Dissertation Research

Consequences of the Delegation of Authority: Monitoring & Regulating Ambient Air Pollution

Local and regional pollutants like particulate matter (PM) have large health impacts and require government regulation due to a tragedy of the commons occurring in the atmosphere. In the United States, a critical component of PM regulation involves enforcing standards on ambient levels of PM using expensive in-situ (ground-based) pollution monitoring stations. These monitors relay measurements back to the regulator to allow for enforcement of the regulation, however monitors can only measure ambient PM at one location. The placement of monitoring sites is therefore crucial for enforcement — only locations with a monitor can be regulated and we only have data on emission sources that blow pollution toward the monitor. However, there has been little published work on the determinants of monitor locations. This project examines the pollution monitor siting decision in an effort to address two groups of questions:

Under the current regime of ambient pollution standards, what are the determinants of existing monitor locations? What groups of people are represented by the existing monitoring network?
How inefficient is our current method of choosing monitor locations? What is the potential welfare gain from moving monitors or optimally adding new monitors to the network?

Ambient Air Pollution Regulation vs Regulating at the source

Air pollutants are emitted from various sources and mixed into the atmosphere, where they chemically react, diffuse, and travel to the locations where they eventually cause damage. Air pollution regulation is designed to control or limit the amount of damage inflicted by air pollution, where one of the major types of air pollution policy is the regulation of ambient air pollution levels. Ambient air pollution is the concentration of pollutants found in the air in and around where people live, at the location where the damages occur. Ambient air pollution regulation is often distinct from air pollution regulations that control the amount of pollution coming from pollution sources. This is necessary because even if many pollution sources are limited to low levels of pollution, geography and weather may push pollution into hotspots, creating an uneven pollution distribution over space. Measuring and regulating air pollution concentrations around population centers has become an important tool in reducing damages of pollution, but measuring air pollution levels accurately at many locations where people live and work is still prohibitively expensive.

The ideal data for implementing ambient air pollution regulation would be measurements of ambient concentrations over time, at all points in space (or at least a reasonably dense number of measurement locations). However, the level of accuracy and precision required for pollution data in order to implement legally enforceable pollution regulation is high, and therefore expensive due to current technology constraints. There are cheaper pollution measurement devices coming online, however these currently only exist for a few pollutants (e.g., PM2.5 and ozone) and do not currently meet the high accuracy needs of regulation. The expensive nature of collecting accurate pollution data, and costs of compliance, implies a budgetary constraint on number of pollution monitoring stations state and federal governments are willing to place. This constraint on the number of monitors is the primary motivation for estimating optimal monitor locations.

Other Ongoing Projects

Ambient air pollution monitoring in the US: placement, timing, network valuation

I utilize Purple Air consumer-grade air pollution monitors to estimate the bias in location and timing of air pollution monitors in the US.

Renewable Energy Siting: Is conservation at odds with the clean energy transition? (with Meredith Fowlie and Charles Taylor)

I am currently collaborating on a project to examine the tension between the placement of renewable energy projects and environmental conservation. My contribution to this project involves simulating solar and wind projects at many locations around the US, including expected generation of electricity, revenues, cost, and tax credits. I am also modeling the network connections between planned projects to control for simultaneous impacts between projects in the estimation.

Value of Satellite Imagery in Pollution Regulation (with Hannah Druckenmiller)

My contribution to this project was to estimate the prediction error of satellite-based air pollution estimates. This utilized Google Earth Engine to run models on large raster datasets merged from various sources.

Identification of parameters in a joint model with AR(1) errors (with Larry Karp)

I am wrapping up a collaboration on a project to rank potential international climate change agreements. I provided the empirical programming to simulate and apply the theoretical model. A notable feature about this work is that I needed to test the various estimation techniques we applied, which requires running many simulations to understand the small sample behavior of different estimators.

Taking the multicollinearity out of survey data to estimate total effects: beekeeper surveys (with Robin Cross)

Previous to my PhD program, I completed a master’s degree in Applied Economics where I focused on econometric techniques applied to survey data. I worked with biologists to publish the analysis. My contribution was to develop the econometric application of the new econometric technique that my advisor published.

Machine learning and satellite imagery: building a cheaper honey bee census from space (with Robin Cross)

To develop additional data for my master’s research, I trained various image classification and object identification models on hand-labeled satellite imagery to count managed honey bee hives in agricultural fields.

Using machine learning and statistical analysis to aid in sample selection: sampling honey bee colonies to test for mites (with Robin Cross)

During surveys, researchers often will need to sample an inventory to get an approximation of some characteristic that is important for policy or decision making. In agricultural surveys for honey bees, researchers cannot test all of the hundreds of managed hives at a given location so must sample a handful of hives to estimate the health of the bees. However, without careful sampling design, sampling 5 of 100 hives may lead to large measurement errors and incorrect policy decisions. I developed an algorithm for optimal sampling of honey bee hives that utilizes machine learning to identify hives in a user-provided image. Given the locations of the hives, and external survey-based estimates of disease probabilities,