemble transform Kalman filter (LETKF) implemented with the Japan Meteorological Agency’s nonhydrostatic model (NHM). The newly developed NHM–LETKF contains an adaptive inflation scheme and a spatial covariance localization scheme with physical distance, and it permits a one-way nested analysis in which a finer-resolution LETKF is conducted by using the output of an outer model. These new features enhance the potential of the LETKF for convective-scale events. The NHM–LETKF was applied to a local severe rainfall event in Japan during 2012. Comparison of the root-mean-square errors between the model first guess and analysis showed that the system assimilated observations appropriately. Analysis ensemble spreads indicated a significant increase around the time torrential rainfall occurred, implying an increase in the uncertainty of environmental fields. Forecasts initialized with LETKF analyses successfully captured intense rainfalls, suggesting that the system could work effectively for local severe weather events. Investigation of probabilistic forecasts by ensemble forecasting indicated that this could become a reliable data source for decision making in the future. A one-way nested data assimilation scheme was also tested. The results demonstrated that assimilation with a finer-resolution model improved the precipitation forecasting of local severe weather conditions.

Corresponding author address: Masaru Kunii, Forecast Research Dept., Meteorological Research Institute, 1-1 Nagamine, Tsukuba, Ibaraki 305-0052, Japan. E-mail: mkunii@mri-jma.go.jp

Abstract

This study seeks to improve forecasts of local severe weather events through data assimilation and ensemble forecasting approaches using the local ensemble transform Kalman filter (LETKF) implemented with the Japan Meteorological Agency’s nonhydrostatic model (NHM). The newly developed NHM–LETKF contains an adaptive inflation scheme and a spatial covariance localization scheme with physical distance, and it permits a one-way nested analysis in which a finer-resolution LETKF is conducted by using the output of an outer model. These new features enhance the potential of the LETKF for convective-scale events. The NHM–LETKF was applied to a local severe rainfall event in Japan during 2012. Comparison of the root-mean-square errors between the model first guess and analysis showed that the system assimilated observations appropriately. Analysis ensemble spreads indicated a significant increase around the time torrential rainfall occurred, implying an increase in the uncertainty of environmental fields. Forecasts initialized with LETKF analyses successfully captured intense rainfalls, suggesting that the system could work effectively for local severe weather events. Investigation of probabilistic forecasts by ensemble forecasting indicated that this could become a reliable data source for decision making in the future. A one-way nested data assimilation scheme was also tested. The results demonstrated that assimilation with a finer-resolution model improved the precipitation forecasting of local severe weather conditions.

Corresponding author address: Masaru Kunii, Forecast Research Dept., Meteorological Research Institute, 1-1 Nagamine, Tsukuba, Ibaraki 305-0052, Japan. E-mail: mkunii@mri-jma.go.jp

1. Introduction

Accurate weather prediction is essential for preventing and mitigating natural disasters, such as tropical cyclones, tornadoes, and local heavy rainfalls. Accurate simulations of these severe phenomena require numerical weather prediction (NWP) models with high resolution and sophisticated physical processes. Another key tool for improving forecasts by NWP models is creating realistic initial conditions through data assimilation methods. Data assimilation techniques couple numerical model outputs with observations so as to obtain the most probable state by considering the accuracy of the two independent estimates.

Current data assimilation methods are based on one of two approaches: variational techniques and ensemble Kalman filter (EnKF) techniques. Variational approaches are used operationally at many NWP centers for assimilating meteorological observations. The four-dimensional variational data assimilation (4DVAR) approach uses the tangent-linear and adjoint versions of forecast models and estimates four-dimensional optimal atmospheric states, fitting the observations over a specified time window of finite length. The results are used for initializing deterministic model forecasts as well as atmospheric analyses.

The EnKF is an approximation of the original Kalman filter approach (Kalman 1960) that runs multiple model forecasts simultaneously with different initial conditions and considers each model output as a statistical sample. EnKF approaches have been widely utilized for data assimilation since this use was originally proposed by Evensen (1994). The advantage of the EnKF method is that it involves a complementary relationship between data assimilation and ensemble forecasting; that is, more realistic analysis increments can be produced in the data assimilation step through its flow-dependent background error estimates from ensemble forecasting, and then ensemble forecasting can be performed by reflecting probabilistic information on analysis errors. These features meet the goal of mitigating natural disasters, especially for extreme events where flow-dependent background errors become essential. In addition, its simplicity of implementation without the use of tangent-linear and adjoint versions of models is appealing.

The local ensemble transform Kalman filter (LETKF; Hunt et al. 2007) is an EnKF scheme in which the algorithm achieves a high efficiency for parallel implementations by taking advantage of independent local analyses of the local ensemble Kalman filter (Ott et al. 2004). Miyoshi and Yamane (2007) attained a computational parallelizing ratio as large as 99.99% by running 80- and 160-member LETKFs. The LETKF scheme has been applied to atmospheric models (e.g., Miyoshi and Aranami 2006; Miyoshi and Yamane 2007; Schutgens et al. 2010; Kang et al. 2011; Miyoshi and Kunii 2012), ocean models (e.g., Shchepetkin and McWilliams 2005; Penny 2011), and even a Martian atmosphere model (Hoffman et al. 2010). These studies produced promising results for numerical simulations, and the method continues to be actively developed.

In this study, LETKF was applied to the nonhydrostatic model (NHM; Saito et al. 2006, 2007; Saito 2012) of the Japan Meteorological Agency (JMA). Miyoshi and Aranami (2006) first implemented the LETKF with the NHM, but the present study used the latest version of the LETKF, which includes the adaptive inflation scheme (Miyoshi 2011) and the spatial covariance localization scheme with physical distance (Miyoshi et al. 2007). In addition, the latest NHM–LETKF version facilitates data assimilation experiments with one-way grid nesting, in which the first guess from a coarser-resolution model is used as a boundary condition for integration of a finer-resolution model. These innovations enhance the potential of the LETKF for simulating convective-scale events. The performance of the NHM as a mesoscale ensemble prediction model has been demonstrated by Kunii et al. (2011).

The NHM–LETKF approach was applied to a heavy rainfall event that occurred in Japan during 2012, and the performance of the system was investigated by running data assimilation as well as ensemble forecast experiments. Since this is a regional forecast model, the influence of boundary perturbations on the LETKF analyses was also examined. This paper is organized as follows. Section 2 describes the NHM–LETKF system. Section 3 presents the experiment setting, and section 4 presents the results. Section 5 offers a summary and the conclusions.

2. The NHM–LETKF system

In contrast to other EnKF schemes, the LETKF is characterized by the independent update of the analysis at each grid point by the simultaneous assimilation of observations within a certain distance from the grid point. An upgrade of the local ensemble Kalman filter, the LETKF does its analysis in the subspace spanned by the ensemble without using singular-value decomposition. This leads to higher parallel efficiency compared with other methods and suits the method for practical application with various numerical models. Once typically applied to global models, the LETKF has recently been implemented within mesoscale regional models.

Miyoshi and Kunii (2012) developed the LETKF with the Weather Research and Forecasting (WRF) Model (Skamarock et al. 2005) and conducted data assimilation experiments for Typhoon Sinlaku of September 2008. The results were promising, with a good overall level of correspondence between the WRF–LETKF analyses and the observations. Kunii and Miyoshi (2012) used the WRF–LETKF model to examine the influence of sea surface temperature (SST) perturbations on analyses. The SST perturbations within EnKF generally improved the analyses and their subsequent forecasts, suggesting the importance of considering SST perturbations within ensemble-based data assimilation. The WRF-LETKF model has also been used for calculating the sensitivity of observations (Kunii et al. 2012).

Subsequently, the LETKF has been implemented within the NHM, sharing numerous common features with the WRF–LETKF model. Critical differences between the WRF–LETKF and NHM–LETKF are in model-dependent elements such as interfaces and analyzed variables. This leads to efficiencies in maintaining both systems, as feedback from WRF–LETKF users can be used to update the NHM–LETKF system.

The NHM–LETKF has an option for performing data assimilation with a one-way nested model. A nested analysis runs independently as a separate cycle using a model of finer resolution that covers a portion of an outer model domain. The output of the inner model does not feed back to the coarser-model solution. This significantly reduces computational costs compared with data assimilation over a whole domain at high resolution. In the nested analysis, lateral boundary conditions are supplied by first-guess fields from the outer model, and analyses are carried forward to the next cycle. In the nested inner model, the LETKF can be used with localization scales and data assimilation intervals that differ from those of the outer model to accommodate higher-resolution configurations.

Quality-controlled observational data used for the JMA operational mesoscale analysis based on the JMA’s nonhydrostatic model–based variational data assimilation approach (JNoVA; Honda et al. 2005) were assimilated in both the outer and inner LETKF domains for simplicity. Because the quality-control processes include a gross error check that excludes data with large differences from guess forecasts, the remaining data may not be optimal for the NHM–LETKF. However, it can be assumed that forecasts initialized on the basis of analyses by NHM–LETKF and JNoVA do not differ significantly; therefore, the quality-control criteria would still apply to the NHM–LETKF.

3. Experiment design

a. Heavy rain event of 2012

In July 2012, torrential rainfall fell on the Japanese island of Kyushu. Figure 1 presents the surface weather map for Japan at 1800 UTC 11 July 2012. A significant subtropical frontal zone (a baiu front) was located north of Kyushu. Moist southwesterly airflow occurred on the southern side of the baiu front from 11 to 14 July, causing torrential rainfall over Kyushu. At Aso city in Kumamoto Prefecture, record-setting rainfall exceeded 100 mm h−1 and 500 mm in a 24–h period. Flooding and debris flows killed more than 20 people and caused widespread landslides and inundation damage.

Fig. 1.
Fig. 1.

Surface weather map at 1800 UTC 11 Jul 2012 analyzed by JMA.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

Figure 2 plots a time series of the accumulated rainfall averaged over the Kyushu area (31°–34°N, 129°–132°E). The amounts were calculated from the JMA’s Radar/Rain Gauge–Analyzed Precipitation (RAP) data interpolated onto 5-km grids. From 11 to 14 July 2012, there were three peaks of precipitation, each exceeding 10 mm (3 h)−1 period as regional averages. The first peak continued for about 21 h as the heavy precipitation area moved southward over Kyushu. Although the second and third peaks had shorter durations, each peak value was comparable to that of the first peak, and each one made the disaster more serious.

Fig. 2.
Fig. 2.

Time series of 3-h accumulated rainfall (mm) averaged over the study region, estimated from the 5-km interpolated rainfall amount.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

b. LETKF analysis and ensemble forecast

This study applied the NHM–LETKF approach to the heavy rain event of July 2012. First, the NHM–LETKF cycles were carried out in an outer domain covering all of Japan (D01 in Fig. 3). For this exercise the NHM consisted of 241 × 193 grid points with a horizontal spacing of 15 km and 50 vertical levels reaching 22 km in the hybrid-vertical coordinate. This grid configuration is similar to that of the JNoVA, which aids comparison between results initialized with the JNoVA and the LETKF methods. The NHM contains a modified Kain–Fritsch convective parameterization scheme (Ohmori and Yamada 2006) along with three ice bulk cloud microphysics schemes (Ikawa and Saito 1991). The Mellor and Yamada level 3 closure model (Nakanishi and Niino 2004, 2006) was implemented for the turbulence scheme. The LETKF analyzes all prognostic variables: three-dimensional wind components u, υ, and w; temperature T; pressure p; water vapor mixing ratio qυ; and water–ice microphysics variables. The LETKF system employs an adaptive inflation scheme (Miyoshi 2011) that estimates the multiplicative inflation factors adaptively at each grid point. Miyoshi and Kunii (2012) investigated the advantage of this method over fixed multiplicative inflation.

Fig. 3.
Fig. 3.

Model domains for the NHM–LETKF analysis (D01, 15-km grid spacing) and the one-way nested analysis (D02, 5-km grid spacing) with elevation (m).

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

The observations included data from radiosondes, surface stations, pilot balloons, wind profilers, aircraft, ships and buoys, and satellite-based winds. The satellite radiances and the RAP data used in the JNoVA analysis were not assimilated because the NHM–LETKF system did not include operators for these observations. Here, a spatial covariance localization scheme was used with observations assimilated with localization parameters of 200 km in the horizontal, 0.2 ln p coordinates in the vertical, and 3 h in time. The localization parameters correspond to the one-sigma length at which the Gaussian localization function becomes e−0.5. These parameters were tuned through a trial-and-error procedure. The assimilation cycle was successively operated with a 6-h time interval, in which observation data were collected into 1-h time slots and assimilated using the four-dimensional LETKF system (Hunt et al. 2004).

The outer NHM–LETKF model was initialized at 0600 UTC 1 July with 50 ensemble members, more than 10 days before the rainfall event. After the NHM–LETKF cycles, subsequent ensemble forecasts were performed by using each LETKF analysis output as the initial conditions; that is, each member of the ensemble forecast was initialized with the LETKF perturbations. The NHM configurations for the ensemble forecasts were similar to those in the data assimilation cycles, but with a finer horizontal resolution of 5 km.

c. One-way nested LETKF analysis

For quantitative simulation of the severe rainfall event, one-way nested data assimilation cycles were executed with a higher resolution of 5 km beginning at 0000 UTC 5 July 2012. The data assimilation cycle in the nested experiment was shorter than that in the outer LETKF simulations, because computational time was saved for spinning up the ensemble-based error covariance by using LETKF perturbations in the outer model as initial estimates in the nested cycles. The domain covered the Kyushu area (D02 in Fig. 3). In the nested system, LETKF analyses in each cycle were carried to the next cycle as initial conditions, and lateral boundary conditions were supplied from guess outputs of the outer cycles. Because the solution of the nested analysis did not need to feed back to the outer model, the nested cycles could be carried out independently of the outer cycles.

In the nested cycles, additional observations such as Doppler radar data could potentially be effective, but in this study the same observations were assimilated in both the inner and the outer cycles so as to focus on the impact of the finer-resolution data assimilation on the simulation of the heavy rainfall event. The assimilation time interval was made shorter at 3 h to maintain the tangent linear assumption used in the four-dimensional LETKF model. Because tighter localization may be more suitable in the higher-resolution analysis, a horizontal localization parameter of 100 km was implemented in the nested LETKF. As with the outer model, subsequent model forecasts initialized with the LETKF analyses were performed using the NHM, but with a horizontal grid interval of 1 km. Here, almost the same configurations were used as for the 5-km NHM run, but the convective parameterization scheme was not activated. The specifications of the outer and nested LETKF experiments are summarized in Table 1.

Table 1.

Specifications of the outer and nested LETKF experiments.

Table 1.

4. Results

a. Lateral boundary perturbations

Miyoshi and Kunii (2012) found that the influence of unperturbed lateral boundary conditions was not critical for analyses and model forecasts for Typhoon Sinlaku. However, it is doubtful whether that would be true in this study because the domain of the NHM was rather smaller than that of Miyoshi and Kunii (2012) and was in the midlatitudes rather than the tropics. Saito et al. (2012) found that the effect of lateral boundary perturbations on ensemble forecasts was to prevent underestimates near the lateral boundaries and to improve the root-mean-square errors (RMSEs) in ensemble forecasting. Preliminary experiments were done with and without lateral boundary perturbations in this study to estimate their influence on the LETKF analysis. Following Saito et al. (2012), lateral boundary perturbations were derived from the JMA operational 1-week ensemble prediction system by subtracting the ensemble mean from each ensemble member and then added to the unperturbed boundary conditions. The following verifications are based on results of the outer NHM–LETKF cycles with a horizontal resolution of 15 km.

Figures 4a and 4b depict the ensemble spreads of zonal wind at the 500-hPa level without and with lateral boundary perturbations. When lateral boundary perturbations were not considered in LETKF cycles, the ensemble spread near the lateral boundaries became too small, and this underestimation traveled in the downwind direction (Fig. 4a). In cycles with lateral boundary perturbations, the ensemble spread increased near the boundaries as expected (Fig. 4b). The lack of ensemble spread near the lateral boundaries made the LETKF overconfident in its guess forecast, leading to universally low analysis increments in these regions (Fig. 4c). In the simulation that included lateral boundary perturbations, the effect of the added uncertainty was to cause the analysis to spread the information from the observations in a more realistic manner (Fig. 4d).

Fig. 4.
Fig. 4.

Ensemble spreads of zonal wind at the 500-hPa level in the outer LETKF (a) without and (b) with lateral boundary perturbations (m s−1). (c),(d) As in (a),(b), but for the analysis increment.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

Statistical verification based on measures such as bias and RMSE was conducted to examine the overall impact of perturbed lateral boundaries in the NHM–LETKF system. Figure 5 presents the verifications of 6-h forecasts relative to radiosonde observations averaged over 10 days (0000 UTC 5 July–0000 UTC 15 July). Because the perturbations of the JMA 1-week ensemble prediction system were generated by singular vectors, some inconsistency between the inner and boundary perturbations might have been expected. However, the RMSEs for all variables showed a consistent advantage from perturbing the lateral boundary, especially for the zonal wind at the middle and upper levels. This result suggests that the small spread near the boundaries seen in Fig. 4a does not indicate small forecast errors, but rather the underestimation of the spread resulting from using identical lateral boundaries in all ensemble members. Insufficient ensemble spread generally leads to systematic underestimation of the observation weight. Therefore, perturbing lateral boundaries, even with other perturbation schemes, should improve the LETKF application with regional models. Based on these results, lateral boundary perturbations were introduced in the following experiments.

Fig. 5.
Fig. 5.

Verification of 6-h forecasts relative to radiosonde observations for the (a) zonal wind component (m s−1), (b) temperature (K), and (c) relative humidity (%), averaged over 10 days (from 0000 UTC 5 Jul to 0000 UTC 15 Jul 2012) in the outer LETKF cycles. Solid (dashed) lines correspond to experiments with (without) lateral boundary perturbations (LBP and CTL). Thick (thin) lines indicate RMSEs (biases).

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

b. Evaluation of analysis fields

To evaluate the general performance of the NHM–LETKF system, Fig. 6 plots the analysis and 6-h forecast RMSEs for the zonal wind component and relative humidity at 850 hPa in the outer cycles. As this verification measure relies on radiosonde data, it does not specifically represent the forecast errors over the whole domain, but instead represents the forecast errors mostly over land, where most radiosonde data were collected. The RMSEs of the guess fields were consistently reduced in the corresponding analyses, indicating that the LETKF assimilated observations appropriately. Figure 6 demonstrates the stability of the system by showing that the RMSEs did not steadily increase or decrease during the experiment period.

Fig. 6.
Fig. 6.

Time series of 6-h forecast (GUES) and analysis (ANAL) RMSEs relative to radiosonde observations for the (a) zonal wind component and (b) relative humidity at 850 hPa in the outer LETKF cycles.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

Previous studies have pointed out that the warm and moist southwesterly wind was a key factor in increasing the precipitation near the baiu front (e.g., Davidson et al. 1998; Ninomiya 2000). Figure 7 presents analysis ensemble spreads for the zonal wind component and relative humidity at 850 hPa, averaged over the entire model domain and over the southwest portion of Kyushu (27.5°–35.0°N, 125.0°–132.5°E) in the outer LETKF cycles. For the zonal wind component, both ensemble spreads fluctuated during the experimental period, but the spread averaged over the Kyushu district increased around 1200 UTC 11 July 2012 (Fig. 7a). A similar feature can be seen in relative humidity (Fig. 7b), with a rapid increase in the ensemble spread around the same time. Since the ensemble spread variability is positively correlated with variations in skill predictability (e.g., Whitaker and Loughe 1998; Scherrer et al. 2004), these spread surges may represent an increase in the uncertainty of environmental fields corresponding to the occurrence of heavy rainfall (Fig. 2). In this case, the large ensemble spread occurred mainly over the ocean, where observation data are generally scarce. Recent studies have suggested that satellite radiance and Global Positioning System radio occultation data are significant in improving model forecasts (Zhu and Gelaro 2008; Cardinali 2009; Gelaro and Zhu 2009). Since these satellite-based observations are available even over the ocean, assimilating these data could help reduce forecast uncertainty over the ocean and further improve the quantitative model forecast.

Fig. 7.
Fig. 7.

Time series of analysis ensemble spreads for the (a) zonal wind component and (b) relative humidity at 850 hPa, averaged over the entire model domain (ALL) and over the southwestern portion of the Kyushu district (LMT, 27.5°–35.0°N, 125.0°–132.5°E).

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

c. Forecast verifications

General verification measures were tried first to ascertain the overall impact of the NHM–LETKF analyses on single deterministic runs from the ensemble-mean analyses. The subsequent model forecasts were conducted using NHM with a horizontal resolution of 5 km, and the results were compared to forecasts initialized with the JNoVA analyses, which approximates the JMA operational mesoscale forecast. Both experiments used the same configurations except for the initial conditions.

For the verification, the threat score (TS) and bias score (BS) were evaluated by comparing precipitation forecasts with the RAP data. These scores are defined as
e1
e2
where F is the number of points where the predicted precipitation exceeds a certain threshold, O is the number of points where the observed precipitation exceeds a threshold, and H is the number of successful predictions. The threat score ranges from 0 to 1, with 1 signifying a perfect forecast.

Threat scores and bias scores for accumulated rainfall over 3 h at the 12-h forecast time, averaged over 13 initial times from 1200 UTC 10 July to 1200 UTC 13 July, are depicted in Figs. 8a and 8b. These scores were calculated with the 5-km NHM forecast outputs. For weak rains of less than 5 mm (3 h)−1, the scores of JNoVA were substantially better than those of LETKF, whereas the LETKF scores were superior for moderate and intense rainfalls. In particular, the bias score of LETKF showed essentially no decay even for intense rainfalls. These characteristics were also observed in forecasts with a longer lead time (Figs. 8c,d).

Fig. 8.
Fig. 8.

(a) TS and (b) BS for 3-h accumulated rainfall in the 12-h NHM forecasts with 5-km horizontal resolution initialized with the JNoVA (dashed lines) and LETKF (solid lines) analyses, averaged over 13 initial times from 1200 UTC 10 Jul to 1200 UTC 13 Jul. (c),(d) As in (a),(b), but for the 24-h forecasts.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

In contrast to the NHM–LETKF approach, the JNoVA analysis assimilates satellite radiance and RAP data. These observations may contribute to the amelioration of moisture fields in the model, leading to better threat and bias scores for weak precipitation events. However, scores for heavy rainfalls indicate the general superiority of the NHM–LETKF system. This result implies that the EnKF schemes providing flow-dependent analysis increments are promising approaches to the simulation of intense precipitation, and that the assumptions for the static background errors used in the 4DVAR may not hold well under severe weather conditions.

Focusing on the forecasts for the intense rainfall case, Fig. 9 presents the 3-h accumulated rainfalls on 0000 UTC 12 July 2012, when heavy rainfall was observed. Comparing the 18-h forecast results initialized at 0600 UTC 11 July 2012 with the RAP observations (Fig. 9c) shows that the forecast initialized with JNoVA failed to capture the heavy rainfall in the middle of Kyushu (Fig. 9a), indicating that generating accurate forecasts for local severe weather remains challenging. A deterministic forecast with the LETKF analysis (Fig. 9b) successfully captured the linear rainband extending from southwest to northeast over the Kyushu area in the 18-h model forecast. Also, the rainfall over other regions was well reproduced compared with the observations (not shown). Whereas JNoVA assimilated more observations, including the satellite radiance and analyzed rainfall amount, the forecast initialized with the LETKF analyses provided better results in this case.

Fig. 9.
Fig. 9.

Simulated mean sea level pressure (hPa; solid line) and 3-h accumulated precipitation (mm; color) in 18-h NHM forecasts with a horizontal resolution of 5 km at 0000 UTC 12 Jul 2012, initialized with the (a) JNoVA and (b) LETKF analyses. (c) JMA RAP observations from this time period.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

In addition to this deterministic forecast, ensemble forecasts were carried out by using LETKF analysis outputs as initial conditions. By using ensemble outputs, the probability of precipitation (POP) can be estimated at each grid point. Figure 10 illustrates the POP over the Kyushu district above a threshold of 50 mm (3 h)−1. Figure 10d corresponds to the deterministic forecast of Fig. 9b with the same initial times, and exhibits remarkable consistency with the observations (Fig. 9c). It should be noted that at the 18-h forecast, the maximum POP was as great as 50% over the area where torrential rain occurred.

Fig. 10.
Fig. 10.

Forecasts of the POP exceeding 50 mm (3 h)−1 as of 0000 UTC 12 Jul 2012, calculated from the 5-km NHM ensemble outputs, with lead times of (a) 36, (b) 30, (c) 24, (d) 18, (e) 12, and (f) 6 h.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

Providing timely, reliable information is important for disaster mitigation. Figure 10 also depicts the length of lead time needed to capture a rainfall event in ensemble forecasting. It is obvious that the reliability of POP estimates increases as lead times shorten to an 18-h forecast, resulting in POP distributions over Kyushu with stronger peaks and narrower geographic extent. Although the difference in the POP in the 12- and 6-h forecasts is not obvious in Kyushu, precipitation over the Shikoku region to the east is slightly better captured with shorter lead times. In this case, a forecast of a very high POP within a lead time of 24 h can be considered reliable for sites subsequently affected by catastrophic flooding. A lead time of 24 h may be enough to transfer information and issue warnings for local disaster responders. Although the probabilistic information from these forecasts needs further refinement for practical use, this trial suggests that such forecasts could become a reliable data source for future decision making.

d. One-way nested LETKF

The one-way nesting option in the NHM–LETKF system can make more efficient use of computer resources by conducting high-resolution data assimilation over limited regions. This advance can be expected to improve precipitation forecasting by enabling the use of higher-resolution models and more sophisticated physical processes to simulate local severe events. In the nested analysis, lateral boundaries that are supplied from outer cycles allow consistent perturbations of lateral boundaries. In addition, it should be possible to use observations more effectively by assimilating high-density data without reducing them to a coarser grid. However, in this study the same observations were assimilated in both the outer and nested analyses in order to isolate the impact of the one-way nested LETKF on model forecasts.

To assess the impact of the one-way nested analysis, the 1-km NHM outputs initialized with the outer and the nested LETKF analyses were compared. The DA15 experiment used a simple downscaled forecast, conducting the 1-km NHM forecast using the outer 15-km LETKF analyses as initial conditions. The DA05 experiment used nested LETKF analyses with a horizontal resolution of 5 km that were then used as initial conditions for the 1-km NHM forecast. As the forecasts in both experiments had the same lateral boundary conditions, the impact of different initial conditions could be estimated.

Figure 11 depicts 6-h forecasts initialized with each analysis. In both experiments, the 1-km NHM forecasts successfully captured the rainband with good correspondence to the observations. Because environmental fields were corrected by the outer LETKF method, the difference in the location of the linear rainband was rather small. However, DA05 was closer than DA15 to the observations in its prediction of rainfall amount, differing in 3-h accumulated rainfall by as much as 50 mm. Threat scores for 3-h accumulated rainfall, averaged between 3 and 15 h, were substantially better for DA05 than for DA15 (Fig. 12a), showing the advantage of the one-way nested LETKF for local severe weather.

Fig. 11.
Fig. 11.

Simulated mean sea level pressure (hPa) and 3-h accumulated precipitation (mm) in 18-h NHM forecasts with a horizontal resolution of 1 km, initialized with the (a) outer and (b) nested LETKF approaches.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

The precipitation forecasts from 1-km NHM were verified by comparing the forecast outputs with the RAP data. Figures 12b and 12c show the accumulated precipitation amounts from the RAP data along with the forecasts from the DA15 and DA05 experiments at two locations where intense rainfall occurred during the forecast period. At one location (Fig. 12b), the accumulated precipitation from DA05 exceeded 150 mm in the 15-h forecast and was closer to the observations than DA15, although the amount was still underestimated. At the other location (Fig. 12c), DA05 successfully captured intense rainfall of over 200 mm in 15 h whereas the DA15 forecast was 150 mm.

Fig. 12.
Fig. 12.

(a) Threat scores for 3-h accumulated rainfall averaged between 3 and 15 h in the DA15 and DA05 experiments. Accumulated precipitation in the RAP data (OBS) and the DA15 and DA05 experiments at (b) 32.5°N, 130.6°E and (c) 32.5°N, 130.8°E, respectively. The initial time of the forecast is 1800 UTC 11 Jul 2012.

Citation: Weather and Forecasting 29, 5; 10.1175/WAF-D-13-00032.1

High-resolution data assimilation over the whole domain would be the preferable choice for predicting local severe events, but the results obtained here indicate that the nested LETKF scheme can improve model forecasts with computational efficiency and could enhance practical convective-scale forecasts. However, there remain difficulties in predicting local intense rainfalls with both temporal and spatial accuracy (Fig. 12b). Assimilating high-resolution observations such as Doppler radar and satellite-based rapid-scan winds could further improve convective-scale forecasts and would be an attractive area of future research.

5. Summary and conclusions

This study implemented the LETKF scheme within the JMA mesoscale model, NHM. The newly developed NHM–LETKF approach contains an adaptive inflation scheme, a spatial covariance localization scheme with physical distance, and a nesting option in which the LETKF is applied to a finer-resolution model by using the guess outputs of the parent model as boundary conditions. These enhancements favor the practical application of the LETKF to local severe events.

The NHM–LETKF approach was applied to a severe rainfall event during July 2012 on Kyushu. Compared with the operational results from JMA, the forecast initialized with the LETKF analysis was clearly improved, successfully capturing torrential rainfall. The ensemble forecast experiments provided probabilistic information for precipitation that visually corresponded well with the observed rainfall amounts. Experimental results using one-way nested data assimilation demonstrated that this technique improved precipitation forecasts of local severe weather conditions, although there are still difficulties in predicting these severe events with adequate temporal and spatial accuracy in NWP models.

For simplicity, the NHM–LETKF system assimilates quality-controlled observations used in the JMA operational analysis. Because the quality-control processes include thinning of the observational data, observations would be over- or underthinned for LETKF analyses at different resolutions. This study focused on the general performance of the NHM–LETKF method for local severe events, but adjustments in data thinning may be indispensable for further improvements of the system. Other limitations of this study included the absence of Doppler radar data and surface observations. Some previous studies (e.g., Snyder and Zhang 2003; Zhang et al. 2004; Tong and Xue 2005) demonstrated the potential of radar data and surface observations for convective-scale analysis in the EnKF, producing promising results. Assimilating these data is important for further improving one-way nested LETKF.

Acknowledgments

The author thanks the members of the Forecast Research Department of the Meteorological Research Institute and Takemasa Miyoshi of the RIKEN Advanced Institute for Computational Science for fruitful discussions. The valuable comments of the anonymous reviewers helped improve the manuscript significantly. This work was supported by the 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 Prediction of Local Heavy Rainfall) and the Strategic Programs for Innovative Research (SPIRE). A portion of the results were obtained by using the K computer at the RIKEN Advanced Institute for Computational Science (Proposal hp120282).

REFERENCES

  • Cardinali, C., 2009: Monitoring the observation impact on the short-range forecast. Quart. J. Roy. Meteor. Soc., 135, 239250, doi:10.1002/qj.366.

    • Search Google Scholar
    • Export Citation
  • Davidson, N. E., Kurihara K. , Kato T. , Mills G. , and Puri K. , 1998: Dynamics and prediction of a mesoscale extreme rain event in the baiu front over Kyushu, Japan. Mon. Wea. Rev., 126, 16081629, doi:10.1175/1520-0493(1998)126<1608:DAPOAM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Evensen, G., 1994: Sequential data assimilation with a nonlinear quasigeostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99, 10 14310 162, doi:10.1029/94JC00572.

    • Search Google Scholar
    • Export Citation
  • Gelaro, R., and Zhu Y. , 2009: Examination of observation impacts derived from observing system experiments (OSEs) and adjoint models. Tellus, 61A, 179193, doi:10.1111/j.1600-0870.2008.00388.x.

    • Search Google Scholar
    • Export Citation
  • Hoffman, M. J., and Coauthors, 2010: An ensemble Kalman filter data assimilation system for the Martian atmosphere: Implementation and simulation experiments. Icarus, 209, 470481, doi:10.1016/j.icarus.2010.03.034.

    • Search Google Scholar
    • Export Citation
  • Honda, Y., Nishijima M. , Koizumi K. , Ohta Y. , Tamiya K. , Kawabata T. , and Tsuyuki T. , 2005: A pre-operational variational data assimilation system for a non-hydrostatic model at the Japan Meteorological Agency: Formulation and preliminary results. Quart. J. Roy. Meteor. Soc., 131, 34653475, doi:10.1256/qj.05.132.

    • Search Google Scholar
    • Export Citation
  • Hunt, B. R., and Coauthors, 2004: Four-dimensional ensemble Kalman filtering. Tellus, 56A, 273277, doi:10.1111/j.1600-0870.2004.00066.x.

    • Search Google Scholar
    • Export Citation
  • Hunt, B. R., Kostelich E. J. , and Szunyogh I. , 2007: Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter. Physica D, 230, 112126, doi:10.1016/j.physd.2006.11.008.

    • Search Google Scholar
    • Export Citation
  • Ikawa, M., and Saito K. , 1991: Description of a nonhydrostatic model developed at the Forecast Research Department of the MRI. MRI Tech. Rep. 28, 238 pp.

  • Kalman, R. E., 1960: A new approach to linear filtering and prediction problems. J. Fluids Eng., 82, 3545, doi:10.1115/1.3662552.

  • Kang, J.-S., Kalnay E. , Liu J. , Fung I. , Miyoshi T. , and Ide K. , 2011: “Variable localization” in an ensemble Kalman filter: Application to the carbon cycle data assimilation. J. Geophys. Res., 116, D09110, doi:10.1029/2010JD014673.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., and Miyoshi T. , 2012: Including uncertainties of sea surface temperature in an ensemble Kalman filter: A case study of Typhoon Sinlaku (2008). Wea. Forecasting, 27, 15861597, doi:10.1175/WAF-D-11-00136.1.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., and Coauthors, 2011: Verification and intercomparison of mesoscale ensemble prediction systems in the Beijing 2008 Olympics Research and Development Project. Tellus, 63A, 531549, doi:10.1111/j.1600-0870.2011.00512.x.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., Miyoshi T. , and Kalnay E. , 2012: Estimating the impact of real observations in regional numerical weather prediction using an ensemble Kalman filter. Mon. Wea. Rev., 140, 19751987, doi:10.1175/MWR-D-11-00205.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., 2011: The Gaussian approach to adaptive covariance inflation and its implementation with the local ensemble transform Kalman filter. Mon. Wea. Rev., 139, 15191535, doi:10.1175/2010MWR3570.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Aranami K. , 2006: Applying a four-dimensional local ensemble transform Kalman filter (4D-LETKF) to the JMA Nonhydrostatic Model (NHM). SOLA, 2, 128131, doi:10.2151/sola.2006-033.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Yamane S. , 2007: Local ensemble transform Kalman filtering with an AGCM at a T159/L48 resolution. Mon. Wea. Rev., 135, 38413861, doi:10.1175/2007MWR1873.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Kunii M. , 2012: The local ensemble transform Kalman filter with the Weather Research and Forecasting model: Experiments with real observations. Pure Appl. Geophys., 169, 321333, doi:10.1007/s00024-011-0373-4.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., Yamane S. , and Enomoto T. , 2007: Localizing the error covariance by physical distances within a local ensemble transform Kalman filter (LETKF). SOLA, 3, 8992, doi:10.2151/sola.2007-023.

    • Search Google Scholar
    • Export Citation
  • Nakanishi, M., and Niino H. , 2004: An improved Mellor–Yamada level 3 model with condensation physics: Its design and verification. Bound.-Layer Meteor., 112, 131, doi:10.1023/B:BOUN.0000020164.04146.98.

    • Search Google Scholar
    • Export Citation
  • Nakanishi, M., and Niino H. , 2006: An improved Mellor–Yamada level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397407, doi:10.1007/s10546-005-9030-8.

    • Search Google Scholar
    • Export Citation
  • Ninomiya, K., 2000: Large- and meso-α-scale characteristics of meiyu/baiu front associated with intense rainfalls in 1–10 July 1991. J. Meteor. Soc. Japan, 78, 141157.

    • Search Google Scholar
    • Export Citation
  • Ohmori, S., and Yamada Y. , 2006: Development of cumulus parameterization scheme in the nonhydrostatic mesoscale model at the Japan Meteorological Agency. Research Actions in Atmospheric and Oceanic Modelling 35, CAS/JSC WGNE, 2 pp. [Available online at http://www.wcrp-climate.org/WGNE/BlueBook/2006/individual-articles/04_Ohmori_Shiro__12487.pdf.]

  • Ott, E., and Coauthors, 2004: A local ensemble Kalman filter for atmospheric data assimilation. Tellus, 56A, 415428, doi:10.1111/j.1600-0870.2004.00076.x.

    • Search Google Scholar
    • Export Citation
  • Penny, S., 2011: Data assimilation of the global ocean using the 4D local ensemble transform Kalman filter (4D-LETKF) and the Modular Ocean Model (MOM2). Ph.D. thesis, University of Maryland, 141 pp. [Available online at http://hdl.handle.net/1903/11716.]

  • Saito, K., 2012: The Japan Meteorological Agency 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, 12661298, doi:10.1175/MWR3120.1.

    • Search Google Scholar
    • Export Citation
  • Saito, K., Ishida J. , Aranami K. , Hara T. , Segawa T. , Narita M. , and Honda Y. , 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan, 85B, 271304, doi:10.2151/jmsj.85B.271.

    • Search Google Scholar
    • Export Citation
  • Saito, K., Seko H. , Kunii M. , and Miyoshi T. , 2012: Effect of lateral boundary perturbations on the breeding method and the local ensemble transform Kalman filter for mesoscale ensemble prediction. Tellus, 64A, 11594, doi:10.3402/tellusa.v64i0.11594.

    • Search Google Scholar
    • Export Citation
  • Scherrer, S. C., Appenzeller C. , Eckert P. , and Cattani D. , 2004: Analysis of the spread–skill relations using the ECMWF Ensemble Prediction System over Europe. Wea. Forecasting, 19, 552565, doi:10.1175/1520-0434(2004)019<0552:AOTSRU>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Schutgens, N. A. J., Miyoshi T. , Takemura T. , and Nakajima T. , 2010: Applying an ensemble Kalman filter to the assimilation of AERONET observations in a global aerosol transport model. Atmos. Chem. Phys., 10, 25612576, doi:10.5194/acp-10-2561-2010.

    • Search Google Scholar
    • Export Citation
  • Shchepetkin, A., and McWilliams J. C. , 2005: The Regional Oceanic Modeling System: A split-explicit, free-surface, topography-following-coordinate ocean model. Ocean Modell., 9, 347404, doi:10.1016/j.ocemod.2004.08.002.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., Klemp J. B. , Dudhia J. , Gill D. O. , Barker D. M. , Wang W. , and Powers J. G. , 2005: A description of the advanced research WRF version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v2.pdf.]

  • Snyder, C., and Zhang F. , 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131, 16631677, doi:10.1175//2555.1.

    • Search Google Scholar
    • Export Citation
  • Tong, M., and Xue M. , 2005: Ensemble Kalman filter assimilation of Doppler radar data with a compressible nonhydrostatic model: OSS experiments. Mon. Wea. Rev., 133, 17891807, doi:10.1175/MWR2898.1.

    • Search Google Scholar
    • Export Citation
  • Whitaker, J. S., and Loughe A. F. , 1998: The relationship between ensemble spread and ensemble mean skill. Mon. Wea. Rev., 126, 32923302, doi:10.1175/1520-0493(1998)126<3292:TRBESA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., Snyder C. , and Sun J. , 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Wea. Rev., 132, 12381253, doi:10.1175/1520-0493(2004)132<1238:IOIEAO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhu, Y., and Gelaro R. , 2008: Observation sensitivity calculations using the adjoint of the Gridpoint Statistical Interpolation (GSI) analysis system. Mon. Wea. Rev., 136, 335351, doi:10.1175/MWR3525.1.

    • Search Google Scholar
    • Export Citation
Save
  • Cardinali, C., 2009: Monitoring the observation impact on the short-range forecast. Quart. J. Roy. Meteor. Soc., 135, 239250, doi:10.1002/qj.366.

    • Search Google Scholar
    • Export Citation
  • Davidson, N. E., Kurihara K. , Kato T. , Mills G. , and Puri K. , 1998: Dynamics and prediction of a mesoscale extreme rain event in the baiu front over Kyushu, Japan. Mon. Wea. Rev., 126, 16081629, doi:10.1175/1520-0493(1998)126<1608:DAPOAM>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Evensen, G., 1994: Sequential data assimilation with a nonlinear quasigeostrophic model using Monte Carlo methods to forecast error statistics. J. Geophys. Res., 99, 10 14310 162, doi:10.1029/94JC00572.

    • Search Google Scholar
    • Export Citation
  • Gelaro, R., and Zhu Y. , 2009: Examination of observation impacts derived from observing system experiments (OSEs) and adjoint models. Tellus, 61A, 179193, doi:10.1111/j.1600-0870.2008.00388.x.

    • Search Google Scholar
    • Export Citation
  • Hoffman, M. J., and Coauthors, 2010: An ensemble Kalman filter data assimilation system for the Martian atmosphere: Implementation and simulation experiments. Icarus, 209, 470481, doi:10.1016/j.icarus.2010.03.034.

    • Search Google Scholar
    • Export Citation
  • Honda, Y., Nishijima M. , Koizumi K. , Ohta Y. , Tamiya K. , Kawabata T. , and Tsuyuki T. , 2005: A pre-operational variational data assimilation system for a non-hydrostatic model at the Japan Meteorological Agency: Formulation and preliminary results. Quart. J. Roy. Meteor. Soc., 131, 34653475, doi:10.1256/qj.05.132.

    • Search Google Scholar
    • Export Citation
  • Hunt, B. R., and Coauthors, 2004: Four-dimensional ensemble Kalman filtering. Tellus, 56A, 273277, doi:10.1111/j.1600-0870.2004.00066.x.

    • Search Google Scholar
    • Export Citation
  • Hunt, B. R., Kostelich E. J. , and Szunyogh I. , 2007: Efficient data assimilation for spatiotemporal chaos: A local ensemble transform Kalman filter. Physica D, 230, 112126, doi:10.1016/j.physd.2006.11.008.

    • Search Google Scholar
    • Export Citation
  • Ikawa, M., and Saito K. , 1991: Description of a nonhydrostatic model developed at the Forecast Research Department of the MRI. MRI Tech. Rep. 28, 238 pp.

  • Kalman, R. E., 1960: A new approach to linear filtering and prediction problems. J. Fluids Eng., 82, 3545, doi:10.1115/1.3662552.

  • Kang, J.-S., Kalnay E. , Liu J. , Fung I. , Miyoshi T. , and Ide K. , 2011: “Variable localization” in an ensemble Kalman filter: Application to the carbon cycle data assimilation. J. Geophys. Res., 116, D09110, doi:10.1029/2010JD014673.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., and Miyoshi T. , 2012: Including uncertainties of sea surface temperature in an ensemble Kalman filter: A case study of Typhoon Sinlaku (2008). Wea. Forecasting, 27, 15861597, doi:10.1175/WAF-D-11-00136.1.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., and Coauthors, 2011: Verification and intercomparison of mesoscale ensemble prediction systems in the Beijing 2008 Olympics Research and Development Project. Tellus, 63A, 531549, doi:10.1111/j.1600-0870.2011.00512.x.

    • Search Google Scholar
    • Export Citation
  • Kunii, M., Miyoshi T. , and Kalnay E. , 2012: Estimating the impact of real observations in regional numerical weather prediction using an ensemble Kalman filter. Mon. Wea. Rev., 140, 19751987, doi:10.1175/MWR-D-11-00205.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., 2011: The Gaussian approach to adaptive covariance inflation and its implementation with the local ensemble transform Kalman filter. Mon. Wea. Rev., 139, 15191535, doi:10.1175/2010MWR3570.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Aranami K. , 2006: Applying a four-dimensional local ensemble transform Kalman filter (4D-LETKF) to the JMA Nonhydrostatic Model (NHM). SOLA, 2, 128131, doi:10.2151/sola.2006-033.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Yamane S. , 2007: Local ensemble transform Kalman filtering with an AGCM at a T159/L48 resolution. Mon. Wea. Rev., 135, 38413861, doi:10.1175/2007MWR1873.1.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., and Kunii M. , 2012: The local ensemble transform Kalman filter with the Weather Research and Forecasting model: Experiments with real observations. Pure Appl. Geophys., 169, 321333, doi:10.1007/s00024-011-0373-4.

    • Search Google Scholar
    • Export Citation
  • Miyoshi, T., Yamane S. , and Enomoto T. , 2007: Localizing the error covariance by physical distances within a local ensemble transform Kalman filter (LETKF). SOLA, 3, 8992, doi:10.2151/sola.2007-023.

    • Search Google Scholar
    • Export Citation
  • Nakanishi, M., and Niino H. , 2004: An improved Mellor–Yamada level 3 model with condensation physics: Its design and verification. Bound.-Layer Meteor., 112, 131, doi:10.1023/B:BOUN.0000020164.04146.98.

    • Search Google Scholar
    • Export Citation
  • Nakanishi, M., and Niino H. , 2006: An improved Mellor–Yamada level-3 model: Its numerical stability and application to a regional prediction of advection fog. Bound.-Layer Meteor., 119, 397407, doi:10.1007/s10546-005-9030-8.

    • Search Google Scholar
    • Export Citation
  • Ninomiya, K., 2000: Large- and meso-α-scale characteristics of meiyu/baiu front associated with intense rainfalls in 1–10 July 1991. J. Meteor. Soc. Japan, 78, 141157.

    • Search Google Scholar
    • Export Citation
  • Ohmori, S., and Yamada Y. , 2006: Development of cumulus parameterization scheme in the nonhydrostatic mesoscale model at the Japan Meteorological Agency. Research Actions in Atmospheric and Oceanic Modelling 35, CAS/JSC WGNE, 2 pp. [Available online at http://www.wcrp-climate.org/WGNE/BlueBook/2006/individual-articles/04_Ohmori_Shiro__12487.pdf.]

  • Ott, E., and Coauthors, 2004: A local ensemble Kalman filter for atmospheric data assimilation. Tellus, 56A, 415428, doi:10.1111/j.1600-0870.2004.00076.x.

    • Search Google Scholar
    • Export Citation
  • Penny, S., 2011: Data assimilation of the global ocean using the 4D local ensemble transform Kalman filter (4D-LETKF) and the Modular Ocean Model (MOM2). Ph.D. thesis, University of Maryland, 141 pp. [Available online at http://hdl.handle.net/1903/11716.]

  • Saito, K., 2012: The Japan Meteorological Agency 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, 12661298, doi:10.1175/MWR3120.1.

    • Search Google Scholar
    • Export Citation
  • Saito, K., Ishida J. , Aranami K. , Hara T. , Segawa T. , Narita M. , and Honda Y. , 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan, 85B, 271304, doi:10.2151/jmsj.85B.271.

    • Search Google Scholar
    • Export Citation
  • Saito, K., Seko H. , Kunii M. , and Miyoshi T. , 2012: Effect of lateral boundary perturbations on the breeding method and the local ensemble transform Kalman filter for mesoscale ensemble prediction. Tellus, 64A, 11594, doi:10.3402/tellusa.v64i0.11594.

    • Search Google Scholar
    • Export Citation
  • Scherrer, S. C., Appenzeller C. , Eckert P. , and Cattani D. , 2004: Analysis of the spread–skill relations using the ECMWF Ensemble Prediction System over Europe. Wea. Forecasting, 19, 552565, doi:10.1175/1520-0434(2004)019<0552:AOTSRU>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Schutgens, N. A. J., Miyoshi T. , Takemura T. , and Nakajima T. , 2010: Applying an ensemble Kalman filter to the assimilation of AERONET observations in a global aerosol transport model. Atmos. Chem. Phys., 10, 25612576, doi:10.5194/acp-10-2561-2010.

    • Search Google Scholar
    • Export Citation
  • Shchepetkin, A., and McWilliams J. C. , 2005: The Regional Oceanic Modeling System: A split-explicit, free-surface, topography-following-coordinate ocean model. Ocean Modell., 9, 347404, doi:10.1016/j.ocemod.2004.08.002.

    • Search Google Scholar
    • Export Citation
  • Skamarock, W. C., Klemp J. B. , Dudhia J. , Gill D. O. , Barker D. M. , Wang W. , and Powers J. G. , 2005: A description of the advanced research WRF version 2. NCAR Tech. Note NCAR/TN-468+STR, 88 pp. [Available online at http://www.mmm.ucar.edu/wrf/users/docs/arw_v2.pdf.]

  • Snyder, C., and Zhang F. , 2003: Assimilation of simulated Doppler radar observations with an ensemble Kalman filter. Mon. Wea. Rev., 131, 16631677, doi:10.1175//2555.1.

    • Search Google Scholar
    • Export Citation
  • Tong, M., and Xue M. , 2005: Ensemble Kalman filter assimilation of Doppler radar data with a compressible nonhydrostatic model: OSS experiments. Mon. Wea. Rev., 133, 17891807, doi:10.1175/MWR2898.1.

    • Search Google Scholar
    • Export Citation
  • Whitaker, J. S., and Loughe A. F. , 1998: The relationship between ensemble spread and ensemble mean skill. Mon. Wea. Rev., 126, 32923302, doi:10.1175/1520-0493(1998)126<3292:TRBESA>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhang, F., Snyder C. , and Sun J. , 2004: Impacts of initial estimate and observation availability on convective-scale data assimilation with an ensemble Kalman filter. Mon. Wea. Rev., 132, 12381253, doi:10.1175/1520-0493(2004)132<1238:IOIEAO>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Zhu, Y., and Gelaro R. , 2008: Observation sensitivity calculations using the adjoint of the Gridpoint Statistical Interpolation (GSI) analysis system. Mon. Wea. Rev., 136, 335351, doi:10.1175/MWR3525.1.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Surface weather map at 1800 UTC 11 Jul 2012 analyzed by JMA.

  • Fig. 2.

    Time series of 3-h accumulated rainfall (mm) averaged over the study region, estimated from the 5-km interpolated rainfall amount.

  • Fig. 3.

    Model domains for the NHM–LETKF analysis (D01, 15-km grid spacing) and the one-way nested analysis (D02, 5-km grid spacing) with elevation (m).

  • Fig. 4.

    Ensemble spreads of zonal wind at the 500-hPa level in the outer LETKF (a) without and (b) with lateral boundary perturbations (m s−1). (c),(d) As in (a),(b), but for the analysis increment.

  • Fig. 5.

    Verification of 6-h forecasts relative to radiosonde observations for the (a) zonal wind component (m s−1), (b) temperature (K), and (c) relative humidity (%), averaged over 10 days (from 0000 UTC 5 Jul to 0000 UTC 15 Jul 2012) in the outer LETKF cycles. Solid (dashed) lines correspond to experiments with (without) lateral boundary perturbations (LBP and CTL). Thick (thin) lines indicate RMSEs (biases).

  • Fig. 6.

    Time series of 6-h forecast (GUES) and analysis (ANAL) RMSEs relative to radiosonde observations for the (a) zonal wind component and (b) relative humidity at 850 hPa in the outer LETKF cycles.

  • Fig. 7.

    Time series of analysis ensemble spreads for the (a) zonal wind component and (b) relative humidity at 850 hPa, averaged over the entire model domain (ALL) and over the southwestern portion of the Kyushu district (LMT, 27.5°–35.0°N, 125.0°–132.5°E).

  • Fig. 8.

    (a) TS and (b) BS for 3-h accumulated rainfall in the 12-h NHM forecasts with 5-km horizontal resolution initialized with the JNoVA (dashed lines) and LETKF (solid lines) analyses, averaged over 13 initial times from 1200 UTC 10 Jul to 1200 UTC 13 Jul. (c),(d) As in (a),(b), but for the 24-h forecasts.

  • Fig. 9.

    Simulated mean sea level pressure (hPa; solid line) and 3-h accumulated precipitation (mm; color) in 18-h NHM forecasts with a horizontal resolution of 5 km at 0000 UTC 12 Jul 2012, initialized with the (a) JNoVA and (b) LETKF analyses. (c) JMA RAP observations from this time period.

  • Fig. 10.

    Forecasts of the POP exceeding 50 mm (3 h)−1 as of 0000 UTC 12 Jul 2012, calculated from the 5-km NHM ensemble outputs, with lead times of (a) 36, (b) 30, (c) 24, (d) 18, (e) 12, and (f) 6 h.

  • Fig. 11.

    Simulated mean sea level pressure (hPa) and 3-h accumulated precipitation (mm) in 18-h NHM forecasts with a horizontal resolution of 1 km, initialized with the (a) outer and (b) nested LETKF approaches.

  • Fig. 12.

    (a) Threat scores for 3-h accumulated rainfall averaged between 3 and 15 h in the DA15 and DA05 experiments. Accumulated precipitation in the RAP data (OBS) and the DA15 and DA05 experiments at (b) 32.5°N, 130.6°E and (c) 32.5°N, 130.8°E, respectively. The initial time of the forecast is 1800 UTC 11 Jul 2012.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 387 67 4
PDF Downloads 310 51 6