Smart flood forecasting infrastructure with uncertainties. Georges Kesserwani University of Sheffield

Transcription

1 Smart flood forecasting infrastructure with uncertainties Georges Kesserwani University of Sheffield 25 January 29/01/ TSUNAMI-CODE - Advancing our ability to forecast Urban multi-scale Flood Modelling flooding Seminar

2 Acknowledgements

3 1. Strategic Importance Need for improved flood forecasting tools POLICY-MAKERS WATER INDUSTRY ACADEMIC CALLS Increased budgets won t be enough Increased Colorado versatility Aug. Adapt 2015 itself to the and intelligence optimum scale Reliable warning maps for the public Efficient handling of uncertainties More accurate and multi-purpose Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

4 2. Present Challenges RESOLUTION ACCURACY Robust flood model TECHNIQUE Error control Data transfer Fitness for purpose Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

5 2. Present Challenges RESOLUTION ITERATIONS ACCURACY Robust flood model TECHNIQUE Error control Data transfer Fitness for purpose Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

6 2. Present Challenges RESOLUTION UNCERTAINTY QUANTIFICATION ITERATIONS ACCURACY Robust flood model TECHNIQUE Error control Data transfer Fitness for purpose Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

7 3. Vision Intelligent TECHNIQUE Holistic UNCERTAINTY QUANTIFICATION Automated decision-making RESOLUTION ACCURACY Various scales of grid and domain sizes ITERATIONS Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

8 4. Technique MULTI-WAVELETS POLYNOMIAL ACCURACY UNCERTAINTY STATISTICS RESOLUTION DECISION Single model structure Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

9 5. Concept Fine Coarse Details Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

10 5. Concept Retain the significant details by truncation according to a threshold ε: d j (lev) = d j (lev) if d j lev > ε 2 lev L 0 otherwise Encode the details to decide an adaptive mesh and then carry out the calculation. Only one parameter is needed, which is the ε: 0 < ε < 1 Dr G Kesserwani Smart 29/01/2018 flood forecasting infrastructure with uncertainties TSUNAMI-CODE - Advancing our ability to forecast multi-scale flooding

11 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε 0.001

12 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε Levels( = 0.1) HFV( = 0.1) Levels( = ) HFV( = ) Levels( = 0.01) Levels( = 0.001) HFV( = 0.01) HFV( = 0.001) (a) ε = 0.01 ε = 0.04 ε = h+z (m) 5 5 Levels x (m)

13 h+z (m Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) 2 N = 1 (coarsest datum represented by 1 cell) ε e-16 0 Levels( = 0.1) Levels( = ) Levels( = 0.01) Levels( = 0.001) HFV( = 0.1) HFV( = ) HFV( = 0.01) HFV( = 0.001) (a) (b) 9 5e x (m) q (m 2 /s) e-17 h+z (m) 5 5 Levels -1e x (m) x (m)

14 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε 0.001

15 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε ε = s T = 2.5s (Exact) T = 0s (Num.) T = 0s (Levels) T = 2.5s (Num.) T = 2.5s (Levels) T = 40s (Num.) T = 40s (Levels) h+z (m) s 2. 5s Levels x (m)

16 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε ε = ε = ε = s

17 6. Feasible basis for Finite Volume models L = 9 (max. level of resolution allowing 2 9 = 512 cells) N = 1 (coarsest datum represented by 1 cell) 0.1 ε ε = nd order 1 st order 2. 5s

18 7. Current and future work MULTI-WAVELETS POLYNOMIAL ACCURACY UNCERTAINTY STATISTICS RESOLUTION DECISION Dr G Kesserwani Single model structure Polynomial accuracy in 2D for realistic applications (J. Ayog) Agent-based version on GPUs (M. Shirvani) Stochastic formulation (pending) Smart forecasting: joined-up flood forecasting (FF) infrastructure with uncertainties

19 Polynomial accuracy in 2D for realistic applications (J. Ayog) Fully well-balanced discontinuous Galerkin flood model J. Ayog, G. Kesserwani and D. Dau

20 Well-balanced: Discontinuous topography J L Ayog Advanced flood risk assessment based on Discontinuous Galerkin flood model formulation

21 Results: Not well-balanced for the slopes Slope coefficients J L Ayog Advanced flood risk assessment based on Discontinuous Galerkin flood model formulation

22 Results: Well-balanced for the slopes All coefficients J L Ayog Advanced flood risk assessment based on Discontinuous Galerkin flood model formulation

23 Agent-based version on GPUs (M. Shirvani) A dynamic multi-agent based flood modelling on GPUs M. Shirvani, G. Kesserwani and P. Richmond

24 Numerical flood model Finite Volume Method to solve shallow water equations U n+1 i,j = U n i,j t x F n i+ 1 2,j n F 1 i 2,j t y G n i,j+ 1 2 n G 1 i,j 2 + t S(U)

25 Main challenge Cannot do sequential calculation

26 Main challenge

27 Thanks for listening. Questions? Georges Kesserwani: