High Resolution Integrated Weather- Flood Modelling Framework

Transcription

1 High Resolution Integrated Weather- Flood Modelling Framework IBM Research Team : Ulisses Mello, Lucas Villa Real, Vaibhav Saxena, Thomas George, Rashmi Mittal, Yogish Sabharwal IBM Research : Merlin Davis (IIT Madras)

2 SWAT Contents Flooding Problem Modelling Framework 2D flood inundation model: Diffusive wave Approximation Numerical Discretization 2D dam break test case and results Work in progress Code parallelization Initial case study Parameterizations 2

3 SWAT Flooding is a common challenge over the world Flooding is the most common of all environmental hazards and it regularly claims over 20,000 lives per year and adversely affects around 75 million people world-wide 2008 Iowa Flooding Event Source: NOAA $7B in Damage! 3 3

4 SWAT Motivation: 5-6 April 200 Flooding Event in Rio Coastal storm with heavy rains (up to 284mm in 24 hours) starting at about 700 BRT on 5 April 200 heaviest recorded compared to the previous 48 years One of the most significant global weather events of 200 Local flooding leading to mudslides, killed over 00 people in Rio and left thousands homeless Widespread disruption of transportation systems (e.g., road closures, airport and rail delays) Rio de Janeiro mayor admitted that the flooding of April 200 uncovered significant gaps in the city s preparedness to respond to major emergencies/crises.. 4

5 SWAT Motivation: Brunei Natural Disaster Events 962: Major flood 980s: Fires in water village 987: Rasau gas blow-out in Belait District 99: Poor air quality resulting from Mount Pinatubo eruption in the Philippines 998: Regional haze 999: Flash flood during La Niña 2008: Temburong flash flood 2009: Extensive flash flood in Muara, Tutong and Belait districts 20 January 2009 Flash Flood in Brunei Heavy rainfall for 4 to 5 hours with a record 45.8mm in 24 hours. Widespread disruption of electric distribution, transportation and communications systems 5

6 SWAT Urban Flood Forecasting Framework Precipitation Estimates Weather Prediction and/or Rainfall Measurements Analysis of Precipitation Flood Prediction Actual Flood Impacts Refine Sensor Network and Model Calibration Model Calibration Impact Estimates 6

7 SWAT Main Elements of Modelling framework rainfall evaporation Quantitative Approach to Estimate Runoff friction runoff topography infiltration 7

8 SWAT Flood Analytics Modelling of the water transport on the land surface and within river channel can be done by physical models Efficient and large scale modelling of the water transport is the basic building block of many different flood related applications Fully dynamic D model for efficient within-river modeling 2D run-off models connecting precipitation to river channel 8 Detailed 2D routing model for accurate flood plain estimate at selected high-risk urban locations 8

9 9 Hydraulic 2D : Surface Runoff Model Assumptions Surface runoff problems have usually the following characteristics: The average depth of the layer of fluid is very small in relation to its other dimensions. The vertical velocity is usually smaller than the horizontal ones. Free surface The hydrostatic balance is a good approximation in the vertical direction : p = ρ g z

10 0 Complete Shallow Water Model (Including Friction) ( uh) ( vh) h + + = R I E ± C t x y ( ) ( ) ( ) uh uhu vhu w ng x u + v h ς + + = τ x u gh /3 + t x y h x x ( ) ( ) ( ) rainfall infiltration evaporation canals vh uhv vhv ng w y u + v h ς + + = τ y v gh /3 + t x y h y y + Initial condition: topography, water depth and velocities known + Boundary conditions: Prescription of values or derivatives of unknowns at the boundary

11 Diffusive Wave Approximation of Complete Shallow Water Model ( uh) ( vh) h + + = R I E ± C t x y elimination of the inertial terms 0 0 ( ) ( ) ( ) uh uhu vhu w ng x u + v h ς + + = τ x u gh /3 + t x y h x x ( ) ( ) ( ) vh uhv vhv ng w y u + v h ς + + = τ y v gh /3 + t x y h y y Diffusive wave Approximation inertia term is taken negligible

12 2 Diffusion Model for the Shallow Water ( uh) ( vh) h + + = R I E ± C t x y () u 2/3 = /2 n H / s x h ( h+ς ) x v h ( h+ ς ) n H / s y 2/3 = /2 y (2) where the wind stresses were neglected for the sake of simplicity. Replacing Eqs. (2) in (), we obtain the following equation:

13 3 Numerical Scheme Finite Volume Method Forward time stepping Locally and globally conservative 2 nd order accurate in space and time Structured Cartesian Mesh

14 4 Finite volume principle Continuity equation in flux form: t h t n+ t n Ω + h t ( hv) = Q Integrate over one time step t and the 2D finite volume Ω with area A: dωdt + t n+ Ω t n n+ ( hv) dtdω = Qdt t t n Ω t Apply the Gauss divergence theorem: n+ tn+ t n dh dt + dt F ndl ˆ = A Ω Ω t n Qdt F n ˆ : : h : Unit normal vector Time-averaged numerical flux Cell-averaged pressure thickness

15 5 Discrete Form : Orthogonal fluxes across cell interfaces tn+ t n dh dt dt t + A Ω Ω F ndl ˆ = 0 Flux form ensures mass conservation G i,j+/2 h n+ t 4 n = h AΩ i= F i nˆ l i i F i-/2,j (i,j) F i+/2,j Upwind-biased Finite volume method h n+ i, j = h n i, j t t A i, j y i+, j 2 G A xi, j + i, j + i, j 2 2 F i+, j 2 y x G i, j 2 i, j 2 i, j 2 F i, j 2 Flow direction G i,j-/2 F: fluxes in x direction G: fluxes in y direction

16 6 Test Case : Ideal Dam Break Problem Test Case Description: A channel was designed to connect two reservoirs, up hill- completely filled with water, downhill- completely empty Ideal dam removed. Water flows completely to the empty one

17 7 Result : Dam Break Problem Santillana et al Test case Simulation for 00 seconds of simulation IBM model

18 8 GSSHA Simulations DEM- SRTM Watershed Delineation Grid creation Land use, soil, vegetation etc. Precipitation flow model run

19 9 Work in Progress Code parallelization : OpenMP + MPI Real World Case Studies Parameterizations : infiltration, evapotraspiration etc Code Parallelization: F i- /2,j Flow direction G i,j+/ 2 F (i,j) G i,j-/2 i+/2, j T F i- /2,j Flow direction G i,j+/ 2 (i,j) G i,j-/2 F i+/2, j T2 Shared Memory parallelization using OpenMP Parallel computations for X and Y edges Parallel computations for grid cells T T2 T3

20 2 0 Need your Attention Thanks!!

21 2 Hydraulic models GSSHA CHL, USACE CCHE2D NCCHE, Mississippi university, USDA MIKE SHE DHI ISIS 2D Infoworks ICM

22 2 2 Two Approaches Hydrologic modelling Physics of the process is unknown Conceptual models Derived from experience and data statistics Hydraulic Modelling Physics based equations Involves Partial differential equations Need numerical methods to solve