Development and Validation of Real-time Tsunami Hazard Mapping System
Abstract
A real-time tsunami hazard mapping system has been developed in order to serve highly accurate prediction of tsunami inundation, to support people's evacuation, and to understand damaged areas promptly in the disaster aftermath. It utilizes off-shore tsunami observation data. The system consists of (1) input of observed data, (2) inversion of tsunami source, (3) tsunami propagation and inundation simulation, and (4) output of predicted information. An early tsunami inundation forecasting system is originally proposed by Tatsumi and Tomita (2013). In this study, we develop an automatic system including the four elements above. The algorithms of tsunami source inversion were refined in order to improve the prediction accuracy. A GPU was introduced to reduce calculation time for source inversion and tsunami propagation and inundation. System validation was conducted by simulation-based experiments. In the experiments, we assume a scenario tsunami generated by predicted M9 Nankai trough earthquake. Tsunami source is estimated by using the spatio-temporal inversion method (Takagawa & Tomita, 2012). In the tsunami source inversion with the spatial resolution of 15 km and temporal resolution of 1 min, the number of unknown parameters to be estimated is more than 2,000. Although it had taken more than 10 minutes to estimate thousands parameters in a CPU calculation, our newly developed GPU (NVIDIA Tesla C2075) implementation can estimate those within 5 seconds. Such high-performance calculation makes it possible to apply a highly accurate method to real-time usage. In order to raise the prediction accuracy, a priori information about spatio-temporal smoothness is needed in the method. The system can optimize the hyper parameters of the smoothness on the basis of observation data by a cross-correlation method. Tsunami inundation is simulated based on the nonlinear long wave theory. The momentum and continuity equations are discretized by a staggered grid in space and integrated temporally by the leap-frog scheme. To balance the calculation cost, resolution and prediction accuracy of inundation, nested grids were used. Grid intervals were 810, 270, 90 and 30 m in the validation. Two-way interaction between the grids with different intervals is considered. The simulation program was also accelerated by GPU. It takes 75 seconds to simulate 2 hour tsunami propagation and inundation. In our implementation, the calculation speed of GPU is about 700 times greater than single thread CPU. In the simulation-based validation, the focus area was set to Nagoya port. A tsunami was observed by 8 GPS buoys along the Nankai trough. A comparison between the scenario tsunami waveform and predicted waveforms reveals the prediction accuracy. Even in the limited observation within six minutes after the earthquake occurrence, the prediction error of first wave peak was 4 % in height and 2 % in arrival time. Our system can provide the predicted tsunami waveforms and spatial distribution of inundation more than one hour before the tsunami attack in the focus area.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2013
- Bibcode:
- 2013AGUFMNH41B1716T
- Keywords:
-
- 4315 NATURAL HAZARDS Monitoring;
- forecasting;
- prediction;
- 4341 NATURAL HAZARDS Early warning systems