The exposures I study are invisible to the systems built to catch them, and impossible to miss if you live there. I count them.
Coastlines are Natural Earth. Sites and exposures are illustrative, not measurements.
Risk indices add hazard scores together, which only works if hazards are free to co-occur. Most are not. Some turn up together far more often than chance, others almost never share a place at all.
Those tendencies cancel when you sum them, so the average looks independent and the index looks sound. The departure sits in the tail, among the places carrying several hazards at once, which is exactly what the index was built to find.
Where that overlap lands is not random. Two burdens settling on the same neighbourhood open a wider gap than either one, measured alone, would lead you to expect.
And a shared hazard is not equally survivable. Comparing outcomes inside one place, in one year, separates how much exposure people receive from how much harm it does them. The question is not only who is exposed, but who withstands it.
Environmental hazards are studied one at a time and scored as though they add up. My own work asks whether that holds, who ends up inside the overlap, and what it costs them.
Most hazard pairs are not independent. Some cluster, some almost exclude one another, and the two tendencies cancel when you add them up — so an index built on averages looks safe and understates the places carrying the most.
01Joint exposure is not spread evenly across a city. Measuring where two burdens land on the same neighbourhood shows a wider gap than either one measured on its own would suggest.
02The same hazard is not equally survivable everywhere. Comparing outcomes within one place, in one year, separates how much exposure people receive from how much harm it does them.
03Work with the Community Noise Lab, at the other end of the scale. One school district, one neighbourhood, a few hundred buildings — places with no monitoring record at all.
Soil contamination, hearing screening, a statewide sign census. Datasets that did not exist before, and the calibration work that lets other people trust them.
04Shops, signs, streets and derelict lots, treated as exposure. Not chosen, not easily avoided, and rarely counted by anyone.
05Spatial dependence, spillover between neighbours, and coverage-aware frames that reconstruct where a survey looked rather than only what it found.
06Environmental exposure data is spatially dependent, measured with error, and observed under incomplete coverage. These are the approaches I work with.
Telling apart the case where one area's exposure changes its neighbour's outcome from two areas simply resembling each other.
Bayesian models with spatially structured effects, using which areas touch which to steady estimates where observation is thin.
Modelling where events actually fall rather than counting them inside boundaries, so the answer survives a change of map.
Turning scattered measurements into a continuous surface, then matching it to the population units decisions are made in.
Handling how often two exposures turn up together, and how often both are extreme at once, without assuming they are unrelated.
Checking cheap sensors against reference instruments, then testing the correction at sites left out of the calculation.
Recording where the survey actually went, so an area with no data reads as unvisited rather than as somewhere nothing was found.
Combining separate measurements into one comparable number, keeping the parts visible so nobody takes the total on trust.
Following how long each shop or site survives, so change within a place is told apart from change in what is being counted.
Allowing for repeat visits to the same site, and for households, neighbourhoods and counties each adding their own variation.
Combining what is already known with what the data show, then checking how much of the answer depends on the assumptions it started from.
Checking that two people applying the same protocol record the same answer, which is what makes a community survey usable.
Most of the state sits far from the nearest official air monitor, so people had no way to know what they were breathing. Monitors went into communities that asked for them, and were checked against professional-grade instruments to make sure the readings could be trusted anywhere else.
Cities that most need to keep track of derelict buildings are usually the least able to pay for it. High school researchers surveyed an entire city from public imagery, and the method keeps a record of where they looked, so an empty patch on the map means nobody went there rather than nothing being wrong.
Testing soil for heavy metals is slow and costly, so most dumping sites never get tested. A quick reading taken on site with a handheld instrument turns out to track how contaminated the soil really is, which gives a city a cheap way to work out where to look properly first.
Cities clear illegal dumping sites by taking away what can be seen, and the soil underneath is almost never tested. Sampling at the surface and below it across every reported site in one city found lead, copper and zinc enriched, at depth close to surface levels. Clearing the waste does not lift the burden. The elements driving calculated health risk turn out to be largely natural, while the one breaching regulation is the one that was dumped.
In much of the rural South the nearest audiologist is hours away, so hearing problems in children go unnoticed for years. Screening run inside schools with portable equipment shows this can be done without a clinic behind it.
Short-haul drivers spend their working lives surrounded by noise but fall outside most workplace hearing programmes. Testing them found hearing worse than expected for their age, and almost none had ever been checked.
After years of water crises, a test result on its own persuades nobody. Students tested water in their own community alongside an education programme, and who was doing the asking mattered as much as what the readings said.
Hot days and polluted days are usually studied separately, but they often land together, and not evenly across a city. Looking at both at once shows the overlap sits heaviest in neighbourhoods that were marked out for disinvestment generations ago.
In the same place and the same year, death rates climb more steeply with combined climate exposure for Black residents than for White residents. The gap is not explained by which counties people live in, which points at conditions inside places rather than differences between them.
It is usually assumed a disadvantaged neighbourhood carries every environmental burden at once. Measuring noise, air pollution and night-time light directly across an entire state shows they do not line up the way that assumption predicts.
Noise and air pollution are often folded into a single environmental score. Measuring both in the same places at the same times shows they behave differently. Noise depends on where you are, air pollution on when you measure, and one combined number hides that.
Over a generation, dollar stores spread across the state while grocery stores closed. Following every shop that opened and closed shows how quickly the places people can buy food changed, and where the change was sharpest.
Most measures of a neighbourhood's food supply treat every shop the same and describe only the present day. This one accounts for what dollar stores actually stock, and follows how each area changed over a generation.
Lower-income neighbourhoods have more alcohol and unhealthy food outlets, and that gap is usually put down to history. Following every opening and closure over a decade shows it is being actively maintained by where new businesses choose to open.
Billboards are something residents cannot opt out of, and work in large cities reports they cluster in Black and Latino neighbourhoods. Checking every billboard in a state on the ground shows education, income and housing tenure track exposure while racial composition does not. What differs between neighbourhoods is how much advertising residents meet, not what it says.
India set out to build a national digital health system, and by the middle of the decade adoption had split sharply between large urban hospitals and rural primary care. This traces what the policy actually delivered, where it stalled, and what the gap between the two says about digitising health care in a country that size.
When a factory pollutes one neighbourhood, its neighbours are affected too, and the usual statistical methods often cannot tell that apart from a place simply resembling the ones around it. This review sets out which approach to use depending on which of those is actually happening.
A large share of the world's waste is dumped or burned outside any controlled system. Cities cannot clean up what they never write down, and the gap between what is happening and what gets recorded is itself the problem.
Public health has a mature toolkit for data that went missing, and almost none for measurements nobody ever attempted. Six methods scattered across ecology, econometrics, official statistics and disease surveillance turn out to solve the same problem under different names. This maps them onto one another and sets out which one a given dataset can actually support.
Complaint records tell you where people reported dumping, not where dumping happens, because that depends on who complains and where the city chooses to look. This sets out a shared record so decisions rest on what has been surveyed rather than what has been phoned in.
Illegal dumping clusters tightly inside the part of a city where derelict buildings concentrate. But nothing measured predicted how contaminated any given site would be, and the clustering vanishes when the same data is summarised at the ward scale the city is actually governed by. Records held only at administrative resolution will not show convergence even where it exists.
A new adolescent cohort in a region where environmental monitoring has been sparse, running its own air, sound, soil and water measurement rather than relying on national models. At the first visit a third of participants had obesity. Two findings were not expected: the cohort is metropolitan rather than rural, and its clearest environmental burden is drinking water, where distrust tracks a documented violation record rather than running ahead of it.
Between an instrument download and a published number sit choices that are rarely written down. Testing them on a long indoor air record shows one common choice inflates the headline result by a third while leaving the usual quality checks unchanged. Several instrument faults also leave traces detectable from the data alone.
Some hazards are absent from every official record and completely obvious to anyone who walks past them. Filling that gap produces small datasets, and the research system rewards large ones. This argues a first count should be judged on whether the authors can say where they looked, not on how much they found.
It took me a while to end up on the other side of both lessons. The places with the least environmental data are not the places with the least exposure; they are the places furthest from anyone who would come and measure. I would like to spend part of my career closing that gap where I am from, measuring what we have put into the ground and what it does to the people standing on it.
I grew up in a remote village in Andhra Pradesh, India, where school was a distance away and higher education further still. Getting to either was the hard part, not the studying. My parents got there, later than they should have. The difficulty was never ability. It was access.
In those same classrooms I was taught to call the earth our mother, पृथ्वी or अवनी, Pṛthvī or avanee, before I was taught anything about it. It was said so often that it stopped registering, and it took me years to hear it as anything more than a phrase.
Biotechnology taught me what a measurement actually means. Biomedical engineering taught me how to build the instrument that produces one, and how to prove it is telling the truth.
What that measurement can honestly support once it stops being a reading on a screen and becomes a claim about the health of people who live somewhere.
All six World Marathon Majors before I turn thirty. That is the goal, and I have not started yet, which makes it a goal rather than a plan.
Not only the racing. The strategy, the engineering, the aerodynamics, and most of what decides a race never makes the broadcast.
No particular destination. I work the route out on the way, usually on a bike or in something with a manual gearbox.
Not a serious practice. I carry a camera and stop for whatever the weather is doing, or for the places where a city runs out and something older takes over. Lately more of it is on film, and I am still learning that part.
If any of this interests you, get in touch about research, Formula One, running, trips, or anything you know more about than I do, with ice cream optional but strongly encouraged.