Research

Inverse Modeling

I have been working on inverse modeling to improve emissions inventories using long time series of hourly measurements at fixed sites.

The inverse model was used to estimate emissions of elemental mercury from lake surface outgasing, forest fires, power plants and unreported sources using measurements in Milwaukee (de Foy et al., 2012).

It was then expanded to analyze sources of reactive mercury and to differentiate the impacts of direct emissions, transport from the free troposphere and different oxidation pathways (de Foy et al., 2014).

I am currently working on using it to evaluate diurnal and annual patterns of emissions of elemental and organic carbon (de Foy et al, 2014).

The model is based on a least squares simplification of the classic Bayesian formulation. This can be done if we consider only the transport of pollutants or if we can represent the impact of chemical formation with a time series at the measurement site. Using Iteratively Reweighted Least Squares (IRLS) makes the results much more robust to outliers. The beauty of the least squares inversion is the speed at which it can be done. Uncertainty estimates can therefore be obtained by using block-bootstrapping. If we assume that measurement errors are randomly distributed, and if model errors are different on any given day, then we can simulate the uncertainty in the results by selecting the days to be included in the inversion at random. This provides larger, yet more realistic, error bars than the normal least squares standard errors. Another strength is that it does not require a priori estimates of the uncertainty as in Monte Carlo Error Propagation, but rather deduces what these are from the data itself. A second advantage of the least squares method is that it is straightforward to include different types of sources in the analysis from different models. For example, the model can combine both Eulerian and Lagrangian models.

Satellite Data Analysis

Most air quality products from satellites have a grid size larger than 10 by 10 km². With swath data that has varying resolution and grids in different places from day to day, it is possible to use oversampling to obtain information on finer grids. I used this in Mexico City to obtain high resolution maps of the sulfur dioxide plumes from the Popocatepetl volcano and from an industrial complex (de Foy et al., 2009).

I am working on getting improved estimates of emissions from large point sources as well as chemical lifetimes. The accuracy of different methods was tested using a numerical transport model (de Foy et al., 2014).

Mexico City

I did my postdoc with Prof. Mario Molina and Dr. Luisa Molina in the Mexico City Project at MIT. During this time, I was involved as a weather forecaster and computer modeler in the MCMA-2003 and the MILAGRO field campaigns. Many of the papers from the field campaigns are available in Atmospheric Chemistry and Physics in the special issues for MCMA-2003 and for MILAGRO.

I used numerical simulations from MM5 and WRF as input to the FLEXPART particle trajectory model. Even though winds in the Mexico City basin are very complex and there are large errors in the simulations of instantaneous wind fields, we showed that overall the atmospheric transport is correctly characterized by the models (de Foy et al., 2009). The simulations were used by multiple groups to help analyze the field campaign measurements.

In de Foy et al., 2011, we compared measurements made with the High Spectral Resolution Lidar (HSRL) with particle trajectory simulations to analyze the sources and transformation of particles in the atmosphere in Mexico City. This showed that the HSRL detected atmospheric tar balls in fresh biomass plumes.

In de Foy et al., 2007 we analyzed surface and column measurements, along with Concentration Field Analysis and Residence Time Analysis to study the transport of pollutants and to evaluate emissions inventories for Mexico City.

As a result of the field campaigns, we found that there was rapid and effective ventilation of pollutants in Mexico City (de Foy et al., 2006). On most days, the pollutants are flushed out of the basin on the same day. In this way, Mexico City is actually more similar to Houston where the sea breeze cleans out the city every day than it is to Los Angeles where a capping inversion leads to multi-day accumulation.

Urban Heat Island

Many tropical cities have an urban heat island at night and a cool island during the day. The behavior of the heat island varies with the seasons. There are different ways of measuring the urban heat island. One way is to compare meteorological observations of the air temperature in the city and in the surrounding country side. Another way is to use satellite remote sensing of the land surface temperature. My student and I showed that the seasonal behavior of the urban heat island is different depending on which measurement method is used (Cui and de Foy, 2012). We also showed that daytime urban heat island were a function of vegetation differentials between the city and its surroundings, and that the nighttime urban heat island was a function of atmospheric stability.

Humming earth due to ocean waves

In an interdisciplinary project, I contributed model and satellite data analysis to studies of the earth's hum by Keith Koper which led to several joint publications starting with Koper and de Foy, 2008.