Work

Event-Based Flood Risk Assessment on Massive Terrains

Year: 2019
Type: other
Abstract: Flooding is a major societal problem today as extreme weather phenomena are becoming more common.When it rains, rainwater flows into drains and collects in sewers, is absorbed by soil, or flows into existing crooks and rivers and ends up in a lake or ocean.However, when the amount of rain exceeds what can be absorbed by soil and sewers, water collects into large pools that fill up depressions of the terrain.Thanks to advances in LiDAR sensing technology, highly detailed digital elevation models are increasingly becoming available.By refining the digital elevation model of an area to incorporate underground pipes, tunnels, and waterways under bridges, the resulting surface model enables a wide variety of terrain analysis applications, which includes simulating water flow to predict which areas will be flooded in various kinds of extreme weather.However, the size of the terrain models available today easily exceed the main memory (RAM) capacity of a regular desktop computer.Since reading a piece of data from the disk is a million times slower than reading data from RAM, algorithms must be I/O-efficient, that is, they must be designed to minimize the number of times data is accessed on disk.In this thesis, we develop new I/O-efficient algorithms to model and simulate flooding from rainfall and sea-level rise on massive digital elevation models.Contrary to previous work, our algorithms have the capacity to model non-uniform extreme weather events, such as rainfall that varies in intensity over the terrain or varying sea level along the coastline.Since e.g.local extreme flash floods can affect downstream areas that do not receive heavy rainfall directly, it is important to model such nonuniform events.To aid in the implementation of I/O-efficient algorithms such as the ones we present in this thesis, we also develop a new software framework to support modular and maintainable implementations of algorithms while automatically minimizing the I/O-overhead between modules.Finally, we consider modeling flood risk on terrains with uncertainty, since large high-resolution terrain models only approximate the ground truth and are subject to a degree of uncertainty; for instance, the raw data acquired with LiDAR sensing is subject to measurement error, and the refinement step to incorporate underground pipes and other waterways into the terrain model is not perfect, which causes uncertainty in the resulting terrain model.By representing the uncertainty in terrain data explicitly, we develop methods for flood risk analysis that properly incorporate terrain uncertainty when reporting what areas are at risk of flooding.v Resumé Oversvømmelse er et omfattende samfundsmæssigt problem i dag, da ekstreme vejrfænomener bliver mere almindelige.Når det regner, ender regnvandet med at flyde ned i afløb og samles i kloakker, at blive absorberet af jordbunden, eller at flyde hen i eksisterende bække og floder for til sidst at ende i en sø eller et hav.Når mængden af regnvand derimod overstiger hvad der kan absorberes af jordbunden og kloakkerne, samles vandet i store vandmasser, der oversvømmer lavninger i terrænet.Takket være udviklingen i LiDAR-sensing-teknologi bliver meget detaljerede digitale højdemodeller i stigende grad tilgængelige.Ved at tilrette højdemodellen for at inkorporere underjordiske rør, tunneller, og vandføringsveje under broer kan den benyttes til mange forskellige analyser, herunder at forudse hvilke områder der oversvømmes ved forskellige former for ekstreme vejrfænomener.Størrelsen på de tilgængelige højdemodeller overstiger dog nemt den interne hukommelse (RAM) i en almindelig computer.Da det tager en million gange længere at læse et stykke data fra en harddisk, end det tager at læse data fra RAM, skal algoritmer være I/O-effektive, dvs.de skal designes til at minimere antallet af gange data læses fra harddisken.I denne afhandling udvikler vi nye I/O-effektive algoritmer til at modellere og simulere oversvømmelse fra regn og havniveaustigninger på massive digitale højdemodeller.I modsætning til tidligere forskningsresultater kan vores algoritmer bruges til at modellere uensartede vejrfænomener, såsom når nedbørsmængden varierer hen over landet, eller når havniveauet varierer langs kysten.Eftersom eksempelvis lokale skybrud kan oversvømme områder nedstrøms, der ikke selv har fået store mængder nedbør, er det vigtigt at modellere sådanne uensartede hændelser.For at støtte op om implementeringen af I/O-effektive algoritmer, såsom dem vi præsenterer i denne afhandling, udvikler vi også et nyt software-framework der understøtter modulære og vedligeholdbare implementationer af algoritmer, samtidig med at de ekstra I/O-operationer mellem moduler automatisk mindskes.Endelig undersøger vi oversvømmelsesrisiko på højdemodeller med usikkerheder, da massive detaljerede højdemodeller kun er en tilnærmelse af virkeligheden; eksempelvis er rådata fra LiDAR-sensing forbundet med måleusikkerhed, og processen forbundet med at tilrette højdemodellen for at inkorporere underjordiske rør og andre vandføringsveje er ikke perfekt, hvilket forårsager usikkerhed i den resulterende højdemodel.Ved at repræsentere usikkerheden i højdemodellen direkte udvikler vi metoder til oversvømmelsesrisikovurdering der tager højde for datausikkerhed når det rapporteres hvilke områder der er i risiko for oversvømmelse.vii I would like to thank my advisor Lars Arge who took me in as a student programmer seven years ago and showed me the road to becoming an independent researcher.Aarhus University officially shuns the term "advisor" and instead calls Lars my "supervisor", which is a term commonly defined as "someone who is charge of somebody or something and makes sure that everything is done correctly"; I find that term pejorative, since the goal of the PhD program is to qualify the PhD student to carry out research independently.Lars has made sure that I was always on track, and he has always been focused and present when I have met with him to ask for his advice.Without his acute intuition, deep knowledge and experience within terrain analysis and I/O-algorithms, I would have followed only unfruitful ideas and missed all opportunities to go beyond the state of the art.Lars's red pen has taught me how to improve the impact and readability of my own writing, and by studying his public talks and presentations I have learned how to present algorithms in a way that everyone can understand. Thanks to my belovedAmalie who has supported me throughout my PhD studies and helped me relax when I was stressed and cheer up when all seemed lost.You are always by my side, and I promise to always be by yours.Thanks to Pankaj K. Agarwal who was my host at Duke University in Spring 2017 and to Pankaj's student Aaron Lowe who insisted on going the extra mile to obtain truly great research results during my half-year stay at Duke.Thanks to all the people in Durham, North Carolina who made my stay abroad feel like a stay at home-away-from-home. Thanks to Ian Zerny and Mads Ager
Author: Rav, Mathias
Institution: Aarhus University
Language: English
Open Access status: green
OpenAlex ID: W4412306097
Cites: 57
Cited by:
Related to: 10