Spatial Reduction Algorithm for Numerical Modeling of Atmospheric Pollutant Transport

Yevgenii Rastigejev, Philippe LeSager

Harvard University

Michael P. Brenner, Daniel J. Jacob

April, 2007

Motivation

Problem: There is an urgent need to improve significantly computational performance of the existing GEOS-Chem code motivated by difficulties with numerical simulation of immerging applications

Solution: To create a multi-scale numerical algorithm which provides a

significant (order of magnitude) reduction in computational cost

Our algorithm is inspired by the observation that atmospheric emissions have most of their fast reactants and their quickly decomposing reaction products localized near the emitter, typically on the ground. Far from the emitter, the fast reactants do not play a significant role.

Solution: Spatial Reduction

In the Slow region

For the algorithm to be efficient:

size Slow Region >> size Fast Region

comp. cost Slow Region << comp. cost Fast Region

Fast Region – the species is fully accounted as a

dependent variable in the system of transport

equations

Slow Region – the species concentration

is found through extrapolation

Gradual Spatial Reduction

Consider a two species reaction system, comprised of chemicals A and B

The central idea of our algorithm is to track a series of chemical boundary layers (CBL) within which the gradual, continuous spatial reduction is employed from the domain where the full mechanism is needed to the domain where the most simplified mechanism is sufficient.

Domain partitioning

• Fast-Slow domain partitioning algorithm
• Algorithms for Reduced Model construction
• Matching procedure
• Efficient methodology for adaptation in time

Prerequisites for Spatial Reduction Numerical Method

Reduced model – exponential extrapolation into Slow region

Matching conditions

matching procedure for concentration values and mass fluxes ensures good mass conservation and continuity properties of the numerical scheme

Adaptive algorithm summary

Chemical model

oxidation of a number of C3H6 in an OH oxidizing atmosphere, with NOx serving as catalyst through the cycling of HOx radicals

15 chemical species and 24 reactions

Fast

Slow

Fast – Slow domain partitioning for the atmosphere with horizontal shear layer flow at time

Comparison of species distribution obtained by the spatial reduction algorithm (spheres and cubes) and conventional method (solid line) depicted in blue and red colors at and at correspondingly at time

Max err = 2-3 %

Average err = 0.5 %

Some preliminary conclusions

• A gradual continuous spatial reduction computational method for Atmospheric chemistry has been developed.
• Unlike conventional time-scale separation methods, the spatial reduction algorithm allows to speed up not only a "chemical solver" but also advection-diffusion numerical integration.
• The method produced accurate results at 5-6 times lower computational cost for the 2-dimensional Atmospheric chemistry problems tested.
• Currently we are working on the algorithm implementation into GEOS-Chem three-dimensional solver.

To be continued with three-dimensional

description …

G-C Implementation

Old SOLVER

(SMVGEAR)

New SOLVER

(DLSODES)

Criteria for CBLs

[ n ] > Threshold Values

OR

Net Production > 10^-3 Max Value

54% of ODE solved

averaged RMS error of 7.2%

Run Time = 0.66 Full Resolution Case

Nb of Species

Short lived species : Max error in the outskirts of concentration peaks

Transported tracer: Max error further away from concentration peaks.

To be continued…

1. Better CBLs (combining absolute/relative criteria, discriminating b/w species, fine tuning)

​

• Extrapolation mechanism will
• substantially reduce the error in the slow region
• noticeably reduce the size of the fast region

​

• Full advantage of sparse structure
• substantial speed up

​