Thesis HACIA EL MODELADO DE TSUNAMIS EN TIEMPO REAL: ¿RAPIDEZ O PRECISIÓN?
Date
2019
Authors
MIGLIARO GAJARDO, FRANCO JAVIER
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Los tsunamis en Chile a veces pueden proporcionar un intervalo de tiempo lo suficientementeacotado entre su generación y su impacto en la costa como para descartar simulaciones entiempo real para modelar y predecir su inundación. Es por lo anterior que se hace necesariobuscar métodos que permitan disminuir los tiempos de simulación y que a su vez no arrojenresultados con una pérdida de precisión que se traduzca en predicciones muy alejadas de larealidad.En el presente documento se analizan en paralelo dos formas que buscan disminuir el tiemporequerido para la simulación computacional o pronóstico de la inundación de tsunamisproducidos por terremotos de falla convergente, sobre los cuales se realiza un análisis decalidad de predicción o de precisión. Estos son, en primer lugar la utilización de Tsunami-HySEA, software que modela inundación por tsunami desarrollado por el grupo EDANYAde la Universidad de Málaga, cuya característica principal es estar optimizado para trabajaren GPU y en segundo lugar, el método NearTIF, desarrollado por Gusman et al., 2014, queen forma indirecta y basado en una serie de escenarios permite encontrar similitudes en lasseries de tiempo y pronosticar inundación.Cabe destacar que ambos análisis se sometieron al tsunami generado por el terremoto delMaule 2010 para las ciudades de San Antonio, Constitución y Talcahuano.Se utilizó Tsunami-HySEA para generar cinco configuraciones topo-batimétricas sobre lascuales se analizaron diversos parámetros con fin de pesquisar el efecto en la calidad delresultado al variar la resolución de la morfología de las localidades en cuestión. Se concluyóque es posible utilizar ciertas configuraciones específicas, dependiendo de la calidad deinformación a emplear, sin perder presición excesiva y logrando disminuir los tiempos decálculo a un cuarto del tiempo original (modelo completo).Para poder verificar el comportamiento de la metodología NearTIF, se generaron suscomponentes (base de datos y algoritmo de búsqueda) y se ejecutó, teniendo resultadosdispares entre las tres ciudades bajo análisis. En tanto San Antonio y Talcahuano tuvieron unpronóstico deseable, Constitución tuvo una gran sobreestimación en el pronóstico deinundación.En términos generales, se observó que ambos métodos logran disminuir los tiempos desimulación lo suficiente como para incluirlos en un sistema de alerta temprana deaproximadamente una hora a unos cuantos minutos (entre uno y seis, dependiendo delmétodo), considerando algunas mejoras que afectaron la precisión de las predicciones enalgunos casos particulares.
The limited time interval between the generation of tsunamis in Chile and their impact on thecoast rules out simulations in real time to model and predict their flooding. For this reason,it is necessary to look for methods that reduce the simulation times and that in turn do notgive results with a loss of precision that translates to predictions very far from the reality.In this document, two methods are analyzed in parallel that seek to reduce the time requiredfor computational simulation or flood forecasting of convergent fault tsunamis, on which ananalysis of prediction or precision quality is performed. These are, in the first place the useof Tsunami-HySEA, software developed by the EDANYA group of the University of Málagaby Macías et al., (2017), whose main characteristic is to be optimized to work on GPU andsecondly, the NearTIF method, developed by Gusman et al., (2014).The analyses were performed using the tsunami generated by the 2010 Maule earthquake, forthe cities of San Antonio, Constitución and Talcahuano.Tsunami-HySEA was used to generate five topo-bathymetric configurations on whichvarious parameters were analyzed in order to investigate the effect on the result’s quality.The previous varying the resolution of the morphology of the localities of interest. It wasconcluded that it is possible to use some specific configurations, depending on the quality ofthe information to be used, without losing excessive precision, reducing the calculation timesto approximately one quarter of the complete model.In order to verify the behavior of the NearTIF methodology, its parts (database, andalgorithm) were generated and executed, with varying results among the three cities underanalysis. While San Antonio and Talcahuano had a desirable forecast, Constitution had alarge overestimation in the forecast.In general terms, it was observed that both methods manage to reduce simulation timesenough to add it to a warning system, considering some improvements that affected theaccuracy of the predictions in some particular cases.
The limited time interval between the generation of tsunamis in Chile and their impact on thecoast rules out simulations in real time to model and predict their flooding. For this reason,it is necessary to look for methods that reduce the simulation times and that in turn do notgive results with a loss of precision that translates to predictions very far from the reality.In this document, two methods are analyzed in parallel that seek to reduce the time requiredfor computational simulation or flood forecasting of convergent fault tsunamis, on which ananalysis of prediction or precision quality is performed. These are, in the first place the useof Tsunami-HySEA, software developed by the EDANYA group of the University of Málagaby Macías et al., (2017), whose main characteristic is to be optimized to work on GPU andsecondly, the NearTIF method, developed by Gusman et al., (2014).The analyses were performed using the tsunami generated by the 2010 Maule earthquake, forthe cities of San Antonio, Constitución and Talcahuano.Tsunami-HySEA was used to generate five topo-bathymetric configurations on whichvarious parameters were analyzed in order to investigate the effect on the result’s quality.The previous varying the resolution of the morphology of the localities of interest. It wasconcluded that it is possible to use some specific configurations, depending on the quality ofthe information to be used, without losing excessive precision, reducing the calculation timesto approximately one quarter of the complete model.In order to verify the behavior of the NearTIF methodology, its parts (database, andalgorithm) were generated and executed, with varying results among the three cities underanalysis. While San Antonio and Talcahuano had a desirable forecast, Constitution had alarge overestimation in the forecast.In general terms, it was observed that both methods manage to reduce simulation timesenough to add it to a warning system, considering some improvements that affected theaccuracy of the predictions in some particular cases.
Description
Catalogado desde la version PDF de la tesis.
Keywords
METODOLOGIA NEARTIF , MODELADO DE TSUNAMIS , PREDICCION DE INUNDACIONES , SISTEMA DE ALERTA TEMPRANA