Ensemble experiments using a nested LETKF system to reproduce intense vortices associated with tornadoes of 6 May 2012 in Japan
© Seko et al. 2015
Received: 22 April 2015
Accepted: 5 November 2015
Published: 26 November 2015
Experiments simulating intense vortices associated with tornadoes that occurred on 6 May 2012 on the Kanto Plain, Japan, were performed with a nested local ensemble transform Kalman filter (LETKF) system. Intense vortices were reproduced by downscale experiments with a 12-member ensemble in which the initial conditions were obtained from the nested LETKF system analyses. The downscale experiments successfully generated intense vortices in three regions similar to the observed vortices, whereas only one tornado was reproduced by a deterministic forecast. The intense vorticity of the strongest tornado, which was observed in the southernmost region, was successfully reproduced by 10 of the 12 ensemble members. An examination of the results of the ensemble downscale experiments showed that the duration of intense vorticities tended to be longer when the vertical shear of the horizontal wind was larger and the lower airflow was more humid. Overall, the study results show that ensemble forecasts have the following merits: (1) probabilistic forecasts of the outbreak of intense vortices associated with tornadoes are possible; (2) the miss rate of outbreaks should decrease; and (3) environmental factors favoring outbreaks can be obtained by comparing the multiple possible scenarios of the ensemble forecasts.
KeywordsEnsemble forecast Tornadoes LETKF
Local heavy rainfalls and tornadoes cause severe damage due to flash floods, landslides, and strong winds. Accurate forecasts of these phenomena, for example, by numerical models, could reduce the damages they cause. However, numerical models are not perfect, and the initial conditions of numerical forecasts include errors. In particular, forecasts of convection cells generated in areas of weak convergence, which frequently occur in urban areas such as the Tokyo Metropolitan Area in summer, are affected by small errors in the initial conditions. This means that forecasts of the position and timing at which a thunderstorm will be generated from a weak convergence are sensitive to errors in the initial conditions. Because of the resulting uncertainty in forecasts of convection cell generation, ensemble forecast techniques must be employed to produce probabilistic forecasts. In addition to probabilistic forecasts, ensemble forecasts produce multiple possible scenarios. Thus, ensemble forecasts have the following merits: (1) analyzed fields of ensemble forecasts (i.e., the ensemble average of multiple scenarios) are statistically more accurate than those of deterministic forecasts and (2) multiple possible scenarios provide information regarding uncertainty (i.e., when there is large scatter among the scenarios, the uncertainty is larger).
The ensemble forecasts operationally performed by the Japan Meteorological Agency (JMA) with the highest horizontal resolution are typhoon forecasts and 1 week forecasts. However, these ensemble forecasts cannot represent convection cells of thunderstorms that cause local heavy rainfall and tornadoes because the horizontal grid interval (40 km) is too large to resolve the convection cells. Recently, several ensemble forecast experiments in which heavy rainfall or local heavy rainfall was reproduced have been performed with a local ensemble transform Kalman filter (LETKF; Hunt et al. 2007; Miyoshi and Aranami 2006). For instance, Kunii (2013) reproduced a heavy rainfall in northern Kyushu by assimilating JMA’s operational observation data with a LETKF system with a grid interval of 5 km. Probability distributions for a 3 h rainfall amount exceeding 50 mm were obtained by changing the initial time of the ensemble forecasts. Heavy rainfall of the northern Kyushu heavy rainfall event could be forecast with high probability even when the initial time of the ensemble forecasts was 12 h before the event.
One method of representing convection cells is to perform downscale forecasts with the model with a grid interval of a few kilometers and to obtain the initial conditions by spatial and temporal interpolation of the analyzed or predicted fields of a data assimilation system. For example, a local heavy rainfall, which caused a flash flood in the Toga River in Kobe that swept away six people, was reproduced by downscale experiments with JMA’s non-hydrostatic model (JMANHM), the initial fields of which were obtained with a LETKF system (Seko et al. 2011). Though the linear band of intense rainfall responsible for the local heavy rainfall was represented as a weak rainfall region by the NHM-LETKF system with a grid interval of 20 km, downscale simulations with a grid interval of 1.6 km were successful in reproducing the linear rainfall band. This result is one example of a local heavy rainfall being reproduced by modifying the environment around the rainfall system through the assimilation of observed data.
A local heavy rainfall that occurred on the Osaka Plain on 5 September 2008 was reproduced by a nested LETKF system (Seko et al. 2013). In this experiment, high-resolution data (Doppler radar and GPS data) were assimilated to reproduce the local heavy rainfall. The number of ensemble members that reproduced the local heavy rainfall was increased by the assimilation of high-resolution data, compared with the number when the analyzed fields were obtained with assimilating only low-resolution conventional data. The results of these ensemble experiments show that the use of a LETKF system is a promising approach to the analysis and prediction of local heavy rainfall events.
In this study, we applied a LETKF system to the prediction of intense vortices associated with tornadoes. To date, no numerical experiments have been conducted using ensemble Kalman filters with a horizontal resolution of a few kilometers to reproduce intense vortices associated with tornadoes occurring in Japan, although a number of studies have used ensemble Kalman filters to reproduce or to investigate tornadoes that occurred in the USA. For instance, Snook et al. (2011) showed that it was possible to obtain probabilistic forecasts by using an ensemble Kalman filter analysis system that assimilated Doppler radar data from the Collaborative Adaptive Sensing of the Atmosphere and Weather Surveillance Radar-1988 systems. Yussouf et al. (2013) conducted simulations with a mesoscale and convective scale ensemble data assimilation system consisting of the Weather Research and Forecasting model and an ensemble adjustment Kalman filter that used a double-moment microphysics scheme, and they successfully reproduced a supercell storm. Clark et al. (2013) investigated the relation between the path lengths of tornadoes and simulated updraft helicity by using an ensemble Kalman filter system with a horizontal grid interval of 1.25 km. Recently, Snook et al. (2015) demonstrated that use of an ensemble Kalman filter system could provide skillful analyses and ensemble-based probabilistic forecasts of tornadic mesoscale convective systems in the USA.
Tornadoes in Japan are different from those in the USA; Japanese tornadoes have an intensity equal to or less than F3 on the Fujita scale, and the low-level atmosphere in Japan is much more humid. Ensemble experiments to reproduce Japanese tornadoes are desirable for improving the forecast accuracy of tornadoes in Japan and for understanding the generation mechanisms of Japanese tornadoes.
We applied a nested LETKF system to a tornado event on 6 May 2012, during which three tornadoes were generated on the Kanto Plain. To express the vortices associated with these tornadoes, we performed downscaling experiments using the JMANHM model (Saito et al. 2006) with a horizontal grid interval of 350 m (NHM-350) and the analyzed fields produced by the LETKF system. We expected the multiple possible scenarios obtained from these ensemble forecast experiments to be useful for inferring environmental factors favoring the generation of intense vortices.
The rest of this paper is organized as follows: In the “Methods” section, the tornadoes of 6 May 2012 are first briefly described, and then, the LETKF system and downscale forecast experiments are explained. In the “Results and discussion” section, we present the results of the ensemble downscale experiments and discuss the factors that favor the generation of the intense vortices associated with tornadoes. In the “Conclusions” section, we present our conclusions.
Observed features of the tornadoes of 6 May 2012
JMA determined the paths of the tornadoes that were generated on 6 May from the damage left in their wake (Japan Meteorological Agency, Meteorological Research Institute, Tokyo District Meteorological Observatory and Sendai District Meteorological Observatory 2012). Three tornadoes were generated at around 12:30 JST over the northern Kanto Plain. The intensity of the southernmost tornado was F3 on the Fujita scale, the highest intensity of tornadoes observed in Japan, and this tornado damaged about 800 houses, killed one person, and injured 37 people. The two northern tornadoes had intensities of F2 and F1.
The Meteorological Research Institute (MRI) has analyzed the environmental conditions when these tornadoes were generated by using JMA’s operational mesoscale analysis data (Meteorological Research Institute 2012). The water vapor distribution at 12:00 JST on 6 May showed that a moist airflow supplied the intense rainfall regions where the tornadoes were generated. In particular, a moist airflow with a water vapor mixing ratio that exceeded 12 g kg−1 contributed to the generation of these tornadoes (Meteorological Research Institute 2012). To investigate the detailed structure of the convection cell that produced the southernmost tornado, MRI conducted a deterministic experiment using a model with a horizontal grid interval of 250 m. Although this experiment reproduced only the southernmost tornado, it showed that the rainfall system that generated the tornadoes had a supercell structure. In the present study, we performed ensemble experiments to further investigate these three tornadoes generated on the northern Kanto Plain.
Outline of the experiments performed with the nested LETKF system
The assimilation window of the inner LETKF was 1 h, and the slot interval was 10 min. In this study, conventional data (described above) from JMA were assimilated in both LETKFs. Although higher resolution observation data (e.g., radial wind from Doppler radars and GPS precipitable water vapor) are expected to be better able to reproduce smaller scale distributions of the initial conditions than conventional data, we did not use such high-resolution data in this study because our main purpose was to explore the merits of ensemble forecasts. Yokota et al. (2015) have shown that the assimilation of high-resolution data (e.g., radial winds from Doppler radars and dense surface observation data of AMeDAS and the Environmental Sensor Network of NTT DOCOMO, Inc.) could improve the positions of the reproduced vortices of the tornadoes of 6 May 2012. The analyzed fields of the inner LETKF were reflected to those of the outer LETKF every 6 h, at which time the analyzed fields of both the inner and outer LETKFs were available. The data assimilation cycles of the outer and inner LETKFs began at 09:00 JST on 3 September and 03:00 JST on 6 September, respectively.
NHM-350, which performed the downscale experiments to represent the intense vortices associated with the tornadoes, had 600 × 600 horizontal and 70 vertical grids, and its initial time was set to 10:30 JST on 6 September. The initial and boundary conditions for NHM-350 were produced from the output of the inner LETKF. NHM-350 used a double-moment microphysics parameterization scheme and the Deardorff (1980) planetary boundary scheme.
Results and discussion
Intense vortices produced by the inner LETKF and NHM-350
Ensemble members #004 and #007 generated the strongest and weakest vorticities among the ensemble members, respectively, at the height of 20 m. An intense vortex was generated near the southern tip of the intense convection by #004 at 11:45 JST (Fig. 7b), 45 min earlier than the observed southernmost tornado was generated, but only weak vorticity was generated by #007 (Fig. 7c). The position of the southernmost intense vortex relative to the rainfall region in #004 was the same as that of the observed vortex relative to the observed rainfall region. This result indicates that #004 reproduced the observed rainfall region that generated the tornado. An intense vorticity was not reproduced by #007, even though #007 produced more intense rainfall than #004 did. The relationship between vorticity and rainfall intensity in #007 suggests that rainfall intensity might not be a factor controlling the generation of intense vortices. The convergence reproduced by #004 at the southern edge of the northern rainfall region (red circle in Fig. 7b) corresponds to the northern intense vortex that was reproduced by #006 (Fig. 7a), though the vorticity generated by #004 was weaker than 0.1 s−1. The stream lines, which show the small-scale distributions of horizontal wind, indicate that in #004 (Fig. 7b) the airflow that converged at the position of the northern vortex passed near the southern rainfall region. If the southern rainfall region modified the airflows that were supplied to the northern rainfall region, it would prevent the development of the northern rainfall region.
We chose ensemble members #002 and #006 for the trajectory analysis because intense vorticity was generated near the northern rainfall region by #006 but not by #002 (Fig. 7).
Both members showed most airflows approaching the southern tip of the northern rainfall region from the south. In #002, the airflows passed the southern rainfall region at 11:30 JST, and their potential temperature decreased before their arrival at the northern rainfall region. In #006, in contrast, some airflows passed far to the east of the southern rainfall region, and their temperature did not decrease. After the passage of these airflows from the east of the southern rainfall region, their direction of movement was changed by the presence of low pressure near the rainfall region, causing them to approach the southern tip of the northern rainfall region. That is, the southern rainfall region in #006 is deduced to have a smaller effect on the airflows that were supplied to the northern rainfall region. This difference of the paths, as a result, might be one reason why #006 was able to generate intense vorticity near the northern rainfall region. This relationship between the southern rainfall and the intense vortex near the northern rainfall region was obtained with the NHM-350 system. To confirm the existence of a corresponding observed relationship, observation data of airflows within the lower atmosphere over the Kanto Plain would be needed.
where V(z) is the horizontal wind vector at height z, k is the vertical unit vector, and C is the moving vector of the storm. To estimate SReH, the moving speed C of the southern rainfall region is needed. In this study, the moving speeds of the reproduced southern rainfall regions were used. A scatter diagram of the durations of intense vortices and the values of certain environmental factors shows that the duration became longer as the low-level layer became more humid and the vertical wind shear became larger (Fig. 10a). The duration also became longer when the SReH in the averaged region was larger.
In addition, comparison of duration with the number of grids in which water vapor and vertical wind shear in the averaged area exceeded 12 g kg−1 and 17 m s−1, respectively (Fig. 10c), showed that the durations of the intense vortices became longer as the number of grids exceeding these thresholds became larger. This result was expected because the number of grids in which values exceeded these thresholds was correlated positively with the averaged values shown in Fig. 10a. A larger number of grids with values exceeding these thresholds indicated that the favorable region for intense vortex generation within the averaged area, that is, within the rectangle on the inflow side of the southern rainfall region, was larger. Therefore, these results indicate that ensemble members reproducing large areas with a favorable environment reproduce vortices with a long duration. Thus, favorable environmental factors can be extracted by comparing the outputs of the ensemble members.
Environmental factors affecting intense vortex generation
where w is the weight of ensemble member #004.
As w became smaller, the maximum vorticity generated during FT = 0 min to 180 min became weaker (Fig. 11). This result is expected because the impact of #004 is reduced when w is smaller; the timing of the maximum vorticity, however, did not change linearly from that of #007 as w decreased. When the weight was changed from 0.25 (#004-007_025) to 0.125 (#004-007_0125), the vorticity decreased greatly. This large change of vertical vorticity suggests that the #004-007_025 experiment was particularly sensitive to environmental factors; therefore, we expected this experiment to show the impact of environmental factors more clearly than the others. Thus, we investigated the impact of each environmental factor by replacing water vapor, horizontal wind, or potential temperature in the initial conditions of #004-007_025 with those of #007.
When the horizontal wind or water vapor was replaced with those of #007 (#004-007_025_UV, #004-007_025_QV), the vorticities became weaker. In contrast, when the potential temperature was replaced with that of #007 (#004-007_025_PT), vorticity became more intense. These results indicate that the horizontal wind and water vapor in the initial conditions of #004 were favorable for the generation of an intense vortex but that the temperature distribution of #004 suppressed the generation of an intense vortex.
The intense vortices associated with tornadoes occurring on 6 May 2012 were reproduced by a nested LETKF system that simulated the environments and convection cells simultaneously. The results of this study demonstrated three merits of ensemble forecasts based on the outputs of the nested LETKF system. The occurrence probability of an intense vortex with a vertical vorticity exceeding 0.1 s−1 during this tornado event was 83 %. This result shows the first merit: probabilistic forecasts of the outbreak of intense vortices associated with tornadoes are possible. In addition, the multiple possible scenarios revealed two other merits of ensemble forecasts. Although a deterministic forecast had reproduced only the southernmost of the three tornadoes, several ensemble members reproduced the intense vortices associated with the two northern tornadoes. Thus, the second merit is that ensemble forecasts can be expected to decrease the miss rate of outbreaks because they can reproduce low-probability phenomena. Comparison of the multiple possible scenarios showed that the duration of the southernmost tornado was related to the low-level water vapor and the vertical wind shear between heights of 0.8 and 2.5 km. Thus, the third merit is that the multiple possible scenarios of the ensemble forecast make it possible to determine the environmental factors favorable for the formation of severe weather phenomena such as intense vortices associated with tornadoes.
Although we demonstrated the merits of ensemble forecasts in this paper, there are several areas where improvements are needed. First, the initial seeds used might influence the probability of outbreaks of intense vortices. In this study, the initial seeds of the outer LETKF were provided by JMA’s mesoscale model analysis fields from before the occurrence of the observed tornadoes, not by adding random perturbations. In this experiment, differences among the initial seeds might have been too small to cause dispersion of intense vortices to be large.
The second area is the number of ensemble members. The merits of the ensemble forecasts of convective scale phenomena, such as the possible distribution of intense vortices, can be demonstrated with as few as 12 ensemble members. For quantitative discussion, however, more ensemble members would be preferable. In this study, we investigated the relationships between environmental parameters and intense vortices in each ensemble member, but we did not discuss the ensemble mean distributions in depth. As shown in Fig. 3, a region of intense southerly winds existed east of the rainfall regions. Because of the small ensemble size, the ensemble mean distribution, if it was produced by simple averaging of the outputs of the ensemble members, would show several separate regions of intense southerly winds; as a result, the relationships between the southerly wind and the rainfall regions would be obscured. Therefore, the relationships in this study were extracted from the output of each ensemble member. It would also be possible to extract these relationships from composite distributions relative to the positions and timings of the occurrence of intense vortices; this approach was discussed by Yokota et al. (2014).
The third area is the assimilation data of the inner LETKF. High-resolution and high-frequency data are needed to reproduce convection cells in rainfall regions. Because such data were not used in this study, the intense vortices were not reproduced at the positions where they were actually observed. For more realistic reproduction of rainfall regions and tornadoes, high-resolution and high-frequency data should be used.
Automated Meteorological Data Acquisition System
Japan Meteorological Agency
Japan Meteorological Agency non-hydrostatic model
Japan Standard Time
Local Ensemble Transform Kalman Filter
Meteorological Research Institute
Non-Hydrostatic Model with a grid interval of 350 m
The authors express their gratitude to Drs. Kazuo Saito, Teruyuki Kato, and Wataru Mashiko of MRI, to Mr. Hiroshi Yamauchi of JMA, and to two anonymous reviewers for valuable comments. Mesoscale analysis data and conventional observation data from JMA were used to produce the initial seeds and boundary conditions of the outer LETKF and in the data assimilation. “Kasaneru 3D” and “Multi-screen monitoring tool” were used to produce Figs. 2, 4, and 5. The authors extend their gratitude to the Tokyo District Meteorological Observatory and the Observation Department and Numerical Prediction Division of JMA. This study was supported in part by “Strategic Programs for Innovative Research (SPIRE), Field 3 (ID: hp120282, hp130012, hp140220)” and “Tokyo Metropolitan Area Convection Study for Extreme Weather Resilient Cities (TOMACS).”
Science section to which this research article belongs: 2) Atmospheric and hydrospheric sciences
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Clark AJ, Gao J, Marth PT, Smith T, Kain JS, Correia Jr J, et al. Tornado pathlength forecasts from 2010 to 2011 using ensemble updraft helicity. Monthly Weather Review. 2013;28:387–407.Google Scholar
- Deardorff JW. Stratocumulus-capped mixed layers derived from a three-dimensional model. Bound-Layer Meteor. 1980;18:495–527.View ArticleGoogle Scholar
- Houze Jr RA. Cloud dynamics. San Diego: Academic Press; 1994.Google Scholar
- Hunt BR, Kostelich EJ, Szunyogh I. Efficient data assimilation for spatiotemporal chaos: a local ensemble transform Kalman filter. Physica D: Nonlinear Phenomena. 2007. 230:112–126.View ArticleGoogle Scholar
- Japan Meteorological Agency, Meteorological Research Institute, Tokyo District Meteorological Observatory and Sendai District Meteorological Observatory. Tornadoes generated on 6th May 2012. 2012. http://www.jma.go.jp/jma/menu/tatsumaki-portal/tyousa-houkoku.pdf, Accessed 21 July 2014 (in Japanese).Google Scholar
- Kunii M. Mesoscale data assimilation for a local severe rainfall event with the NHM-LETKF system. Weather and Forecasting. 2013;29:1093–105.View ArticleGoogle Scholar
- Meteorological Research Institute. Tornadoes generated near Tsukuba, Ibaraki on 6th May 2012. 2012. http://www.jma.go.jp/jma/press/1205/11c/120511tsukuba_tornado.pdf, Accessed 21 July 2014 (in Japanese).Google Scholar
- Miyoshi T, Aranami K. Applying a four-dimensional local ensemble transform Kalman filter (4D-LETKF) to the JMA nonhydrostatic model (NHM). SOLA. 2006;2:128–31.View ArticleGoogle Scholar
- Saito K, Fujita T, Yamada Y, Ishida J, Kumagai Y, Aranami K, et al. The operational JMA nonhydrostatic mesoscale model. Monthly Weather Review. 2006;134:1266–98.View ArticleGoogle Scholar
- Seko H, Miyoshi T, Shoji Y, Saito K. Data assimilation experiments of precipitable water vapour using the LETKF system: intense rainfall event over Japan 28 July 2008. Tellus A. 2011;63:402–14.View ArticleGoogle Scholar
- Seko H, Tsuyuki T, Saito K, Miyoshi T. Development of a two-way nested-LETKF system for cloud-resolving model. In: Park SK, Xu L, editors. Data assimilation for atmospheric, oceanic and hydrological applications, vol. 2. Springer: Heidelberg; 2013. p. 489–507.View ArticleGoogle Scholar
- Snook N, Xue M, Jung Y. Analysis of a tornadic mesoscale convective vortex based on ensemble Kalman filter assimilation of CASA X-band and WSR-88D Radar data. Monthly Weather Review. 2011;139:3446–68.View ArticleGoogle Scholar
- Snook N, Xue M, Jung Y. Multiscale EnKF assimilation of radar and conventional observations and ensemble forecasting for a tornadic mesoscale convective system. Monthly Weather Review. 2015;143:1035–57.View ArticleGoogle Scholar
- Yokota S, Kunii M, Seko H. Doppler radar radial wind assimilation for the tornado outbreak on May 6, 2012. CAS/JSC WGNE Research Activities in Atmospheric and Oceanic Modelling. 2014;1:29–1.30.Google Scholar
- Yokota S, Kunii M, Seko H, Yamauchi H. Assimilation of rainwater estimated by the polarimetric radar for tornado outbreaks on 6 May 2012. CAS/JSC WGNE Research Activities in Atmospheric and Oceanic Modelling. 2015;1:27–1.28.Google Scholar
- Yussouf N, Mansell ER, Wicker LJ, Wheatley DM, Stensrud DJ. The ensemble Kalman filter analyses and forecasts of the 8 May 2003 Oklahoma City tornadic supercell storm using single- and double-moment microphysics schemes. Monthly Weather Review. 2013;141:3388–412.View ArticleGoogle Scholar