Norihiko Sugimoto, Minh Tuan Pham, Kanta Tachibana, Tomohiro Yoshikawa, and Takeshi Furuhashi
Abstract We propose a high speed non-empirical method to detect centers of tropical cyclones, which is useful to identify tropical cyclones in huge climatology data. In this method, centers of tropical cyclones are detected automatically by iteration of streamline in down-stream direction from some initial positions. We also bend the path of streamline successively to converge on the center of tropical cyclone rapidly. Since this method is free from empirical conditions used in the conventional method, the accuracy is independent of these conditions. Moreover, because the proposed method does not need to check these at all grid points, computational cost is significantly reduced. We compare the accuracy and effectiveness of the method with those of the conventional one for tropical cyclone identification task in observational data. Our method could find almost all tropical cyclones, some of which were not identified by the conventional method. This method will be useful for future huge climatology data, since computational cost does not depend on the number of grid points.
Global warming is a very important issue in the climate research fields. As the Intergovernmental Panel on Climate Change (IPCC 2007) reported, global warming gives not only increase of global average temperature, but change of frequencies and intensities of extreme events. There is a social need to study causality between global warming and these extreme events, since their impact on our life and environment will be serious in future climate.
Among these extreme events, tropical cyclone is one of the most important topics. It is argued eagerly that the effect of global warming on the frequencies and intensities of the tropical cyclones [Emanuel 1987, Holland 1997]. In order to predict future climate change and extreme events such as tropical cyclones, many
J.B. Elsner and T.H. Jagger (eds.), Hurricanes and Climate Change, 251
doi: 10.1007/978-0-387-09410-6, © Springer Science + Business Media, LLC 2009
numerical investigations has been done using general circulation model [Broccoli and Manabe 1990, Haarsma et al. 1993, Bengtsson et al. 1996, Krishnamurti et al. 1998, Sugi et al. 2002, Yoshimura et al. 2006, Bengtsson et al. 2007]. However, there is some ambiguity in these studies because of the parameterization process due to finite resolution of the numerical model. Some studies suggest that while tropical cyclones will appear less frequently, intense tropical cyclones will appear more frequently. It is necessary to use a higher resolution model to discuss features of future tropical cyclones in detail. Although recent numerical investigation achieves fine grid size about 20 km [Oouchi et al. 2006], further investigation will be needed to have a consensus understanding of future tropical cyclones. This tendency to develop and use high resolution model is accelerating and will continue.
One of the main weak points of these high resolution numerical studies is that there is no definite method to identify tropical cyclones in huge output data. In the case of the observational data, the conventional way is an empirical identification by experts using satellite images. However, it is not reasonable for an identifier of tropical cyclones in the model output data to create surrogate images to satellite ones, since the reliable data are distributed uniformly at all grid points. There is a room of exploring an alternative method to identify tropical cyclones effectively. Although several methods have been proposed to identify tropical cyclones in the model output data [Haarsma et al. 1993, Bengtsson et al. 1995, Sugi et al. 2002, Oouchi et al. 2006], these methods have some problems as the following. Since these methods use empirical conditions of criteria, first it is difficult to determine criteria themselves. Second, there is some possibility to overlook and miss the tropical cyclones in the datasets, since the accuracy depends on these criteria. Third, as these conditions need to be checked at all grid points, the computational cost is enormously in the case of high resolution model. The finer the model resolution is, the larger the size of the output data becomes. Thus, sometimes we have to analyze several gigabyte data to discuss future tropical cyclones induced by climate change [Oouchi et al. 2006].
In the present study, we propose a new high speed non-empirical method to detect centers of vortices automatically, which enables us to identify tropical cyclones in huge climatology data. To perform this method effectively, we first transform coordinates. Since a tropical cyclone has its wind flow as concentric circles, we make use of streamline. For effective searching, we bend the path of streamline successively. This method allows us to save large amount of computational time even in the case of high resolution model, since our method does not need to check conditions at all grid points.
The rest of this paper is organized as follows. In section 2, we show several methods used in this study. First, empirical conditions of criteria used in the conventional method are shown. Then we propose a new high speed non-empirical method to detect centers of tropical cyclone automatically. In Section 3, we compare the effectiveness of the proposed method with those of the conventional one for tropical cyclone identification task in the observational data. We give summary and future works in Section 4.
In the present study, we use datasets of National Centers for Environmental Prediction (NCEP) global reanalysis project [Kalnay et al. 1996] as a meteorological data. The horizontal resolution is 2.5°lat x 2.5°lon grid, and the total grid number is 10244. In the conventional method of identifying tropical cyclones, fourth-daily geopotential height, horizontal velocity, and temperature at several pressure levels are used. In the proposed method, we only use fourth-daily horizontal velocity at one pressure level, i.e. either of 850[hPa] or 1000[hPa]. Although mesh size of this data is too coarse to express structures of tropical cyclones, this dataset is enough to detect tropical cyclone-like phenomena. In the present study, we emphasize the accuracy of the proposed method even in the coarse resolution.
We also use ''best track'' dataset as a correct answer to evaluate the accuracy of the methods. This dataset is official record of tropical cyclones in Northwest Pacific, compiled by the Japan Meteorological Agency since 1951. The dataset contains records such as center position, central pressure, and maximum sustained wind speed for every three to six hours. We regard the data as ground truth data, since these data are provided and assessed by the experts after receiving all information of tropical cyclones. We compare the identification of tropical cyclones in the several methods with the right answer of ''best track''.
First, we introduce the conventional method of tropical cyclone identification. Traditionally, tropical cyclones are identified by several empirical conditions. For example, [Bengtsson et al. 1995] use the following 5 conditions of criteria. For the position i,
(2) H/000 < min¡2Nm H/000 and above j 3k 2 N(j, 48), |wf00| > 15.
3 A700 , A500 , A300 > 3 (3) Ai,48 + Ai,48 + Ai,48 > 3:
Was this article helpful?