Wednesday, 9 January 2019
Hall 4 (Phoenix Convention Center - West and North Buildings)
Handout (221.5 kB)
Meteorological disasters of railways caused by heavy rain include the collapse of embankments, inundation due to rise of water level, and so on. In Japan, to prevent from encountering these disasters, train operation control such as operation suspension are carried out based on values observed by rain gauges installed along railway lines. For localized heavy rain, spatially scale is smaller than arrangement interval of rain gauges, so it is not necessarily possible to detect rainfall accurately.
We are developing a system of disaster prevention against localized heavy rain in small-river in urban area of JAPAN in the framework of Council for Science, Technology and Innovation (CSTI), Cross-ministerial Strategic Innovation Promotion Program (SIP). The system consists of forecasting occurrence of inundation, flooding and landslide disasters by using precipitation information, analyzing stop position of trains avoiding these influence area, and examining passenger evacuation places. Precipitation data input to this system is synthetic precipitation by XRAIN (eXtended RAdar Information Network) and predicted precipitation (Kato, et al. 2017) developed by National research Institute for Earth science and Disaster prevention. Here, XRAIN is an observation network capable of observing precipitation in near real time by X-band MP and C-band radars installed in Japan. It is conceivable that there will have difference between precipitation observed on the ground and synthetic rainfall by radars because it is raindrops in the air that are observed by radars. And there is a possibility that predicted precipitation data has some displacement errors about rainfall area because it uses nowcast and numerical simulation. Because water catchment area of small-scale rivers is narrow, rainfall amount difference between precipitation input to the system and one observed actually may have an influence on prediction result of immersion depth in the system. Therefore, to prevent underestimating results of inundation or flooding analysis in the system, we are studying methods to convert precipitation data.
In this paper, we show results of an investigation for method to estimate ground rainfall using precipitation intensity observed by radars.
In this study, we extracted 20 localized heavy rain instances that the maximum values of 1 hour rainfall was 30 mm or more observed at observation points in Kanto area in 2016 among the ground weather observation networks on a nationwide (AMeDAS: Automated Meteorological Data Acquisition System) operated by Japan Meteorological Agency. If 1 hour rainfall of 30 mm or more was observed at 2 or more AMeDAS point by the same raincloud, we extracted as different instances. In this study, ground analysis stations with 1 hour rainfall of 30 mm or more in each precipitation case were analyzed. Therefore, one ground observation dataset was analyzed for one extracted case. 10 minute precipitation amount observed by AMeDAS point (hereafter ground precipitation) and 10 minute precipitation amount at each grid of XRAIN composite rainfall (hereafter estimated precipitation) in the vicinity of the AMeDAS point were compared. Here, the value obtained by XRAIN is precipitation intensity obtained at every minute. We assumed that precipitation intensity obtained at each time continued with the same intensity in previous 1 minute, we obtained estimated precipitation of 1minute by dividing precipitation intensity by XRAIN by 60. Then, estimated precipitation was obtained by integrating the value for 10 minutes. We compared ground precipitation and estimated precipitation at grid point of XRAIN including ground observation site. As a result, there were difference between these two rainfall amounts. Estimated precipitation is obtained spatially. Therefore, in order to investigate whether there were grids where estimated precipitation with smaller difference from ground precipitation was observed, the simultaneous correlation between estimated precipitation and ground precipitation was calculated in the vicinity of grid including ground observation point. The period of calculating simultaneous correlation coefficient was set to 2 hours from 90 minutes before the time when the maximum value of 1 hour precipitation was observed at the ground observation point. Consequently, there were grids within 3 kilometers horizontally from the ground observation point that correlation coefficient becomes higher than that obtained by grid of XRAIN including the ground observation point.
This displacement seems to occur due to the time between observation raindrops by radars and observation of rainfall on the ground caused by falling them. Therefore, we assumed that error between these two precipitation amounts became smaller by using estimated precipitation obtained a few minutes before ground precipitation was obtained. We set time difference between the time to calculate ground precipitation and that to calculate estimated precipitation in 1 up to 9 minutes. And then, for each the extracted cases, we investigated time difference at which root mean square error (RMSE) between ground precipitation and estimated precipitation was the smallest. Here, the period to calculate RMSE of two kinds of precipitation was set in the same 2 hours as the period for calculating simultaneous correlation coefficient mentioned above. As a result, although there were different cases, it was found that RMSE was the smallest in the case that time difference was 2 minutes (about 2mm), and the larger time difference became, the larger RMSE tended to become (about 4mm in difference 9min). In the case that the terminal speed of raindrops is 10 m/s, raindrops will drop 1200 m in 2 minutes. Since this value is within the range of observation altitude of the radar, it is considered effective to consider the time difference in order to estimate ground precipitation. Therefore, it is thought that ground precipitation are estimated with less error by using synthetic precipitation observed 2 minutes before.
We will also present the result of examination of the estimation method considering the ground wind or wind around 1km from ground level at the presentation day.
We are developing a system of disaster prevention against localized heavy rain in small-river in urban area of JAPAN in the framework of Council for Science, Technology and Innovation (CSTI), Cross-ministerial Strategic Innovation Promotion Program (SIP). The system consists of forecasting occurrence of inundation, flooding and landslide disasters by using precipitation information, analyzing stop position of trains avoiding these influence area, and examining passenger evacuation places. Precipitation data input to this system is synthetic precipitation by XRAIN (eXtended RAdar Information Network) and predicted precipitation (Kato, et al. 2017) developed by National research Institute for Earth science and Disaster prevention. Here, XRAIN is an observation network capable of observing precipitation in near real time by X-band MP and C-band radars installed in Japan. It is conceivable that there will have difference between precipitation observed on the ground and synthetic rainfall by radars because it is raindrops in the air that are observed by radars. And there is a possibility that predicted precipitation data has some displacement errors about rainfall area because it uses nowcast and numerical simulation. Because water catchment area of small-scale rivers is narrow, rainfall amount difference between precipitation input to the system and one observed actually may have an influence on prediction result of immersion depth in the system. Therefore, to prevent underestimating results of inundation or flooding analysis in the system, we are studying methods to convert precipitation data.
In this paper, we show results of an investigation for method to estimate ground rainfall using precipitation intensity observed by radars.
In this study, we extracted 20 localized heavy rain instances that the maximum values of 1 hour rainfall was 30 mm or more observed at observation points in Kanto area in 2016 among the ground weather observation networks on a nationwide (AMeDAS: Automated Meteorological Data Acquisition System) operated by Japan Meteorological Agency. If 1 hour rainfall of 30 mm or more was observed at 2 or more AMeDAS point by the same raincloud, we extracted as different instances. In this study, ground analysis stations with 1 hour rainfall of 30 mm or more in each precipitation case were analyzed. Therefore, one ground observation dataset was analyzed for one extracted case. 10 minute precipitation amount observed by AMeDAS point (hereafter ground precipitation) and 10 minute precipitation amount at each grid of XRAIN composite rainfall (hereafter estimated precipitation) in the vicinity of the AMeDAS point were compared. Here, the value obtained by XRAIN is precipitation intensity obtained at every minute. We assumed that precipitation intensity obtained at each time continued with the same intensity in previous 1 minute, we obtained estimated precipitation of 1minute by dividing precipitation intensity by XRAIN by 60. Then, estimated precipitation was obtained by integrating the value for 10 minutes. We compared ground precipitation and estimated precipitation at grid point of XRAIN including ground observation site. As a result, there were difference between these two rainfall amounts. Estimated precipitation is obtained spatially. Therefore, in order to investigate whether there were grids where estimated precipitation with smaller difference from ground precipitation was observed, the simultaneous correlation between estimated precipitation and ground precipitation was calculated in the vicinity of grid including ground observation point. The period of calculating simultaneous correlation coefficient was set to 2 hours from 90 minutes before the time when the maximum value of 1 hour precipitation was observed at the ground observation point. Consequently, there were grids within 3 kilometers horizontally from the ground observation point that correlation coefficient becomes higher than that obtained by grid of XRAIN including the ground observation point.
This displacement seems to occur due to the time between observation raindrops by radars and observation of rainfall on the ground caused by falling them. Therefore, we assumed that error between these two precipitation amounts became smaller by using estimated precipitation obtained a few minutes before ground precipitation was obtained. We set time difference between the time to calculate ground precipitation and that to calculate estimated precipitation in 1 up to 9 minutes. And then, for each the extracted cases, we investigated time difference at which root mean square error (RMSE) between ground precipitation and estimated precipitation was the smallest. Here, the period to calculate RMSE of two kinds of precipitation was set in the same 2 hours as the period for calculating simultaneous correlation coefficient mentioned above. As a result, although there were different cases, it was found that RMSE was the smallest in the case that time difference was 2 minutes (about 2mm), and the larger time difference became, the larger RMSE tended to become (about 4mm in difference 9min). In the case that the terminal speed of raindrops is 10 m/s, raindrops will drop 1200 m in 2 minutes. Since this value is within the range of observation altitude of the radar, it is considered effective to consider the time difference in order to estimate ground precipitation. Therefore, it is thought that ground precipitation are estimated with less error by using synthetic precipitation observed 2 minutes before.
We will also present the result of examination of the estimation method considering the ground wind or wind around 1km from ground level at the presentation day.
This work was supported by Council for Science, Technology and Innovation (CSTI), Cross-ministerial Strategic Innovation Promotion Program (SIP), “Enhancement of societal resiliency against natural disasters.” (Funding agency: Japan Science and Technology Agency (JST)).
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner