Grantee Research Project Results

2002 Progress Report: Investigation of Four-Dimensional Data Assimilation Methodologies for Air Quality Models

EPA Grant Number: R827113
Title: Investigation of Four-Dimensional Data Assimilation Methodologies for Air Quality Models
Investigators: Xiu, Aijun , Hanna, Adel , Mathur, Rohit , Zou, Xiaolei
Institution: MCNC / North Carolina Supercomputing Center , Florida State University
EPA Project Officer: Hahn, Intaek
Project Period: October 1, 1998 through September 30, 2001 (Extended to November 8, 2003)
Project Period Covered by this Report: October 1, 2001 through September 30, 2002
Project Amount: $371,737
RFA: Exploratory Research - Physics (1998) RFA Text |  Recipients Lists
Research Category: Air Quality and Air Toxics , Air , Safer Chemicals , Land and Waste Management

Objective:

The objectives of the research project are to: (1) develop useful insights into the relatively unexplored, though potentially beneficial, area of chemical data assimilation; and (2) improve the predictive capability of atmospheric chemistry-transport models through the systematic investigation of the potential use of chemical data assimilation methods. Comprehensive atmospheric chemistry-transport models have played a central role in both research and policy development. Emerging multipollutant issues place additional demands on the role of such models in the design of effective abatement strategies. Although data assimilation methods have been widely used to improve the skill of meteorological models, the application of such methods to tropospheric chemistry problems has been rather limited. Data assimilation of chemical species observations within such models could allow knowledge of pollutant transport and chemistry to be incorporated in the analysis procedure and facilitate the propagation of information from asynoptic observations to model predictions, consequently improving the model's skill.

Progress Summary:

There are two widely used four-dimensional data assimilation methodologies for assimilating observation data into numerical models: the nudging (or Newtonian relaxation scheme) and the four-dimensional variational assimilation (4DVAR) scheme. The nudging scheme relaxes the model state toward observations during the assimilation period by adding to the prognostic equations nonphysical diffusive-type terms based on the difference between the model and the observations. The 4DVAR uses the variational method to define the data assimilation problem as an optimization problem, with the numerical model being a strong constraint. Because of the complex nature of the 4DVAR for dealing with large dimension problems, the adjoint technique is introduced to efficiently calculate the gradient of a cost or distance function with respect to a control variable, which may include initial conditions and/or model parameters.

We have used a chemical box model as a test bed for the investigation of the two data assimilation methods. This chemical box model has the Carbon Bond IV mechanism and uses the Modified Quasi Steady State Approximation solver. The initial conditions are based on the Intergovernmental Panel on Climate Change (IPCC) Photochemical Model Intercomparison Study. We have successfully developed the nudging scheme and the tangent linear model and adjoint model for the chemical box model.

We performed sensitivity tests within the chemical box model with different data assimilation strategies, such as perturbing different species or their combinations, nudging different species or their combinations, and testing different nudging coefficients. Sensitivity tests of the nudging scheme in the chemistry box model illustrate the complexities of applying data assimilation to chemically reactive nonlinear systems. In these isolated chemical kinetic experiments, assimilating O3 improved the predictions for O3, but not the overall quality of the simulation. However, assimilating primary species, such as NOx (NO and NO2), resulted in improved predictions for all species. In reality, however, measurements of ambient NOx are quite limited compared to O3. The accuracy of the nudging method also is sensitive to the nudging coefficient. These results indicate that the design of the chemical data assimilation system should carefully consider the feedbacks and interactions that arise from complex coupled chemical reactions between various atmospheric species.

To investigate the effect of the data assimilation techniques on real world problems, we implemented the nudge scheme in a three-dimensional (3-D) air quality model, the Multiscale Air Quality Simulation Platform (MAQSIP). We ran MAQSIP for a 4-km resolution nested domain over North Carolina, in which we have dense surface O3 observation stations. In the preliminary tests, we only nudged surface O3 observations into the 3-D model. Compared to base case simulations (without data assimilation) and observations, data assimilation had little impact on domain-wide daytime O3 prediction, but helped reduce night-time overpredicted O3 concentrations.

We continued developing and testing the tangent linear and adjoint models for the representative tropospheric chemistry box model. Both chemical and meteorological variables are chosen as active variables to facilitate more flexible applications, and the tangent linear model and adjoint model are developed for the chemistry box model. The chemical variables include species concentration, reaction rates, and other related variables. We tested the correctness of both the tangent linear and adjoint models and found the solution to the linearization problem caused by the use of varying time steps in the box model. We also reduced the computational cost of the adjoint model by increasing a limited amount of memory. We incorporated the limited-memory quasi-Newton method for minimization in the tropospheric chemistry box model.

Future Activities:

Future activities are to perform extensive experiments with the nudging scheme in the 3-D air quality model. We will do the "twin tests" with pseudo observations and nudge different species with various radiuses of influence, nudging coefficients, and time windows. We will assimilate O3 for different time periods, such as all day or just at night, to see how the predicted O3 concentrations are affected. We will continue to develop the tangent linear and adjoint models for the chemistry box model to make them more robust. We will test the minimization routine, the tangent linear model, and the adjoint model by performing the "twin experiment." We believe that the methodologies and results will eventually help the new U.S. Environmental Protection Agency and National Oceanic and Atmospheric Administration program on real-time air quality forecasting. We will prepare a journal article and give presentations at various conferences. We also will prepare the final report for this project.

Journal Articles:

No journal articles submitted with this report: View all 5 publications for this project

Supplemental Keywords:

meteorology, photochemistry, data nudging, adjoint method, real-time air quality forecasting., RFA, Scientific Discipline, Air, Environmental Chemistry, Engineering, Chemistry, & Physics, tropospheric ozone, Atmospheric Sciences, Physics, meterology, photochemistry, real time monitoring, atmospheric models, air quality modeling, air pollution models, atmospheric chemistry transport models, air pollution modeling system, chemical kinetics, assimilation methodologies, chemical transport model, pollutant transport

Relevant Websites:

http://www.cep.unc.edu/empd/projects/abstracts/03061417.htm Exit

Progress and Final Reports:

Original Abstract

1999 Progress Report

2000 Progress Report

2001 Progress Report

2003

Final Report