High-performance computation in hazard and risk research - - PowerPoint PPT Presentation

▶
high performance computation in hazard and risk research
SMART_READER_LITE
LIVE PREVIEW

High-performance computation in hazard and risk research - - PowerPoint PPT Presentation

Mar 05, 2023 250 likes •587 views

High-performance computation in hazard and risk research Friedemann Wenzel, Bruno Merz, Patrick Heneka, Thomas Hofherr, Joachim Miksat Karlsruhe Institute of Technology GeoForschungsZentrum Potsdam International Symposium on Grid Computing

slide-1
SLIDE 1

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

High-performance computation in hazard and risk research

Friedemann Wenzel, Bruno Merz, Patrick Heneka, Thomas Hofherr, Joachim Miksat Karlsruhe Institute of Technology GeoForschungsZentrum Potsdam

slide-2
SLIDE 2

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Center for Disaster Management and Risk Reduction Technology (CEDIM) Mission Risk Mapping Computational Challenges Winterstorm modelling Earthquake ground motion modelling

slide-3
SLIDE 3

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

CEDIM History

The Center of Disaster Management and Risk Reduction Technology (CEDIM, www.cedim.de) is a joint Center of Excellence of University of Karlsruhe (one of eight German Elite-Universities) and two large research institutions of Helmholtz Gesellschaft, the GeoForschungsZentrum (GFZ) Potsdam and the Forschungszentrum Karlsruhe (FZK). It has been established in 2002 by University of Karlsruhe and GFZ; FZK joined in 2007. Currently 30 scientists of the three institutions work under CEDIMs umbrella.

slide-4
SLIDE 4

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

CEDIM Mission

  • CEDIM creates scientific knowledge, technologies and

intelligent tools in these fields by developing synergies between the expertise of its supporting institutions.

  • CEDIM co-operates closely with national risk and crisis

managing agencies but also contributes to key international challenges such as the impact of disasters on megacities and under climate change conditions.

  • As Center of Excellence of a university CEDIM communicates

its experience into the academic sector with the aim of mainstreaming disaster risk reduction in education. Successful risk reduction requires risk assessment and analysis, risk communication and risk management.

slide-5
SLIDE 5

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Partners and key expertise

Engineering: Structural, Electrical, Mechanical, Communication Water Ressource Management Economic engineering Logistics Engineering Geological Hazards Meteorology and Climate Research Decision Support Sustainability Analysis Emergency Medicin Geological Hazards Early Warning Systems Satellite Technology Flood Risk

slide-6
SLIDE 6

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Institutes

  • Institute for Meteorology und Climate Research (IMK)
  • Institute for Hydromechanics (IFH)
  • Institute for Water and River Basin Management (IWG)
  • Geophysical Institute (GPI)
  • Institute of Concrete Structures and Construction Material Technology (IFMB)
  • Lehrstuhl für Versicherungswirtschaft (LVW)
  • Institut für Wirtschaftswissenschaften - Verkehrsnetzwerke (IWW)
  • Institute for Industrial Production (IIP)
  • Institute for Communications Engineering (INT)
  • Geological Institute (AGK)
  • Institute for Technology and Management in Construction (TMB)
  • Geodetic Institute (GIK)
  • Institute for Meteorology und Climate Research (IMK-TRO)
  • Atmospheric Trace Constituents and Remote Sensing (IMK-ASF)
  • Atmospheric Environmental Research Division (IMK-IFU)
  • Institute for Nuclear and Energy Technologies - Accident Consequence Group (IKET-UNF)
  • Institute for Technology Assessment and System Analysis (ITAS)
  • Medical Department (MED)
  • Remote Sensing (Sektion 1.4)
  • Earthquake Risk and Early Warning (Sektion 2.1)
  • Earth's Magnetic Field (Sektion 2.3)
  • Seismology (Sektion 2.4)
  • Deformation und Rheology (Sektion 3.2)
  • Engineering Seismology (Sektion 5.3)
  • Engineering Hydrology (Sektion 5.4)
slide-7
SLIDE 7

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

scientific groups: – asset estimation – earthquake – flood – winter storm – man-made-hazards – synopsis – GIS / data management

earthquake storm GIS / data management synopsis flood man-made hazards asset estimation

CEDIM – Risk Map Germany

slide-8
SLIDE 8

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Source: Merz, B. / Thieken, A. (2004): Flood Risk Analysis: Concepts and Challenges, Österreichische Wasser- und Abfallwirtschaft 56/3-4

Hazard, Exposure, Vulnerability and Risk

slide-9
SLIDE 9

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Regionalization of Values

Regionalization of population and assets based on CORINE landuse data

slide-10
SLIDE 10

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

CEDIM-Riskmaps

Earthquakes T = 475 a Floods T = ca. 500 a Storm T = 2 a … T = 500 a

T = 50 a T = 200 a Damage to residential buildings Reference: 2000

(IKSR-Scenario)

slide-11
SLIDE 11

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Risk Comparison

Damage frequency curve Average annual loss (AAL)

Annual probability of exceedance

Probable maximum loss (PML) (at a given time period) Risk can be compared through these three functions/curves Risk can also be compared in a spatial manner through maps, e.g., mapping the maximum losses at 100 year event for each hazard in each community Risk can be compared through these three functions/curves Risk can also be compared in a spatial manner through maps, e.g., mapping the maximum losses at 100 year event for each hazard in each community

100-year flood loss (euro) 0.001 0.01 1000-year EQ loss (euro) 0.001 0.01 0.001 0.01 loss loss

slide-12
SLIDE 12

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Comparison of risks for Saxony

  • Hazard with

highest damage potential for residential buildings

  • Comparison of

damage per capita

  • Returnperiods

EQ475, ST200, HW200/300

slide-13
SLIDE 13

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

slide-14
SLIDE 14

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

  • Storm Hazard:

exceedance probability of maximum wind speed (grid size: 1km x 1km)

  • Vulnerability:

storm damage function in respect of building structure, exposition, wind speed etc.

  • Storm Damage Risk:

statistical expected loss per zip code for specific levels of exceedance probability Definition of Risk: Risk = Storm Hazard x Vulnerability x Value The assests are taken from the work of the CEDIM Asset Estimation Group.

Orography Storm hazard Vulnerability of technical and natural structures Risk calculation Wind climate

Method Risk Map

Storm Risk

slide-15
SLIDE 15

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Method: 1) selection of the strongest storm event per year 2) numerical simulation of the wind field pattern 3) development of a storm damage function 4) adaption to an extreme value distribution function 5) risk map

Storm damage risk model

Storm Risk

slide-16
SLIDE 16

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

ECMWF Storm Maps

slide-17
SLIDE 17

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Storm Maps – Hazard Curves

slide-18
SLIDE 18

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

KAMM: Karlsruhe Atmospheric Mesoscale Model

Developed at the Institute of Meteorology and Climate Research

'synoptic scale'

  • bservation

prediction equation of motion energy equation 'KAMM' soil- vegetation model terrain height landuse data biogeníc emissions anthropogenic emissions balance equations

  • f the reactive species

'DRAIS' deposition model chemical model 'RADM2 (modified)' aerosol model 'MADEsoot' photolysis rates 'STAR' wind, temperature, humidity, turbulence, radiation, concentrations, deposition, aerosols

Model architecture

Storm Hazard

slide-19
SLIDE 19

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

equations of motion:

ideal gas law:

conservation of mass: first law of thermo- dynamics:

Physics

input data:

  • rography, land use data, large-

scale synoptic weather conditions, initial values for the wind and temperature field

  • utput data:

3-dim. fields of wind speed, temperature, humidity, pressure, shear stress etc. Storm Hazard

slide-20
SLIDE 20

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Implementation

The Forschungszentrum Karlsruhe is operating HPC vector systems since 1986 starting with IBM 3090VF-600 and Fujitsu

  • VP50. Last year the two running vector systems VPP5000-8

and SX-5/8 are substituted by a SX-8R vector system (8 vektorprocessors archiving 36 GFlop/s peak and offer 256 GB shared memory).

Storm Hazard

slide-21
SLIDE 21

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Example: Gale Lore, January, 28 1994

wind field simulation by KAMM (1km x 1km)

characteristics:

  • the North is more

affected than the South

  • modification of the wind

field pattern due to the complexe terrain

  • differences in surface

roughness are visible

Karlsruhe Stuttgart Freudenstadt Lahr

Storm Hazard

3-dim. grid with terrain following coordinates: 320 x 320 x 50 grid points x 50.000 time steps → Calculating the complex differential equations: ~5*109 times

slide-22
SLIDE 22

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Extratropical storms

Method – Hazard: 1) devision into 6 sub-regions* 2) detection of the strongest storm event per year 3) simulation of the wind field pattern by KAMM 4) calculation of the extreme value distribution function at each grid point 5) validation using observational data 6) hazard map └> Approx. computing time: 180 days

* due to the limited calculating capacity

Storm Hazard

slide-23
SLIDE 23

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Storm Maps – Vulnerability Curves

slide-24
SLIDE 24

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Winter Storm Lothar 26.12.99

Total Error: 4% Spatial Correlation: 0.87 Damage per community

  • f winterstorm Lothar
slide-25
SLIDE 25

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Taipei Basin Response Modelling

planar wave front incidence: N S-wave front

Co-operation with Prof. Kou-Liang Wen (National Central University, Taiwan)

slide-26
SLIDE 26

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Model Development - Input

Tertiary Basement (Wang et al., 2004)

750 m 200-300 m

Topography (SRTM) SungShan Formation (Wang et al., 2004)

120 m 40-60 m

  • P- and S-wave

velocities (Wang et al., 2004)

  • SRTM topography

data

slide-27
SLIDE 27

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Model Development - Output

N

  • 3D model size:

27.8 x 27.8 x 2 km3

  • discretisation:

dx = dy = 25 m dz = 12.5 m → 225 ·106 grid points

  • parameters:

vp, vs, density

slide-28
SLIDE 28

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

HP XC6000 at the Scientific Supercomputing Centre Karlsruhe

  • Intel Itanium2 processors (1.5 GHz/1.6 GHz)
  • 101 compute nodes each with 2 proc. and 12 GB main mem.
  • 10 compute nodes each with 8 proc. and 64 GB main mem.

⇒ ~ 2 TB main mem.; theor. peak performance: 1.9 TFLOPS

slide-29
SLIDE 29

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Code & Model

  • 3D Finite-Difference code

(Furumura, 2005): – Fortran – parallelised with MPI

  • 3D model:

– 27.8 x 27.8 x 2 km3 ⇒1132 x 1132 x 176 = 225 ·106 grid points

  • Wave propagation simulation for 30 s

⇒ 20,000 time steps with dt = 1.5 ms

  • Needed resources of the HP XC6000:

– ~ 120 GB of main memory – 16 processors – computation time: 22 h 15 min

slide-30
SLIDE 30

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

Simulated Wavefield

slide-31
SLIDE 31

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

CEDIM Research Topic I: Tools for risk management

  • Develop Tools for Risk Management (Scenario

Generator, Synoptic Risk Evaluation)

  • Develop Real-time Information Tools
  • Include Risk to Infrastructure
  • Include Risk to Industrial Facilities
  • Include Socio-economic Vulnerability
slide-32
SLIDE 32

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan

CEDIM Research Topic II: Risk and Regional Climate Change

Objective: Quantify change of flood risk in small to medium size catchments Method: Combine regional climate models driven by global models with regional/local hydrological models and

  • bservations

Application: Three typical river systems 30 years control time series (1971-2000) 30 years prognosis for 2030-2060)

slide-33
SLIDE 33

International Symposium on Grid Computing 2008 Academia Sinica, Taipei, Taiwan Within CEDIM the high performance computing Center is used for extensive climate simulations.

Implementation

The Forschungszentrum Karlsruhe is operating HPC vector systems since 1986 starting with IBM 3090VF-600 and Fujitsu

  • VP50. Last year the two running vector systems VPP5000-8

and SX-5/8 are substituted by a SX-8R vector system (8 vektorprocessors archiving 36 GFlop/s peak and offer 256 GB shared memory).

Storm Hazard