Vol.16 No.1 JOURNAL OF TROPICAL METEOROLOGY March 2009

Article ID: 1006-8775(2010) 01-077-05 VERIFICATION OF TYPHOON FORECASTS BY THE GRAPES MODEL

1 2 1 SONG Yv (宋 煜), YE Cheng-zhi (叶成志) , Zhen (黄 振), PU Wen-yao 3 (濮文耀)

(1. Dalian Meteorological Observatory, Dalian 116001 ; 2. Hunan Meteorological Observatory, Changsha 410007 China; 3. Dalian Weather Modification Office, Dalian 116001 China)

Abstract: Four landfalling typhoon cases in 2005 were selected for a numerical simulation study with the Global/Regional Assimilation and Prediction System (GRAPES) model. The preliminary assessment results of the performance of the model, including the predictions of typhoon track, landfall time, location and intensity, etc., are presented and the sources of errors are analyzed. The 24-hour distance forecast error of the typhoon center by the model is shown to be about 131 km, while the 48-hour error is 252 km. The model was relatively more skilful at forecasts of landfall time and locations than those of intensity at landfall. On average, the 24-hour forecasts were slightly better than the 48-hour ones. An analysis of data impacts indicates that the assimilation of unconventional observation data is essential for the improvement of the model simulation. The model could also be improved by increasing model resolution to simulate the mesoscale and fine scale systems and by improving methods of terrain refinement processing.

Key words: GRAPES model; landfalling typhoon; verification CLC number: P444 Document code: A doi: 10.3969/j.issn.1006-8775.2010.01.011

1 INTRODUCTION errors of typhoon paths issued by the Central Meteorological Observatory of China were 120 km Landfalls of typhoons in 2005 were and 198 km, respectively. Those of the Joint characterized by large numbers, strong winds, Typhoon Warning Center of the US were 110 km and torrential rainfall, and disasters such as floods, 187 km[2], respectively. The corresponding errors landslides, and mudflows, which claimed human were 101 km and 176 km for Japan [3] and 94 km and lives and caused huge financial and ecological 170 km for , China[4]. However, the losses. Statistics show that 91.4985 million people forecasting skill for typhoon intensity has shown no were affected, among these were included 244 dead similar improvement, as study on the impact of the and 57 missing. In China, the direct losses reached structure of tropical cyclones on the change in RMB75.7 billion, with the provinces of , intensity has been limited [5-6]. Fujian, and ranking at the top of the loss list. The model GRAPES began its experimental Among the landfalling typhoons, Haitang, Matsa, operation at the National Meteorological Center of Talim and Khanum were the most intense and [1] China during the rainy season of 2005 and has caused severe damage . provided good guidance for the forecasters at It is therefore of great importance to be able to different operational levels nationwide. With accurately predict the landfall of a typhoon. GRAPES, the most powerful typhoons of 2005, Knowing the track of the typhoon, in particular, is Haitang, Matsa, Talim and Khanum, were simulated crucial for disaster prevention and mitigation. in this study to assess the forecasts of track, landfall The performances of track forecasting for location and time, and intensity in 24 h and 48 h and landfalling typhoons in the Northwest Pacific and to analyze the error. the South China Sea in different countries improved in 2005; for example, the 24-h and 48-h forecast 2 THE MODEL GRAPES

Received date: 2009-05-08; revised date: 2009-11-12 Foundation item: Key Science Project for Hunan Meteorological Bureau (200602; 200705) Biography: SONG Yv, senior engineer, mainly undertaking the research on marine meteorology. E-mail for correspondence author: [email protected]

PDF created with pdfFactory trial version www.pdffactory.com 78 Journal of Tropical Meteorology Vol.16 The limited-area model GRAPES[7] adopts a entire process from formation to dissipation (See Fig. semi-implicit, semi-Lagrange time advection and a 1c). The 24-h errors are all within 200 km as far as terrain-following topography-dependent coordinates. the forecasts of eye position are concerned with the The semi-Lagrange trace is calculated with the average error at 111 km; the mean error falls on 246 trajectories of a three-dimensional vector particle. km in the 48-h forecast. Horizontal resolution is 30 km and 31 layers in the 3.4 Verification of Khanum vertical, without Bogus for the TCs. The data of surface, sounding, and boat are used, but with no As many as six forecast simulations were satellite or radar data. The model, covering the conducted for Khanum, covering all but the initial region of East Asia, has a time interval of 300 s and and ending stages of the typhoon (See Fig. 1d). The uses simple ice micro-physics, and adopts the errors of the 24-h forecast of the storm eye are Kain-Fritsch convection scheme, Dudhia shortwave mostly smaller than 200 km, averaging at 132 km. and RRTM longwave radiation schemes, and slab Most of the 48-h forecast errors are larger than 200 schemes. The outputs of T213L31 at 2000 are used as km, averaging at 261 km. the initial fields and boundary layer condition. 3.5 Comprehensive analysis The average 24-h and 48-h errors are 131 km and 252 km, respectively, which are determined Relatively speaking, the track forecasts of the based on 25 sample forecasts for 24 h and 22 for centers of Haitang and Khanum are very close to the 48h. The track forecasted for Haitang was the most observation, or even overlap with it. The 24-h track stable, with errors of 32.7 km for 24 h and 64.7 km forecasts of Matsa and Talim are generally parallel for 48 h, while that for Matsa was the worst, with with the observation. For Matsa, the forecast of the errors of 80.8 km and 164.8 km, respectively. The recurvature after landfall is not good and the average errors of all four typhoons were 58.4 km in forecasts of the cyclogenesis and dissipating stages 24 h and 123 km in 48 h. deviate so much that they are almost inapplicable. Compared with the 24-h track forecast, the 48-h one 3 VERIFICATION AND ANALYSIS OF decreases somewhat in accuracy, though maintaining a consistent tendency with that of the TC TRACK FORECAST observation. 3.1 Verification of Haitang Based on the statistics of the 25 cases of the 24-h forecast and 22 cases of the 48-h forecast, the A total of seven forecast simulations were made mean error is shown to be 131 km for 24 h and 252 of the life cycle of Haitang from cyclogenesis to km for 48 h (See Fig. 1e). Haitang is the best decay except for the initial and ending stages (See forecasted in terms of its eye position with the 24-h Fig. 1a). All of the errors for the 24-h forecast of error at 77 km and the 48-h forecast error at 151 km; the eye location are within a range of 150 km, with the movement forecast has the highest applicability. the most within 100 km and an average error of only As shown in the root mean square error (RMSE) of 77 km; the mean error for the 48-h forecast is 151 the centers of these TCs, the forecast is the most km. It turns out that the model is quite skillful in stable for Haitang, being 32.7 km and 64.7 km for predicting the track of . 24 h and 48 h, respectively; the RMSE is the largest 3.2 Verification of Matsa for the forecast of Matsa, being 80.8 km and 164.8 km for 24 h and 48 h, respectively. The average Like the case of Haitang, there are also seven RMSEs of the four TCs are 58.4 km for 24 h and forecast simulations for the whole life cycle of 123 km for 48 h, respectively (See Fig. 1f). Matsa, excluding the initial and ending stages (See Fig. 1b). For the forecast of the typhoon eye, the 4 ACCESSING THE LANDFALL 24-h errors are between 100 km and 300 km with the average as high as 205 km, mainly due to the The landfall forecast for Haitang in Taiwan, relatively large deviations in the landfall and China was relatively later, north of the actual site, recurvature forecasts of Matsa. The 48-h forecast and weaker. The forecast showed that Talim was errors of the typhoon center are also quite large, all passing north to Taiwan and also weaker (Table 1). surpassing 100 km and with an average of 353 km. The model forecast was good for Haitang, Talim 3.3 Verification of Talim and Khanum landing in Fujian and Zhejiang, both in time and location, while the intensities were weaker Five forecasts of all but the initial and ending for Talim and Khanum (Table 2). The error in stages of were made, covering its landfall site for Matsa was about 100 km.

PDF crea78 ted with pdfFactory trial version www.pdffactory.com No.1 SONG Yv (宋 煜), YE Cheng-zhi (叶成志) et al. 79 After initial field analysis, we found that these fields were also weaker than reality (Table 3).

Fig.1 Forecast errors for TC locations. A: Haitang; b: Matsa; c: Talim; d: Khanum; e: average errors; f: RMSE.

Table 1 Comparison of the forecasts of Haitang and Talim for landfall on Taiwan Island. typhoon Valid time Landing time Landing site Min. pressure Max. winds near Dong’ao, Yilan, Observed landfall 1450 Jul.18 940 hPa 45 m/s Taiwan Is. Forecast at 0800 Jul.18 2300 Jul.18 northern tip of Taiwan Is. 977.5 hPa 34.7 m/s just off northern tip of

Forecast at 0800 Jul. 17 0500 Jul.19 990 hPa 24.5 m/s

g Taiwan Is. n a

t

i 0.3 latitude more a 9 h later than 37 hPa weaker ~10 m/s weaker

H northward than observation for 24 h observation Forecast bias 0.5 latitude more 15 h later than northward than 50 hPa weaker ~20 m/s weaker observation for 48 h observation Observed landfall 0600 Sept.1 near Hualian, Taiwan Is. 930 hPa 50 m/s

just off northern tip of

m Forecast at 0800 Aug.31

i 2300 Aug.31 <982.5 hPa 22.5 m/s

l Taiwan Is. a T ~7 h earlier than 1 latitude northward than Bias in forecast 40 hPa weaker 27.5 m/s weaker observation observation

5 CONCLUSIONS AND DISCUSSIONS observed tracks with a high credibility of reference. (2) The model outputs of typhoon landfall Analysis on the tracks and landfalls of the most agreed well with reality in the southeast China area, severe typhoons in 2005 indicates: especially with respect to landfall time and location. (1) Except for the initial and ending stages in The forecasted landfall location of 24 h was better than which the center location errors in 24 h and 48 h that of 48 h. were 131 km and 252 km, the model was stable, and (3) The location errors at the initial and ending the predicted tracks basically paralleled the stages were obvious and the forecasts showed a

PDF created with pdfFactory trial version www.pdffactory.com 79 80 Journal of Tropical Meteorology Vol.16 weakening tendency. The blockage of Taiwan Island the topographic processing for the island should be is worth mentioning, as the model tended to predict improved. the TCs passing the northern tip of the island. Thus,

Table 2 Comparison of forecasts of Haitang, Matsa, Talim and Khanum for landfall on mainland China.

Valid Max. wind speed typhoon Landfall time Landfall site Centre pressure time near the center 0.5 latitude more 6 h earlier than observation southward than 5 hPa weaker Close to observation in 24 h forecast Forecast observation Haitang bias 0.3 latitude more about the same as southward than 2.5 hPa weaker Close to observation observation in 48 h forecast observation 2 h later than observation in 1 latitude more northward 32 hPa weaker 18 m/s smaller 24 h forecast than observation Forecast Matsa Less than 3 latitudes more bias 2 h earlier in 48 h forecast northward than 34 hPa weaker 15 m/s smaller observation about the same as About the same as 6 hPa weaker 7 m/s smaller Forecast observation in 24 h forecast observation Talim bias 1 latitude more northward 2 h earlier in 48 h forecast 18 hPa weaker 18 m/s v than observation 2-5 h later than observation About the same as 46 hPa weaker 29 m/s smaller Forecast in 24 h forecast observation Khanum bias 0.5 latitude southward 2-3 h later in 48 h forecast 50 hPa weaker 30 m/s smaller than observation

Table 3 Comparison of initial values of the model.

Center pressure Typhoon Valid duration Max. wind speed near center / m/s / hPa Initial model values at 0800 Jul. 18 973.1 35.6

. Haitang n s

o I Observation at 0800 Jul. 18 930 55

g n a n

i Initial model values at 0800 Aug. 31 989.3 32.2 w d i n a Talim a T l Observation at 0800 Aug. 31 910 65

Initial model values at 0800 Jul. 19 981.7 32.2

t Haitang s Observation at 0800 Jul. 19 960 40 a

e a h n

t Initial model values at 0800 Aug. 5 984 31 i u h Matsa o C

s Observation at 0800 Aug. 5 950 45

f n o

o Initial model values at 0800 Sept. 1 984.8 30.4 t s

g Talim a n

i Observation at 0800 Setp. 1 960 40 o d c

n Initial model values at 0800 Sept. 11 999.3 27.4 a l Khanum Observation at 0800 Setp. 11 945 50

(4) The initial fields of the model were too weak showed that there was a weak cyclone southeast of and this had a negative impact on the forecasts. Japan (37°N, 127°E) at 0800 July 7. Led by this Based on the analysis above, we present the cyclone and a steering stream at the edge of following analysis of the problems and possible subtropical high pressure, Matsa turned to the reasons for errors. northeast at Wuhu city, Anhui province, but the (1) Inaccurate forecast of the steering stream: model did not predict the cyclone and so the typhoon Subtropical high pressure, westerly troughs, and the forecast failed. Therefore, the tracking error of inner stress of a typhoon will affect its track, i.e., Matsa came from a mistake involving the steering with a different configuration, the typhoon will stream. move differently. This means that an accurate (2) Initial fields: incorrectly described initial typhoon forecast will depend on correct forecasting intensities or locations of TCs led to forecasts of of these systems. The 500-hPa situation of Matsa weak intensity and the wrong track of the TC.

PDF crea80 ted with pdfFactory trial version www.pdffactory.com No.1 SONG Yv (宋 煜), YE Cheng-zhi (叶成志) et al. 81 The error of initial fields comes primarily from Important Meteorological Information [R]. 2005: NO.106, 134, the shortage of conventional observations, and is 135, 141. then further complicated by the low resolution (30 [2] FUKADA E. A Review of the Joint Typhoon Warning Center 2006 Season [C]. 61st km) that cannot reflect severe weather processes. Interdepartmental Hurricane Conference, March 5-9, 2007, The assimilation of unconventional data and higher New Orleans, LA, U.S.A. resolution are of great importance [8-10], and bogus is [3] Annual Report on Activities of the RSMC Tokyo - Typhoon another effective way to improve the initial fields [11]. Center 2005 [EB/OL]. Higher resolution may improve the tracking http://www.jma.go.jp/jma/jma-eng/jma-center/rsmc-hp-pub-eg/ annualreport.html. accuracy when the TC vortex is only being weakly [4] LV Guo-chen. On the Track Forecasting of Typhoons in described. Northwest Pacific 2005 [C]. Proceedings of Seminar on (3) Topography: the errors of turning at the Meteorological Science and Technology Across the Taiwan island of Taiwan and at the ending stage may result Straits 2006. from the inaccurately processed topography under [5] DUAN Yi-hong, YU Hui, WU Rong-sheng. Review of the the condition of an underestimated TC vortex. research in the intensity change of tropical cyclone [J]. Acta Meteor. Sinica, 2005, 63(5): 636-645. Finely processed topography may improve the [6] YE Cheng-zhi, PAN Zhi-xiang, CHENG Rui, et al. The ability to simulate the vortex. numerical research of the primary mechanism of the landfall (4) Background fields: model T213 forecasted the reinforcement of based on AREM [J]. Acta landfall of Matsa 2 degrees of latitude more Meteor. Sinica, 2007, 65(2): 208-220. [12] [7] CHEN De-hui, YANG-Xue-sheng, ZHANG Hong-liang, et northward , which means that T213 should be improved further. It will cause a substantial error to al. Strategy for designing a non-hydrostatic multi-scale community model dynamic core [J]. Quart. J. Appl. Meteor., take the output of T213 as the background of 2003, l4(4): 452-461. GRAPES, but better background fields will [8] TAN Xiao-wei, CHEN De-hui, ZHANG Qing-hong. An contribute to better results. impact study of a new type of data of adaptive or targeting In all, the error of GRAPES came primarily observation on typhoon forecast [J]. J. Trop. Meteor., 2006, from the shortage of data and this led to inaccurate 22(1): 18-25. [9] YE Cheng-zhi, OUYANG Li-cheng, LI Xiang-yu, et al. steering streams and initial errors of the model. The Validation of 2005 heavy rain events over the Yangtze River key to the solution lies in the assimilation of data basin forecast by GRAPES [J]. J. Trop. Meteor., 2006, 22(4): from satellites and radar to improve the initial fields 393-399. of TC, and to increase the resolution of the model. [10] DING Wei-yu, WAN Qi-lin, YAN Jing-hua, et al. Impact Improved methods in detailed topographic of the initialization on mesoscale model prediction in south processing and better background fields will further China [J]. J. Trop. Meteor., 2006, 22(1): 10-17. [11] WAN Qi-lin, HE Xi-cheng. The structure of tropical decrease the error. cyclone by tovs and its application in numerical weather Acknowledgement: We thank Dr. SUN Jian of the Chinese prediction [J]. J. Trop. Meteor.(English edition), 2002, 8(2): Meteorological Academy for his help. 218-224. [12] GUO Qiao-hong. The application and verification of T213 and ECM model products in the forecast of Typhoon Matsa [J]. REFERENCES: J. Zhejiang Meteor., 2006, 27(2): 7-15.

[1] China Meteorological Administration. Special Reports on

Citation: SONG Yv, YE Cheng-zhi, et al. Verification of typhoon forecasts by the GRAPES model. J. Trop. Meteor., 2010, 16(1): 77-81.

PDF created with pdfFactory trial version www.pdffactory.com 81