Abstract
An intense rainband associated with Typhoon 1326 (Wipha) induced a fatal debris flow on Izu Oshima, Japan, on October 15 – 16, 2013. This rainband formed along a local front between the southeasterly humid warm air around the typhoon and the northeasterly cold air from the Kanto Plain. In this paper, the Japan Meteorological Agency Nonhydrostatic Model was optimized for the “K computer”, and ultra-high-resolution (500 – 250 m grid spacing) numerical simulations of the rainband with a large domain were conducted. Two of main factors that affect a numerical weather prediction (NWP) model, (1) grid spacing and (2) planetary boundary layer (PBL) schemes [Mellor–Yamada–Nakanishi–Niino (MYNN) and Deardorff (DD)], were investigated. Experiments with DD (Exps_DD: grid spacings of 2 km, 500 m, and 250 m) showed better reproduc-ibility of the rainband position than experiments with MYNN (Exps_MYNN: grid spacings of 5 km, 2 km, and 500 m). Exps_DD simulated distinct convective-scale up/downdraft pairs on the southeast/northwest sides of the front, whereas those of Exps_MYNN were not clear. Exps_DD yielded stronger cold pools near the surface than did Exps_MYNN. These differences in the boundary layer structures likely had a large impact on the position of the front and the associated rainband. Exps_DD with the 500-m grid spacing showed the best precipitation performance according to the Fractions Skill Score. To check other factors which influence precipitation forecast, model domain sizes, lateral boundary conditions in nesting simulations, and terrain representations were investigated. In the small domain experiments, the rain- shapes were very different from the observations. In the experiment using a nesting procedure, the deterioration of the forecast performance was acceptably reduced. The model with fine terrains reproduced the intense rain over the island. These results demonstrate that the ultra-high-resolution NWP model with a large domain has the possibility to improve predictions of heavy rain.
| Original language | English |
|---|---|
| Pages (from-to) | 25-54 |
| Number of pages | 30 |
| Journal | Journal of the Meteorological Society of Japan |
| Volume | 96 |
| Issue number | 1 |
| DOIs | |
| Publication status | Published - 2018 |
Keywords
- Heavy rainfall
- High-resolution NWP
- JMA-NHM
- K computer
Access to Document
Other files and links
Fingerprint
Dive into the research topics of 'Ultra-high-resolution numerical weather prediction with a large domain using the K computer: A case study of the Izu Oshima heavy rainfall event on october 15–16, 2013'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver
}
In: Journal of the Meteorological Society of Japan, Vol. 96, No. 1, 2018, p. 25-54.
Research output: Contribution to journal › Article › peer-review
TY - JOUR
T1 - Ultra-high-resolution numerical weather prediction with a large domain using the K computer
T2 - A case study of the Izu Oshima heavy rainfall event on october 15–16, 2013
AU - Oizumi, Tsutao
AU - Saito, Kazuo
AU - Ito, Junshi
AU - Kuroda, Tohru
AU - Duc, Le
N1 - Funding Information: A part of this work was supported by the Ministry of Education, Culture, Sports, Science and Technology as the Field 3, the Strategic Programs for Innovative Research (SPIRE), the FLAGSHIP2020 project(Advancement of meteorological and global environmental predictions utilizing observational “Big Data”), and the JSPS Grant-in-Aid for Scientific Research ‘Study of optimum perturbation methods for ensemble data assimilation’ (16H04054). The authors thank Dr. Hiromu Seko of the Meteorological Research Institute for his continuous support and encouragement. Computational results were obtained using the K computer at the RIKEN Advanced Institute for Computational Science (project ID: hp140220, hp150214, and hp160229). A part of input and output data were processed with UV1000 at the Japan Agency for Marine-Earth Science and Technology. Detailed and constructive comments on the manuscript by two anonymous reviewers and the editor, Dr. Masayuki Kawashima of the Hokkaido University, significantly improved the quality of this paper. We give thanks to Dr. Akira Noda of Japan Agency for Marine-Earth Science and Technology for providing useful comments. Funding Information: One NWP model, the Japan Meteorological Agency nonhydrostatic model (JMA-NHM; hereafter “NHM”), has implemented the MellorY amadaNakanishi? Niino (MYNN; Nakanishi and Niino 2004) scheme, which is regarded as one of the most sophisticated PBL schemes available. It is common to use the MYNN scheme when the horizontal resolution is larger than the boundary layer depth (e.g., the horizontal grid spacing is 2 km or more) such as in operational applications. However, when the horizontal resolution becomes sufficiently high to start resolving the boundary layer turbulence, an LES-based model such as the DD scheme seems more suitable. Kato (2011, 2012) investigated the dependency of snowfall forecasts on horizontal (0.5(�넋�5 km) and vertical grid spacings and turbulence schemes (MYNN level 3 and DD). In both studies, they found that the turbulent scheme had a larger impact on the precipitation amount than did the horizontal resolution. However, no experiments have been conducted using a high-resolution NWP model to compare MYNN and DD in heavy rain predictions. It is important to examine how MYNN and DD impact the representation of mesoscale convective systems and the quantitative forecasts of intense rain. Precipitation associated with a frontal system is often enhanced by orographic forcing. Oku et al. (2010) compared for two cloud-resolving models: NHM and the Weather Research and Forecasting model with similar model settings and topography data. They pointed out that not only the differences in model numerics and physics but also slight changes in the representation of topography induce significant differences in the representations of extreme weather. Nunalee et al. (2015) used a high-resolution numerical model (1-km grid spacing) to simulate actual cases of atmospheric flow past mountainous islands using three different topography datasets. They found that the use of accurate orographic boundary conditions is more advantageous when running high-resolution mesoscale models. Roberts et al. (2009) investigated the orographic enhancement of rain in England using the Met Office Unified Model (UM) with different grid spacings of 12, 4, and 1 km. Their results showed the benefit of increased resolution in the UM and the benefit of coupling high-resolution rainfall forecasts to the probability distributed model for flood warnings. In general, the model domain size is determined by a trade-off between the spatiotemporal scales of the targeted phenomena and the available computational resources. Several studies have investigated the influence of the domain size on precipitation predictions in climate models. Larsen et al. (2013) examined the regional climate model (HIRHAM) with different model domain sizes (1400(�넋�5500 km) and different horizontal grid spacings (5.5, 11, and 12 km) for precipitation simulations. They concluded that, the domain sizes were more important than the resolutions in reproducing the precipitation. Bray et al. (2011) investigated the sensitivity of the domain size and the buffer zone to the uncertainty of extreme rainfall downscaling using the Pennsylvania State University? National Center for Atmospheric Research fifth- generation mesoscale model (MM5). They noted that high-resolution experiment (500 to 250-m grid spac-the domain size and the buffer zone have a significant ing) with a large domain and to examine important impact on the model rainfall estimates. Davies (2014) factors for accurate forecasts of heavy rain events. The determined some general guidelines for lateral bound-target event is the Izu Oshima heavy rain that occurred ary condition (LBC) strategies. These strategies imply in October 2013 when a strong typhoon approached that, for a small domain, forecast fields are largely East Japan. The heavy rain was brought on by a rain-determined by LBCs and as a consequence low update band associated with the stationary front, which was frequencies of LBCs cause a loss of information in maintained by the generation of new convection cells forecast fields. on the inflow (warm) side; however, its horizontal Wang et al. (2016) tested the sensitivity of heavy scale of approximately 330 km was larger than those precipitation to various model configurations using described in previous studies of line-shaped precipita-the Australian Community Climate and Earth-Sys-tion with back-building formation. tem Simulator via a series of convection-permitting To accomplish ultra-high-resolution experiments simulations (0.5(�넋�4 km) for three heavy precipitatiowni th a large domain, first we optimized NHM for the cases. They showed that impacts of model resolutions K computer and modified the pre-and post-processes and domain sizes on quantitative precipitation forecast of the NHM. can be compared with impacts of physical param-The important factors, (1) the grid spacing (5 km, eterization schemes and initial conditions. These 2 km, 500 m, and 250 m) and (2) the PBL scheme previous studies independently suggest the benefits (MYNN levels 2.5 and 3 and DD), for heavy rain pre-of finer resolutions and model domains in selected diction were investigated using the optimized NHM cases. However, no study has been conducted for with a large domain (1600 km × 1100 km). To discuss ultra-high-resolution simulations (100-m scale) with the large model domain impacts, we investigated the large model domains for NWP. Such a high-resolution impacts of experiments with a small model domain simulation with a large domain experiment requires (200 km2) and LBCs on the rainband formation. The huge computational resources. However, using a large sensitivity analyses of the model terrain were con-domain has the benefit of reducing the impact of the ducted with two terrain datasets made from different boundary conditions. digital elevation maps. To obtain rigorous results, In Japan, the “K computer” owned by RIKEN, some significant factors were examined with respect whose performance in the LINPACK (LINear equa-to different initial times . The obtained results were tions software PACKage) benchmark floating-point quantitatively evaluated using the Fractions Skill operation was ranked first in the world’s top super-Score. computers in June 2011, and has been used for This paper is organized as follows. In Section 2, the research since September 2012. Half of the computa-details of the Izu Oshima heavy rain event in October tional resources of the K computer has been assigned 2013 are described. In Section 3, the optimization to “High-Performance Computing Infrastructure of the NHM for the K computer is described, and Strategic Program for Innovative Research (SPIRE)” the design of the experiments is presented. Section funded by the Ministry of Education, Culture, Sports, 4 shows the results of the simulations, and the main Science, and Technology (MEXT) of Japan. As a sub factors determining the rainband position are exam-subject of one of the five strategic fields of SPIRE, ined. In Section 5, the impacts of the model domain “Ultra-high Precision Meso-Scale Weather Prediction” size and the LBCs and the different representations of has been conducted (Saito et al. 2013) with three sub-the terrain are discussed. The robustness of this study themes: 1) the development of cloud-resolving four-was investigated for different initial times, and a qual-dimensional data assimilation systems, 2) the devel-itative verification of the precipitation is shown. The opment and validation of a cloud-resolving ensemble summary and concluding remarks are given in Section analysis and forecast system, and 3) basic research 6. Funding Information: A part of this work was supported by the Ministry of Education, Culture, Sports, Science and Technology as the Field 3, the Strategic Programs for Innovative Research (SPIRE), the FLAGSHIP2020 project (Advancement of meteorological and global environmental predictions utilizing observational “Big Data”), and the JSPS Grant-in-Aid for Scientific Research ‘Study of optimum perturbation methods for ensemble data assimilation’ (16H04054). The authors thank Dr. Hiromu Seko of the Meteorological Research Institute for his continuous support and encouragement. Publisher Copyright: © The Author(s) 2018.
PY - 2018
Y1 - 2018
N2 - An intense rainband associated with Typhoon 1326 (Wipha) induced a fatal debris flow on Izu Oshima, Japan, on October 15 – 16, 2013. This rainband formed along a local front between the southeasterly humid warm air around the typhoon and the northeasterly cold air from the Kanto Plain. In this paper, the Japan Meteorological Agency Nonhydrostatic Model was optimized for the “K computer”, and ultra-high-resolution (500 – 250 m grid spacing) numerical simulations of the rainband with a large domain were conducted. Two of main factors that affect a numerical weather prediction (NWP) model, (1) grid spacing and (2) planetary boundary layer (PBL) schemes [Mellor–Yamada–Nakanishi–Niino (MYNN) and Deardorff (DD)], were investigated. Experiments with DD (Exps_DD: grid spacings of 2 km, 500 m, and 250 m) showed better reproduc-ibility of the rainband position than experiments with MYNN (Exps_MYNN: grid spacings of 5 km, 2 km, and 500 m). Exps_DD simulated distinct convective-scale up/downdraft pairs on the southeast/northwest sides of the front, whereas those of Exps_MYNN were not clear. Exps_DD yielded stronger cold pools near the surface than did Exps_MYNN. These differences in the boundary layer structures likely had a large impact on the position of the front and the associated rainband. Exps_DD with the 500-m grid spacing showed the best precipitation performance according to the Fractions Skill Score. To check other factors which influence precipitation forecast, model domain sizes, lateral boundary conditions in nesting simulations, and terrain representations were investigated. In the small domain experiments, the rain- shapes were very different from the observations. In the experiment using a nesting procedure, the deterioration of the forecast performance was acceptably reduced. The model with fine terrains reproduced the intense rain over the island. These results demonstrate that the ultra-high-resolution NWP model with a large domain has the possibility to improve predictions of heavy rain.
AB - An intense rainband associated with Typhoon 1326 (Wipha) induced a fatal debris flow on Izu Oshima, Japan, on October 15 – 16, 2013. This rainband formed along a local front between the southeasterly humid warm air around the typhoon and the northeasterly cold air from the Kanto Plain. In this paper, the Japan Meteorological Agency Nonhydrostatic Model was optimized for the “K computer”, and ultra-high-resolution (500 – 250 m grid spacing) numerical simulations of the rainband with a large domain were conducted. Two of main factors that affect a numerical weather prediction (NWP) model, (1) grid spacing and (2) planetary boundary layer (PBL) schemes [Mellor–Yamada–Nakanishi–Niino (MYNN) and Deardorff (DD)], were investigated. Experiments with DD (Exps_DD: grid spacings of 2 km, 500 m, and 250 m) showed better reproduc-ibility of the rainband position than experiments with MYNN (Exps_MYNN: grid spacings of 5 km, 2 km, and 500 m). Exps_DD simulated distinct convective-scale up/downdraft pairs on the southeast/northwest sides of the front, whereas those of Exps_MYNN were not clear. Exps_DD yielded stronger cold pools near the surface than did Exps_MYNN. These differences in the boundary layer structures likely had a large impact on the position of the front and the associated rainband. Exps_DD with the 500-m grid spacing showed the best precipitation performance according to the Fractions Skill Score. To check other factors which influence precipitation forecast, model domain sizes, lateral boundary conditions in nesting simulations, and terrain representations were investigated. In the small domain experiments, the rain- shapes were very different from the observations. In the experiment using a nesting procedure, the deterioration of the forecast performance was acceptably reduced. The model with fine terrains reproduced the intense rain over the island. These results demonstrate that the ultra-high-resolution NWP model with a large domain has the possibility to improve predictions of heavy rain.
KW - Heavy rainfall
KW - High-resolution NWP
KW - JMA-NHM
KW - K computer
UR - http://www.scopus.com/inward/record.url?scp=85042279037&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85042279037&partnerID=8YFLogxK
U2 - 10.2151/jmsj.2018-006
DO - 10.2151/jmsj.2018-006
M3 - Article
AN - SCOPUS:85042279037
SN - 0026-1165
VL - 96
SP - 25
EP - 54
JO - Journal of the Meteorological Society of Japan
JF - Journal of the Meteorological Society of Japan
IS - 1
ER -