Computational Challenges in Air Pollution Modelling Z. Zlatev National Environmental Research Institute 1. Why air pollution modelling? 2. Major physical.

Slides:



Advertisements
Similar presentations
Steady-state heat conduction on triangulated planar domain May, 2002
Advertisements

1 AirWare : AirWare : C OMPREHENSIVE A IR QUALITY M ODEL WITH E X TENSIONS VERSION 4.30, PBM DDr. Kurt Fedra Environmental Software & Services GmbH A-2352.
Fighting the Great Challenges in Large-scale Environmental Modelling I. Dimov n Great challenges in environmental modelling n Impact of climatic changes.
Computational Methods in Physics PHYS 3437
AIR POLLUTION. ATMOSPHERIC CHEMICAL TRANSPORT MODELS Why models? incomplete information (knowledge) spatial inference = prediction temporal inference.
Total Recall Math, Part 2 Ordinary diff. equations First order ODE, one boundary/initial condition: Second order ODE.
PART 7 Ordinary Differential Equations ODEs
Fast Convolution Algorithm Alexander Eydeland, Daniel Mahoney.
Spatial Reduction Algorithm for Numerical Modeling of Atmospheric Pollutant Transport Yevgenii Rastigejev, Philippe LeSager Harvard University Michael.
/08/2002SMARTER meeting 1 Solution of 2D Navier-Stokes equations in velocity-vorticity formulation using FD Remo Minero Scientific Computing Group.
Atmospheric modelling activities inside the Danish AMAP program Jesper H. Christensen NERI-ATMI, Frederiksborgvej Roskilde.
A Concept of Environmental Forecasting and Variational Organization of Modeling Technology Vladimir Penenko Institute of Computational Mathematics and.
CHAPTER 3: SIMPLE MODELS
Ordinary Differential Equations (ODEs) 1Daniel Baur / Numerical Methods for Chemical Engineers / Implicit ODE Solvers Daniel Baur ETH Zurich, Institut.
Large-Scale Density Functional Calculations James E. Raynolds, College of Nanoscale Science and Engineering Lenore R. Mullin, College of Computing and.
Method of manufactured solutions The first stage in code verification is to test for problems for which we have analytical solutions. However, often we.
Ordinary Differential Equations (ODEs)
Jenny Stocker, Christina Hood, David Carruthers, Martin Seaton, Kate Johnson, Jimmy Fung The Development and Evaluation of an Automated System for Nesting.
Session 9, Unit 17 UAM and CAMx. UAM and CAMx UAM - Urban Airshed Model Currently available versions:  UAM-V 1.24  UAM-V 1.30  Available from Systems.
CMAQ (Community Multiscale Air Quality) pollutant Concentration change horizontal advection vertical advection horizontal dispersion vertical diffusion.
3rd EGEE User Forum Clermont-Ferrand ’2008 – Tz. Ostromsky – 11 February 2008 Enabling Grids for E-sciencE IPP - BAS 1 Long range air pollution transport.
1 CCOS Seasonal Modeling: The Computing Environment S.Tonse, N.J.Brown & R. Harley Lawrence Berkeley National Laboratory University Of California at Berkeley.
Environmental Modelling, Security Measures and Decision Making Zahari Zlatev National Environmental Research Institute Frederiksborgvej 399, P. O. Box.
« Data assimilation in isentropic coordinates » Which Accuracy can be achieved using an high resolution transport model ? F. FIERLI (1,2), A. HAUCHECORNE.
Using Partitioning in the Numerical Treatment of ODE Systems with Applications to Atmospheric Modelling Zahari Zlatev National Environmental Research Institute.
Implementation of the Particle & Precursor Tagging Methodology (PPTM) for the CMAQ Modeling System: Mercury Tagging 5 th Annual CMAS Conference Research.
Regional climate prediction comparisons via statistical upscaling and downscaling Peter Guttorp University of Washington Norwegian Computing Center
Lecture 3.
Sept COMP60611 Fundamentals of Concurrency Lab Exercise 2 Notes Notes on the finite difference performance model example – for the lab… Graham Riley,
Overview of Techniques for Deriving Emission Inventories from Satellite Observations Frascati, November 2009 Bas Mijling Ronald van der A.
Experiences with the Globus Toolkit on AIX and deploying the Large Scale Air Pollution Model as a grid service Ashish Thandavan Advanced Computing and.
© 2011 Autodesk Freely licensed for use by educational institutions. Reuse and changes require a note indicating that content has been modified from the.
7. Introduction to the numerical integration of PDE. As an example, we consider the following PDE with one variable; Finite difference method is one of.
4. Atmospheric chemical transport models 4.1 Introduction 4.2 Box model 4.3 Three dimensional atmospheric chemical transport model.
Serge Andrianov Theory of Symplectic Formalism for Spin-Orbit Tracking Institute for Nuclear Physics Forschungszentrum Juelich Saint-Petersburg State University,
Georgia Institute of Technology Initial Application of the Adaptive Grid Air Quality Model Dr. M. Talat Odman, Maudood N. Khan Georgia Institute of Technology.
Air quality decision support under uncertainty (case study analysis) Piotr Holnicki Systems Research Institute PAS Warszawa, Newelska 6
Outline Numerical implementation Diagnosis of difficulties
TEMPLATE DESIGN © A high-order accurate and monotonic advection scheme is used as a local interpolator to redistribute.
Variational data assimilation: examination of results obtained by different combinations of numerical algorithms and splitting procedures Zahari Zlatev.
THE MODELS-3 COMMUNITY MULTI- SCALE AIR QUALITY (CMAQ) MODEL: 2002 RELEASE – NEW FEATURES Jonathan Pleim, Francis Binkowski, Robin Dennis, Brian Eder,
Discretization Methods Chapter 2. Training Manual May 15, 2001 Inventory # Discretization Methods Topics Equations and The Goal Brief overview.
Discretization for PDEs Chunfang Chen,Danny Thorne Adam Zornes, Deng Li CS 521 Feb., 9,2006.
Types of Models Marti Blad Northern Arizona University College of Engineering & Technology.
Georgia Institute of Technology SUPPORTING INTEX THROUGH INTEGRATED ANALYSIS OF SATELLITE AND SUB-ORBITAL MEASUREMENTS WITH GLOBAL AND REGIONAL 3-D MODELS:
Further Developments and Applications for the Adjoint of CMAQ
Ch 8.2: Improvements on the Euler Method Consider the initial value problem y' = f (t, y), y(t 0 ) = y 0, with solution  (t). For many problems, Euler’s.
___________________________________________________________________________CMAQ Basics ___________________________________________________Community Modeling.
Onlinedeeneislam.blogspot.com1 Design and Analysis of Algorithms Slide # 1 Download From
A Parallel Linear Solver for Block Circulant Linear Systems with Applications to Acoustics Suzanne Shontz, University of Kansas Ken Czuprynski, University.
Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency policy Jonathan Pleim, Shawn Roselle,
Multipole-Based Preconditioners for Sparse Linear Systems. Ananth Grama Purdue University. Supported by the National Science Foundation.
The application of Models-3 in national policy Samantha Baker Air and Environment Quality Division, Defra.
7. Air Quality Modeling Laboratory: individual processes Field: system observations Numerical Models: Enable description of complex, interacting, often.
Math for CS Fourier Transform
8. Box Modeling 8.1 Box Models –Simplest of atmospheric models (simple saves $). –Species enter the box in two ways: 1. source emissions 2. atmospheric.
1 The Interactions of Aerosols, Clouds, and Radiation on the Regional Scale.
Lecture 19 MA471 Fall 2003.
CSE 245: Computer Aided Circuit Simulation and Verification
J-Zephyr Sebastian D. Eastham
Class Notes 18: Numerical Methods (1/2)
Ensemble variance loss in transport models:
topics Basic Transmission Line Equations
Models of atmospheric chemistry
COMP60621 Designing for Parallelism
OVERVIEW OF FINITE ELEMENT METHOD
Conservative Dynamical Core (CDC)
CASA Day 9 May, 2006.
Chapter 2 A Survey of Simple Methods and Tools
Atmospheric modelling of HMs Sensitivity study
Presentation transcript:

Computational Challenges in Air Pollution Modelling Z. Zlatev National Environmental Research Institute 1. Why air pollution modelling? 2. Major physical and chemical processes 2. Major physical and chemical processes 3. Need for splitting 3. Need for splitting 4. Computational difficulties 5. Need for faster and accurate algorithms 6. Different matrix computations 7. Inverse and optimization problems 8. Unresolved problems 4. Computational difficulties 5. Need for faster and accurate algorithms 6. Different matrix computations 7. Inverse and optimization problems 8. Unresolved problems

1. Why air pollution models? n Distribution of the air pollution levels n Trends in the development of air pollution levels n Establishment of relationships between air pollution levels and key parameters (emissions, meteorological conditions, boundary conditions, etc.). n Predicting appearance of high levels

2. Major physical processes n Horizontal transport (advection) n Horizontal diffusion n Deposition (dry and wet) n Chemical reactions + emissions n Vertical transport and diffusion Describe these processes mathematically

3. Air Pollution Models

4. Need for splitting n Bagrinowskii and Godunov 1957 n Strang 1968 n Marchuk 1968, 1982 n McRay, Goodin and Seinfeld 1982 n Lancer and Verwer 1999 n Dimov, Farago and Zlatev 1999 n Zlatev 1995

4. Criteria for choosing the splitting procedure n Accuracy n Efficiency n Preservation of the properties of the involved operators

5. Resulting ODE systems

6. Size of the ODE systems n (480x480x10) grid and 35 species results in ODE systems with more than 80 mill. equations (8 mill. in the 2-D case). n More than time-steps are to be carried out for a run with meteorological data covering one month. n Sometimes the model has to be run over a time period of up to 10 years. n Different scenarios have to be tested.

7. Chemical sub-model n Parallel tasks The calculations at a given grid-point The calculations at a given grid-point n Numerical methods QSSA (Hesstvedt et al., 1978) QSSA (Hesstvedt et al., 1978) Backward Euler (Alexandrov et al., 1997) Backward Euler (Alexandrov et al., 1997) Trapezoidal Rule (Alexandrov et al., 1997) Trapezoidal Rule (Alexandrov et al., 1997) Runge-Kutta methods (Zlatev, 1981) Runge-Kutta methods (Zlatev, 1981) Rosenbrock methods (Verwer et al., 1998) Rosenbrock methods (Verwer et al., 1998) Criteria for choosing the numerical method?

8. Advection sub-model n Parallel tasks The calculations for a given compound The calculations for a given compound n Numerical methods Pseudo-spectral discretization (Zlatev, 1984) Pseudo-spectral discretization (Zlatev, 1984) Finite elements (Pepper et al., 1979) Finite elements (Pepper et al., 1979) Finite differences (up-wind) Finite differences (up-wind) “Positive” methods (Bott, 1989; Holm, 1994) “Positive” methods (Bott, 1989; Holm, 1994) Semi-Lagrangian algorithms (Neta, 1995) Semi-Lagrangian algorithms (Neta, 1995) Wavelets (not tried yet) Wavelets (not tried yet)

9. Discretization of the derivatives

10. Pseudo-spectral discretization

11. Convergence of the Fourier series If f(x) is continuous and periodic and if f´(x) is piece-wise continuous, then the Fourier series of f(x) converges uniformly and absolutely to f(x). Davis (1963)

12. Accuracy of the Fourier series It can be proved (Davis, 1963) that if

13. Drawbacks of the pseudo-spectral method

14. Finite elements The application of finite elements in the advection module leads to an ODE system: P is a constant matrix, H depends on the wind Choice of method

15. Matrix Computations n Fast Fourier Transforms n Banded matrices n Tri-diagonal matrices n General sparse matrices n Dense matrices Typical feature: The matrices are not large, but these are to be handled many times in every sub-module during every time-step

16. Major requirements n Efficient performance on a single processor n Reordering of the operations What about parallel tasks? “Parallel computation actually reflects the concurrent character of many applications” D. J. Evans (1990) D. J. Evans (1990)

17. Chunks on one processor SIZE Fujitsu SGI IBM SMP First line: the straight-forward call of the box routine Last line: the vectorized option Second line: using 192 chunks Owczarz and Zlatev (2000)

18. “Non-optimized” code Module Comp. time Percent Chemistry Advection Initialization Input operations Output operation Total IBM SMP computer, one processor

19. Parallel runs on IBM SMP Processors Advection Chemistry Total IBM “Night Hawk” (2 nodes); NSIZE=48

20. Scalability Process (288x288) (96x96) Ratio Advection Chemistry Total IBM “Night Hawk” (2 nodes); NSIZE=48

22. Why is a good performance needed? Grid Comp. Time Grid Comp. Time (96x96) 424 (45.8) (96x96) 424 (45.8) (288x288) 6209 ( 3.1) (288x288) 6209 ( 3.1) Non-optimized code: IBM “Night Hawk” (2 nodes); NSIZE=48

23. PLANS FOR FUTURE WORK n Improving the spatial resolution of the model used to obtain information. n Object-oriented code n Predicting occurrences where the critical levels will be exceeded. n Evaluating the losses due to long exposures to high pollution levels. n Finding optimal solutions.

24. Unresolved problems n 3-D models on fine grids n Local refinement of the grids n Data assimilation n Inverse problems n Optimization problems Important for decision makers