An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi”

Otsuka, S. et al.

Paper: An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi”

Shigenori Otsuka∗, Seiya Nishizawa∗∗, Takeshi Horinouchi∗∗∗, and Shigeo Yoden∗

∗Division of Earth and Planetary Sciences, Graduate School of Science, Kyoto University Kitashirakawa-Oiwake-cho, Sakyo, Kyoto 606-8502, Japan E-mail: [email protected] ∗∗RIKEN Advanced Institute for Computational Science 7-1-26, Minatojima-minami-machi, Chuo-ku, Kobe, Hyogo 650-0047, Japan ∗∗∗Division of Earth System Science, Graduate School of Environmental Earth Science, Hokkaido University Kita-10, Nishi-5, Kita-ku, Sapporo 060-0810, Japan [Received October 10, 2012; accepted November 22, 2012]

Gfdnavi is a web-based data and knowledge server Ensemble numerical weather prediction (NWP), which program for geophysical fluid data that constructs has been conducted operationally for nearly two decades databases, provides analysis and visualization tools, [1], has enabled us to quantify the uncertainty of predic- and shares knowledge documents. A new Gfdnavi user tion. This kind of probabilistic information is thought to interface for analyzing and visualizing data on web be useful for both appropriate disaster management and browsers is developed to improve the user experience efficient economic activities. The problem of data size be- by providing seamless analysis and visualization op- comes more serious, however, because an ensemble pre- erations, multiple diagram editing, a layer function, diction system performs a lot of simulation (10-100 runs) and so on. An experimental data handling system for at the same time. ensemble numerical weather prediction data is con- Another problem in ensemble prediction is that the structed using Gfdnavi to address such issues as data handling of ensemble data is complicated compared with processing and transfer between weather centers and classical deterministic forecasts. For data transfer from decision makers in various sectors, including that of weather centers to decision makers in various sectors, in- disaster management. Special tools to analyze and vi- cluding that for disaster management, ensemble predic- sualize ensemble numerical weather prediction data tion data must be processed into appropriate formats that are implemented as user-defined Gfdnavi plug-ins. An are easily understood by people who are not NWP special- interactive document that provides basic ideas of how ists. In recent years, this issue is being addressed by the to utilize probabilistic ensemble data information is World Meteorological Organization under the Observing written with the Gfdnavi knowledge documentation System Research and Predictability Experiment (THOR- system, in which hyperlinks enable users to edit dia- PEX) [2]. grams in the document. In this paper, the problem in handling large data is addressed by using Gfdnavi (Geophysical fluid data navi- Keywords: analysis, visualization, database, web appli- gator1), which is a web-based data and knowledge server cation, ensemble numerical weather predictions program for geophysical fluid data [3, 4]. A brief introduction of Gfdnavi and details of a new user interface of 1. Introduction Gfdnavi for the analysis and visualization functions will be described in section 2. In the last century, data types and data sizes in me- To solve the problem of processing ensemble data, we teorology and related fields have been increasing. The develop tools for easy handling of ensemble prediction output of numerical models, reanalysis data, and satellite data. These tools are implemented as user-defined plug- data, for example, are commonly used in recent years. ins of Gfdnavi. A user’s guide for the tool has also been Due to a rapid increase in computational resources and prepared using the knowledge documentation system of advances in satellite observations, the spatiotemporal res- Gfdnavi. This will be described in section 3. olution of these datasets is increasing, as is the types of physical quantities. Data handling is thus getting harder 2. Gfdnavi and downloading whole datasets to local computers is al- most impossible. Data formats vary from data to data, 2.1. Overview which requires extra cost to learn how to handle them. In Gfdnavi can be operated by large data centers, small the field of disaster management, huge amounts of data of groups, and personal users. End users who want to ana- various types are needed. This makes the problem worse. 1. http://www.gfd-dennou.org/arch/davis/gfdnavi/

48 Journal of Disaster Research Vol.8 No.1, 2013 An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi”

lyze data need to access Gfdnavi servers. A typical ﬂow of operation by end users is as follows: 1 Open a web browser and access a server that operates Gfdnavi. 2 Find datasets in Finder or Search windows. When datasets are selected, an Analysis window opens. 3 Analyze and visualize data in the Analysis window. Users obtain image ﬁles as a result of visualization.

4 Write a knowledge document for results in Knowl- Fig. 1. Gfdnavi analysis and visualization window. Two edge window. diagrams are shown. 5 Write a Ruby script (advanced users) to automate the (2)-(4) process. On Gfdnavi servers, Gfdnavi scans selected directories to make a database of metadata in data files of binary formats that are commonly used in the field of meteorology, such as NetCDF, GRIB, and GrADS. The database is used for a web-based viewer. The user interface over web browsers is intuitive and a web service programing interface is also provided for advanced users. Gfdnavi provides server-side data analysis and visualization tools. As long as one uses server-side tools, including user-defined plug-ins, one does not need to down- load original data, and only results, i.e., image files, are transferred from the server to local machines. For those who need advanced analysis and visualization on local machines, downloading a minimum set of analyzed (processed) data is available. Gfdnavi further provides a knowledge documentation system that stores documents into the same database. Users make documents for diagrams that are created with Gfdnavi. Hyperlinks that enable users to reproduce the same diagrams on Gfdnavi are automatically created in Fig. 2. Input tab√ window for the left diagram in Fig. 1. documents. The database of knowledge documents en- A zonal mean of u2 + v2 is obtained with this operation. ables users to share knowledge. Gfdnavi is implemented White arrows show the operation data flow. using the Ruby interface of GFD-Dennou library and Ruby on Rails. See details in Horinouchi et al. (2010) [3] and Nishizawa et al. (2010) [4]. In the next subsection, the new features introduced to appears with default parameters for visualization. Users Gfdnavi version 2.2 will be described. need to change parameters to get what they want to draw. The four buttons atop each diagram – Input, Method & Axis, Options, Action – open tab windows to tune analy- 2.2. Analysis and Visualization User Interface sis and visualization parameters. In the analysis of meteorological data, it is necessary to Users edit multiple diagrams at the same time (Fig. 1), compare multiple quantities and multiple cross-sections play animation, and overlay diagrams. The green panel at the same time. In the new user interface of Gfdnavi, on the left shows selected variables, axes of data, and so parameters for visualization, which were shared by all of on. The button at the top left corner is used to create a the diagrams in the previous version, can be set for each knowledge document with currently editing diagrams. diagram to achieve this. Functions are implemented by Figure 2 shows the Input tab window used to select object-oriented programing in which parameters for a dia- input data of a diagram in which multiple variables are gram is defined as attributes of the object for the diagram, used as input to an analysis method, i.e., a subroutine relating the user interface and the internal expression di- √used to apply mathematical operations. A zonal mean of rectly. u2 + v2 is obtained in the case of Fig. 2.Dataflowin Figure 1 shows the user interface for analysis and vi- analysis is displayed in this window. Users edit variables sualization. When this window is opened for the first and types of operations directly in window, which enables time after the selection of data, a diagram automatically users to operate intuitively. Users perform seamless anal-

Journal of Disaster Research Vol.8 No.1, 2013 49 Otsuka, S. et al.

Fig. 3. Method & Axis tab window for a diagram with two Fig. 5. Action tab window for the left diagram in Fig. 1. layers with the tab user interface for Layer 0 selected.

for visualization (Fig. 4). The General options category contains parameters, such as diagram size and map pro- jections, whereas the Speciﬁc options category contains options for each draw method, i.e., line types of contours, tone colors, and so on. Figure 5 shows the Action tab window for the left ﬁg- ure in Fig. 1. When editing of a diagram is completed, users take various actions for the diagram. Users down- load Ruby scripts that reproduce the same diagrams when executed on the console. By editing downloaded scripts, users perform advanced analysis using Ruby scripting language [4]. Overlaying different diagrams is available through layer functionality. Fig. 3 shows the Method & Axis tab window for a diagram with two layers. Although it is not easy to see the main diagram, which is located beneath the tab window, in Fig. 3, the bottom layer, Layer 0, shows temperature by color tone and contour, whereas the top layer, Layer 1, shows wind vectors. Visualization parameters for each layer can be edited separately through the tab user interface in parameter windows. Layers of diagrams can be added, removed, and swapped, like the layer function of graphic editing programs. Fig. 4. Options tab window for the left diagram in Fig. 1.

3. Application to Ensemble NWP Data ysis and visualization operations through this user interface. The exponential growth of computer power allows us to The Method & Axis tab window provides the user in- perform high-resolution forecasts and increase the num- terface for selecting the Draw method, subroutines used ber of ensemble members. Accurate prediction by ensem- to draw various types of diagrams. In the case of Fig. 3, ble NWP systems would help risk management. Proba- tone contour is selected to draw color tones with con- bilistic weather information also helps decision making tour lines. The Method & Axis tab window also pro- in other sectors. Speciﬁcally, commercial decisions are vides the user interface for specifying areas to be drawn often made, not on the basis of events which are likely for two-dimensional diagrams, variable range for one- to occur, but on the basis of events which are unlikely to dimensional diagrams, and slice points of data. Axes that occur, but which, if they did occur, would involve serious are used for the abscissa and ordinate of the diagram are ﬁnancial loss [5]. also selected in this window. As mentioned in section 1, raw output of ensemble The Options tab window provides parameter settings NWPs is not suitable for dissemination to the public be-

50 Journal of Disaster Research Vol.8 No.1, 2013 An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi”

Fig. 6. End-to-end forecast system as a future global interactive forecast system (WMO 2005) [2].

cause of the larger sizes compared to deterministic NWPs The Japan Meteorological Agency (JMA) Meteorologi- and difficulties in interpretation. This requires data con- cal Research Institute has conducted experimental ensem- version to a user-friendly form before dissemination. Data ble NWPs and storm surge simulation on Cyclone Nargis types and formats required by users vary depending on and used the output is as test data for tools to analyze and their purpose. An ensemble-mean horizontal distribution visualize ensemble NWP data, as detailed in the subsec- might, for example, be the best in some cases, whereas tions that follows. Data is also used to create an interac- statistics on extreme values at a particular point might be tive document on how to utilize probabilistic information helpful in other cases. We have to consider not only the obtained from ensemble NWP data. development of ensemble prediction systems but also the The atmospheric model we use is the JMA nonhydro- development of decision support systems to utilize output static model (NHM) [9]. The period of the experiment is from ensemble NWPs. The decision support system con- from April 30, 2008, 1200 UTC, to May 3, 1200 UTC. veys information from the NWP sector to decision mak- The horizontal resolution is 10 km. Initial and boundary ers in social, economic, and environmental sectors in an conditions for the NHM control run are taken from the appropriate way. This idea is shown as a schematic dia- JMA’s high-resolution global analysis and forecast. Initial gram of the end-to-end forecast system in the THORPEX and boundary perturbations are produced from the JMA’s framework (Fig. 6)[2]. one-week global ensemble prediction system. In total, the For this purpose, we started to develop an experimental ensemble size is 21 – control run and 20 perturbed runs. data handling system for decision support. We have de- The Princeton ocean model (POM) [10] is used for storm veloped tools to store ensemble NWP data in a database surge simulation. Input to POM is output from NHM, i.e., and handle probabilistic information obtained from en- wind at 10 m and sea level pressure. The horizontal reso- semble NWP data, depending on the purpose. Here, Gfd- lution in POM is 3.5 km. See Saito et al. (2010) [11] and navi is suitable to build such a system because Gfdnavi Kuroda et al. (2010) [12] for simulation details. has a database system with analysis, visualization, and knowledge-sharing tools. 3.2. Analysis and Visualization of Ensemble NWP In recent years, it has become possible to install ensem- Data ble NWP systems in developing countries and regions [6]. Gfdnavi is designed to accept user-defined plug-ins. As mentioned above, decision support systems for ensem- Here, plug-ins of analysis and visualization tools for en- ble NWP data are also required at the same time. One of semble NWP data are implemented. Fig. 7 shows a drop- the purposes of the current study is to demonstrate how to down menu for selecting visualization methods on Gfd- develop such a system with free open-source programs. navi. Visualization tools for ensemble prediction data, This section briefly introduces the data handling system ensemble 1D and ensemble 2D, are highlighted by red for ensemble NWP data. ellipses. Users do not need to specify details for mathematical operations to visualize ensemble prediction data 3.1. Test Case: Cyclone Nargis because they are operated within these visualization tools. 2 Cyclone Nargis hit Myanmar in May 2008, killing This enables a quick look at ensemble NWP data. more than 13,800 people and causing economic loss re- 2. The latest versions of tools are available from the web page: http://www- ported to be more than 10 billion dollars. This is the worst mete.kugi.kyoto-u.ac.jp/otsuka/gfdnavi ensemble tools/ (the address is natural disaster in Myanmar’s recorded history [7, 8]. subject to change).

Journal of Disaster Research Vol.8 No.1, 2013 51 Otsuka, S. et al.

Fig. 7. Drop-down menu for selecting visualization methods. Special methods for ensemble prediction data are highlighted by red ellipses.

Fig. 8. Horizontal distribution of maximum sea surface elevation for all time steps and ensemble members of POM 3.2.1. 1D Diagrams of Ensemble NWP Data simulation.

In this section, 1D visualization methods for ensemble NWP data are demonstrated using sea surface elevation mum, and standard deviation are shown. With this plot, it simulated by POM. is possible to see the distribution of extreme values (out- Figure 8 shows the horizontal distribution of maximum liers) that are not shown in the spread. sea surface elevation for all of the time steps and all of the The bottom-right panel in Fig. 9 shows percentiles of ensemble members. The highest sea surface elevation is ◦ ◦ distribution for the same data. For example, most of the obtained at 95.07 E, 16.1 N, which is the mouth of the members show a sea surface elevation of about 1 m at Irrawaddy River (Ayeyarwady River). Although the max- around May 2, 0000 UTC, whereas some members show imum expected sea surface elevation is valuable for disas- a sea surface elevation of nearly 4 m. After May 2, 0000 ter management, it is more valuable for obtaining proba- UTC, in contrast, ensemble members show a variety of bilistic information from the same data. The time evolu- sea surface elevations. tion of sea surface elevation at the mouth of the Irrawaddy As seen above, we can extract many kinds of informa- River will be shown next to quantify probability. tion from 1D diagrams of the same ensemble NWP data The top-left panel in Fig. 9 is a plume diagram of sea by using different plotting methods. surface elevation at the mouth of the Irrawaddy River. A plume diagram consists of superimposed 1D line plots. Typically, time series of all of the ensemble members are 3.2.2. 2D Diagrams of Ensemble NWP Data superimposed. At the initial time, differences among en- In this section, 2D visualization methods for ensemble semble members are small. As time evolves, differences NWP data are demonstrated using the sea level pressure become larger – the reliability of the forecast becomes simulated by NHM. lower –, which makes superimposed line plots spread Three kinds of 2D plots for ensemble NWP data away from each other. (This looks like the plume rising are implemented: ensemble mean and spread, so-called from a chimney.) In this ﬁgure, the ensemble member spaghetti plot, and distributions of probability exceeding that shows the largest sea surface elevation is highlighted a threshold value. The spaghetti plot shows the same in red. The value and timing of the maximum sea surface isopleth lines for all of the ensemble members in a sin- elevation vary much from member to member. gle plot. The higher density of contour lines means the The top-right panel in Fig. 9 shows the ensemble mean higher reliability of the forecast. The lower density of and spread (standard deviation) for the same data. The lines, which shows patterns similar to spaghetti on a plate, spread becomes larger as the sea surface elevation be- means lower reliability. A spaghetti plot of the geopoten- comes higher. tial height at 500 mb with isopleth lines of 5640 m, for The bottom-left panel in Fig. 9 shows a box plot for the example, is shown as a cover illustration of Kalnay (2003) same data. A box plot or box-and-whisker plot is intro- [15]. duced by Tukey (1977) [13]. The original box plot shows Figure 10 shows a spaghetti plot of sea level pressure ﬁve quantities: minimum, lower quartile, median, upper with contour lines of 1000 hPa at the initial time, 24 hours, quartile, and maximum. There are many variations [14]. and 48 hours. As time evolves, the low-pressure region, In the bottom-left panel in Fig. 9, the maximum, mini- which corresponds to Cyclone Nargis, becomes larger and

52 Journal of Disaster Research Vol.8 No.1, 2013 An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi”

Fig. 9. Time series of sea surface elevation at the Irrawaddy River mouth simulated by POM. Diagrams are produced by the tool for one-dimensional visualization of ensemble prediction data. Plume (top left), mean and spread (top right), box and whisker (bottom left), and percentile (bottom right).

Fig. 10. Spaghetti diagram of sea level pressure simulated by NHM at t = 0 (red), 24 (green), and 48 h (blue) with contour lines of 1000 hPa produced by the tool for two-dimensional visualization of ensemble prediction data.

the difference between different members becomes larger. If one makes decisions based only on a single fore- Speciﬁcally, the eastward movement speed of the cyclone cast, the forecast may show minor damage by chance. differs from member to member, which makes differences Some ensemble members, for example, show a maxi- in value and the timing of the maximum sea surface ele- mum sea surface elevation of about 2 m (Fig. 9, top left). vation shown in Fig. 9. Shibayama et al. (2009) [16] reported that water level deviation due to the storm surge was estimated at 3-4 m from 3.2.3. Importance of Advanced Utilization of Ensem- a ﬁeld survey along the Yangon River. If a single forecast ble NWP Data that predicts the maximum sea surface elevation of 2 m From Fig. 9, we point out the importance of utilizing is used for decision making, decision makers may fail to information from all of the ensemble members. make appropriate decisions.

Journal of Disaster Research Vol.8 No.1, 2013 53 Otsuka, S. et al.

Fig. 11. Examples of the diagrams in the interactive document on the decision support system.

If only the ensemble mean is available for decision visualization functions of the system. Generic visualiza- making, the maximum sea surface elevation becomes tion functions include 1D plots (line, marker, and bar), 1.6 m at May 2, 0100 UTC (Fig. 9, top right), which is 2D plots (contour, tone, and vector arrow), and statisti- lower than maximum sea surface elevations of individual cal plots (scattergram and histogram). Specifically, how ensemble members and much lower than estimation from to extract information by slicing multidimensional data a field survey. Although the ensemble mean is often used along each axis is demonstrated for five-dimensional en- as a product of ensemble predictions, this clearly shows semble NWP data as a function of latitude, longitude, that the ensemble mean is not suitable for the current pur- height, time, and ensemble member. pose. Visualization functions specific to ensemble NWP data If information on all of the ensemble members can be include 1D diagrams such as plume diagrams and 2D dia- utilized, a sea surface elevation of nearly 4 m is predicted grams such as the spaghetti plot are described. How to on May 1, 2100 UTC, and May 2, 0100 UTC (Fig. 9, compute the ensemble mean and spread by combining top left), which is comparable to the value obtained by analysis tools in Gfdnavi is explained as analyses with a field survey. It is possible, in addition, to say that the simple mathematical operations. probability that the sea surface elevation exceeds 2 m is Sample diagrams shown in the tutorial document are more than 31% (Fig. 9, bottom right). also provided in an interactive document on the experi- As demonstrated here, it is important to fully utilize en- mental analysis system using the knowledge documenting semble prediction data to provide meaningful information system of Gfdnavi (Fig. 11). A hyperlink just below each for decision making. diagram, Redraw this image, opens the Analysis window. When the window opens, the same diagram with the same parameters appears. This enables us to modify parameters 3.3. User’s Guide of Tools for Ensemble NWP Data of sample diagrams and redraw them. We created an interactive tutorial document on how to This also enables us to save diagrams on the database analyze ensemble NWP data by using the experimental and write documents on diagrams. By using this tool, it is analysis system built into Gfdnavi to obtain probabilistic possible to archive and share knowledge that is needed to information that is useful for the prevention and mitiga- analyze and interpret NWP data. Comment functionality tion of meteorological disasters. also helps communication with other users. By archiving The user’s guide consists of systematic explanations of many case studies, the system will provide useful infor-

54 Journal of Disaster Research Vol.8 No.1, 2013 An Experimental Data Handling System for Ensemble Numerical Weather Predictions Using a Web-Based Data Server and Analysis Tool “Gfdnavi” mation and tools for specialists in ensemble NWP data ming interface,” Lecture Notes in Computer Science, Vol.6193, pp. and the system will also serve as a tool for education. 105-116, 2010. [5] T. N. Palmer, “The economic value of ensemble forecasts as a tool for risk assessment: From days to decades,” Quar. J. Roy. Meteor. Soc., Vol.128, pp. 747-774, 2002. [6] S. Yoden, K. Saito, T. Takemi, and S. Nishizawa, “New stages of 4. Summary international collaborations which serves for mitigation of meteorological disasters in Southeast Asia,” Tenki, Vol.55, pp. 705-708, We have developed a new Gfdnavi user interface for 2008 (in Japanese). [7] P. J. Webster, “Myanmar’s deadly daffodil,” Nature Geoscience, analysis and visualization that will improve the user ex- Vol.1, pp. 488-490, 2008. perience by seamless operations of analysis and visual- [8] H. M. Fritz, C. D. Blount, S. Thwin, M. K. Thu, and N. Chan, “Cy- ization. The new user interface enable users to edit multi- clone Nargis storm surge in Myanmar,” Nature Geoscience, Vol.2, ple diagrams at the same time and also enables overlaying pp. 448-449, 2009. [9] K. Saito, J. Ishida, K. Aranami, T. Hara, T. Segawa, M. Narita, multiple diagrams like a layer function of graphic editing and Y. Honda, “Nonhydrostatic atmospheric models and operational programs. development at JMA,” J. Meteor. Soc. Japan, Vol.85B, pp. 271-304, 2007. We have also introduced an experimental system de- [10] A. F. Blumberg and G. L. Mellor, “A description of a veloped on Gfdnavi to provide analysis and visualiza- three-dimensional coastal ocean circulation model,” In Three- dimensional coastal ocean models, pp. 1-16, American Geophysi- tion tools for ensemble numerical weather prediction data. cal Union, 1987. These tools will help data processing and the transfer of [11] K. Saito, T. Kuroda, M. Kunii, and N. Kohno, “Numerical simu- data from weather centers to decision makers in various lation of Myanmar cyclone Nargis and the associated storm surge Part II: Ensemble prediction,” J. Meteor. Soc. Japan, Vol.88, pp. sectors. An interactive document that provides basic ideas 547-570, 2010. of how to utilize probabilistic information on ensemble [12] T. Kuroda, K. Saito, M. Kunii, and N. Kohno, “Numerical simu- data is written with the knowledge documentation system lations of Myanmar cyclone Nargis and the associated storm surge Part I: Forecast experiment with a nonhydrostatic model and simu- of Gfdnavi in which automatic hyperlinks in the document lation of storm surge,” J. Meteor. Soc. Japan, Vol.88, pp. 521-545, enable users to edit diagrams in the document themselves. 2010. [13] J. W. Tukey, “Exploratory data analysis,” Addison-Wesley Publish- Although Gfdnavi provides a useful platform, it is nec- ing Co., 1977. essary to provide plug-in functionality depending on the [14] R. McGill, J. W. Tukey, and W. A. Larsen, “Variations of box plots,” needs of each user. The analysis and visualization tool for The American Statistician, Vol.32, pp. 12-16, 1978. ensemble prediction data is an example of such plug-ins. [15] E. Kalnay, “Atmospheric modeling, data assimilation, and predictability,” Cambridge University Press, 2003. We would like to improve the basic functions of Gfdnavi [16] T. Shibayama, H. Takagi, N. Hnu, and Y. Aoki, “Field survey of as well as plug-in tools for various needs. Improvement storm surge disaster due to cyclone Nargis in Myanmar,” J. Japan Soc. Civil Engineers B2 (Coastal Engineering), Vol.65, pp. 1376- of the documentation and user interface to edit plug-ins 1380, 2009 (in Japanese). is also important for enabling users to create plug-ins for them.

Acknowledgements This work was supported in part by MEXT Special Coordi- Name: nation Funds for Promoting Science and Technology for FY Shigenori Otsuka 2007-09 “International Research for Prevention and Mitigation of Meteorological Disasters in Southeast Asia” and Kyoto Uni- Affiliation: versity’s Global COE Program for FY 2009-13 “Sustainabil- Postdoctoral Researcher, RIKEN Advanced In- ity/Survivability Science for a Resilient Society Adaptable to stitute for Computational Science Extreme Weather Conditions.” We thank Dr. Kazuo Saito and Dr. Tohru Kuroda for providing experimental ensemble numerical prediction data. Address: 7-1-26 Minatojima-minami-machi, Chuo-ku, Kobe, Hyogo 650-0047, References: Japan [1] R. Mureau, F. Molteni, and T. N. Palmer, “Ensemble prediction us- Brief Career: ing dynamically conditioned perturbations,” Quar. J. Roy. Meteor. 2009 DSc in Graduate School of Science, Kyoto University Soc., Vol.119, pp. 299-323, 1993. 2009-2012 Research Fellow, Division of Earth and Planetary Sciences, [2] WMO, “Accelerating improvements in the accuracy of one-day to Graduate School of Science, Kyoto University two-week high-impact weather forecasts for the benefit of society, 2012- Project Assistant Professor, Educational Unit for Adaptation to the economy and the environment: THORPEX,” 2005, http://www. Extreme Weather Conditions, Center for the Promotion of Interdisciplinary wmo.int/pages/prog/arep/wwrp/new/documents/brochure e.pdf Education and Research, Kyoto University [accessed Jan. 17, 2013] 2013- Postdoctoral Researcher, RIKEN Advanced Institute for [3] T. Horinouchi, S. Nishizawa, C. Watanabe, A. Tomobayashi, S. Ot- Computational Science suka, T. Koshiro, Y.-Y. Hayashi, and GFD Dennou Club, “Gfd- Selected Publications: navi, web-based data and knowledge server software for geophysi- • cal fluid sciences, Part I: Rationales, stand-alone features, and sup- S. Otsuka and S. Yoden, “Temporal-spatial distribution of thin moist porting knowledge documentation linked to data,” Lecture Notes in layers in the midtroposphere over the tropical eastern Pacific,” J. Climate, Computer Science, Vol.6913, pp. 93-104, 2010. Vol.22, pp. 5102-5114, 2009. [4] S. Nishizawa, T. Horinouchi, C. Watanabe, Y. Isamoto, A. To- Academic Societies & Scientific Organizations: mobayashi, S. Otsuka, and GFD Dennou Club, “Gfdnavi, web- • Meteorological Society of Japan (MSJ) based data and knowledge server software for geophysical fluid sciences, Part II: RESTful web services and object-oriented program-

Journal of Disaster Research Vol.8 No.1, 2013 55 Otsuka, S. et al.

Name: Name: Seiya Nishizawa Shigeo Yoden

Afﬁliation: Afﬁliation: Research Scientist, RIKEN Advanced Institute Division of Earth and Planetary Sciences, Grad- for Computational Science uate School of Science, Kyoto University

Address: Address: 7-1-26 Minatojima-minami-machi, Chuo-ku, Kobe, Hyogo 650-0047, Kitashirakawa-Oiwake-cho, Sakyo, Kyoto 606-8502, Japan Japan Brief Career: Brief Career: 1983 Research Associate, Kyoto University 2005- Researcher, Kyoto University 1985-1987 JSPS Fellow for Research Abroad, University of Washington 2008- Researcher, Kobe University 1987 Associate Professor, Kyoto University 2009- Assistant Professor, Kobe University 2002- Professor of Meteorology, Kyoto University 2012- Research Scientist, RIKEN Advanced Institute for Computational Selected Publications: Science • “Atmospheric predictability,” J. Met. Soc. Japan, Vol.85B, pp. 77-102, Selected Publications: 2007. • S. Nishizawa and S. Yoden, “Distribution functions of a spurious trend in Academic Societies & Scientific Organizations: a finite length data set with natural variability: Statistical considerations • The Meteorological Society of Japan (MSJ) and a numerical experiment with a global circulation model,” • The American Meteorological Society (AMS) J.Geophys.Res., Vol.110, D12105, doi:10.1029/2004JD005714, 2005. • International Commission on the Middle Atmosphere, IAMAS/IUGG • S. Nishizawa, T. Horinouchi, C. Watanabe, Y. Isamoto, A. Tomobayashi, S. Otsuka, and GFD Dennou Club, “Gfdnavi, web-based data and knowledge server software for geophysical fluid sciences, Part II: RESTful web services and object-oriented programming interface,” In M. Yoshikawa et al. (Eds.) DASFAA 2010, LNCS, 6193, pp. 105-116, 2010. Academic Societies & Scientific Organizations: • Meteorological Society of Japan (MSJ) • The Japanese Society for Planetary Sciences (JSPS)

Name: Takeshi Horinouchi

Afﬁliation: Division of Earth System Science, Graduate School of Environmental Science, Hokkaido University

Address: Kita-10, Nishi-5, Kita-ku, Sapporo 060-0810, Japan

56 Journal of Disaster Research Vol.8 No.1, 2013