Bayesian and Adjoint Inverse Model Analysis of PM Sources in the United States Using Observations from Surface, Aircraft, and Satellite PlatformsEPA Grant Number: R832158
Title: Bayesian and Adjoint Inverse Model Analysis of PM Sources in the United States Using Observations from Surface, Aircraft, and Satellite Platforms
Investigators: Jacob, Daniel J. , Seinfeld, John
Institution: Harvard University , California Institute of Technology
EPA Project Officer: Chung, Serena
Project Period: January 1, 2005 through December 31, 2007 (Extended to December 31, 2008)
Project Amount: $450,000
RFA: Source Apportionment of Particulate Matter (2004) RFA Text | Recipients Lists
Research Category: Particulate Matter , Air Quality and Air Toxics , Air
Our objective in this project is to improve knowledge of PM sources in the United States on a continental scale through combined Bayesian and adjoint inverse model analyses integrating observations from surface sites, aircraft, and satellites.
We will exploit data from the multi-agency ICARTT aircraft campaign over eastern North America in summer 2004 (http://www.al.noaa.gov/ICARTT/ Exit ) that will provide extensive mapping of PM composition and ancillary variables in the continental boundary layer. We will also use aerosol optical depth (AOD) data from the MODIS instrument aboard the NASA Terra and Aqua satellites. Information from these new data sets will be combined with PM composition data from IMPROVE and CASTNET surface sites in the United States towards the development of an improved ability to constrain regional PM sources.
Our project entails a comprehensive inverse model analysis of PM sources in the United States using atmospheric observations for a 5-year period [2000-2004] and with specific focus on the ICARTT campaign (summer 2004). We will use both Bayesian synthesis and adjoint modeling, allowing for the first time a comparison of the two approaches. The Bayesian synthesis approach is robust, relatively simple to implement, and provides error statistics as part of its solution. Its main drawback is that it can solve only for a limited number of pre-selected PM source parameters and must assume other parameters (e.g., subregional source distributions, seasonal variations, etc.) to be well known. The adjoint method is far more flexible and powerful in that it provides PM source information at the level of resolution of the driving CTM, but it is less robust and does not provide straightforward error statistics. At this stage of inverse model development it is extremely valuable to pursue and compare the two approaches.
The forward model in our inverse analysis will be the global GEOS-CHEM CTM (http://www-as.harvard.edu/chemistry/trop/geos Exit ) driven by assimilated meteorological data from the NASA Global Modeling and Assimilation Office (GMAO). Horizontal resolution will be 1° x 1°, with 55 levels in the vertical. The global simulation capability will allow the tracking of sources on scales ranging from regional to intercontinental; the latter can be important for dust and forest fire events. More importantly, a global model is essential to interpret satellite AOD observations as these include a substantial and variable component in the free troposphere. The GEOS-CHEM CTM is heavily documented (over 50 journal papers listed on the above web site).
Results from this project will advance in a major way our knowledge of important regional PM sources in the United States. For example, bottom-up source estimates for ammonia and carbonaceous aerosols are recognized to be highly uncertain; our work will provide improved constraints on these sources on a national scale. By improving knowledge of PM sources, our work will provide critical information for designing emission control strategies to meet PM2.5 and PM10 standards.
Our project will also represent a major theoretical advance in the application of inverse modeling methods to optimize PM sources. The adjoint method, recognized to be a particularly powerful and versatile tool, will be applied for the first time to the PM source problem. It will provide not only valuable estimates of sources, but also of model sensitivities to various parameters that govern the distribution and variation of PM. Parallel application of the more standard Bayesian synthesis method is an important feature of this proposal and will allow us to develop confidence in the application of the adjoint method.
A timely contribution from this work will be to explore the potential of aircraft and satellite observations for constraining PM sources. Over the next decade there will be a large increase in PM data from satellites. These will include data from two approved NASA missions: a space-based lidar (CALIPSO) to be launched in 2005 and an aerosol polarimeter (GLORY) to be launched in 2007. Observations from the MODIS instrument are intended to continue through the next decade as part of the NOAA NPOESS satellite program starting in 2010. There is also a strong interest at both NASA and the European Space Agency (ESA) to launch geostationary satellites providing continuous AOD observations over large regions of the world such as North America. Being able to tap into this vast array of PM data from space could improve considerably our long-term ability to quantify PM sources in the United States. By examining the consistency of PM observations collected from satellite, aircraft, and surface platforms within a common analytical framework, our project will greatly increase the ability of the research community to knowledgeably use satellite observations for constraining PM sources.
It should be noted that the optimization of sources enabled by inverse methods improves knowledge of emissions but does not by itself improve understanding. Inverse models represent essentially a statistical fit to observations. Departures from the bottom-up inventories indicate likely errors in the procedures and assumptions used to generate these inventories, and these must be corrected in order to improve understanding and enable better projections of emissions in the future. As part of our project we will initiate this next step by examining the information on activity rates, emission factors, and environmental variables used to construct the bottom-up inventories for which we have identified major flaws. Collaborations with colleagues at EPA and elsewhere that have expertise in the construction of bottom-up inventories will be a logical follow-up to the project.