A midyear roundup of environmental research papers that stood out from January through June 2022—mostly from ES&T and ES&T Letters, with a few papers from other journals mixed in. The selection leans toward studies that are genuinely useful, methodologically interesting, or simply too provocative not to comment on.
January
Desalination, lake sediments, and exposures you can wear
One paper introduced a 3D engineered “tree” for desalination. By treating wooden rods and arranging them in arrays, the system evaporates groundwater more efficiently than actual trees. That makes it potentially valuable for seawater desalination or drinking-water treatment in remote areas. It also used COMSOL simulation, which is still seen mostly as a chemical engineering tool, but there is no reason environmental research should not be using it much more often for practical problems.
Another study reconstructed more than a century of fossil-fuel emissions in southeastern China using radioactive isotope data from sediments in a volcanic lake near Zhanjiang, Guangdong. The record suggested that fossil-fuel emissions have slowed in recent years, which works as an independent line of evidence alongside more conventional climate indicators.
Artificial light at night was examined through its effects on Daphnia magna. The key point is that artificial lighting does not share the same spectrum as natural light: some wavelengths are missing and intensities differ, often in ways the human eye barely notices. Ecological systems, however, may respond strongly to those differences, and the study found impacts on development, reproduction, and antipredator defenses.
A community transmission study of SARS-CoV-2 used an agent-based model and arrived at a familiar but still important pattern: 70% of infected people did not transmit the virus onward, while 13.4% accounted for 80% of secondary transmission. That makes a strong case for modeling risk at the household and contact-history level rather than relying mainly on daily case counts. People living alone or with limited contacts contribute very differently from people whose work or social habits keep them constantly mobile. If prevention resources are limited, the high-contact nodes matter far more than averages.
A crater lake in the Caribbean was used to challenge a number of claims about global preindustrial lead and mercury emissions that have circulated widely, including in popular science books. Once again, volcanic lakes proved to be remarkably useful archives for studying environmental change across the industrial era.
Bromoform, a highly unstable compound that transforms quickly in the environment, is easy to overlook. One study found that industrial-area emissions were far above natural background. That matters not only because it enlarges the anthropogenic pool of brominated compounds, but also because it can confound research on environmental processes often assumed to be dominated by natural bromoform sources.
A methods paper on effect-directed analysis laid out a full workflow for prioritizing features in suspect and nontarget screening. These platform-building papers often matter more than flashy discoveries. New findings may depend partly on luck, but before luck helps, a lab needs an integrated technical system—and that usually represents years of accumulated know-how and a group’s real core competence.
Microplastics were detected in human sputum, with smoking and intubation suggested as possible sources. An oddly practical implication follows: spitting may be one route by which people physically expel microplastics.
A wearable passive-sampling wristband study in South African children used what was, in essence, three PDMS bars inside a watch-style sampler to characterize the external exposome. That seems especially promising for short-term group activities, where traditional exposure monitoring is difficult. For long-duration sampling, though, the uncertainties multiply quickly, and without a very large sample base the conclusions become harder to trust.
Another paper classified dissolved organic matter using MS1 and MS2 data through unsupervised structural approaches. Conceptually it felt very close to molecular networking approaches such as GNPS. The diplomatic version is “different paths, same destination.” The less diplomatic version is “reinventing the wheel.”
February
Ships, toxicity models, seabirds, and a review-heavy month
A study on Chinese fishing vessels suggested that older small boats may be an important source of PM2.5 emissions. Given that China has on the order of 700,000 to 1 million fishing vessels, about half of them small, that is not a niche emissions issue.
An open QSAR-based tool for predicting acute toxicity, STopTox, offered six toxicity endpoints from either a drawn structure or a SMILES string. The predictions are not yet at the level of commercial software, but the upside is that both data and models are open, which makes the tool far easier to use and evaluate.
Another model integrated chemical use, structure, environmental fate, exposure, and toxicity into a mechanistic framework for estimating emissions and risk. It can swap in different component models, which makes it practical as a fast decision-support tool.
Mercury concentrations in Arctic seabirds showed a U-shaped trend over the last 20 years: down first, then back up. The study linked that reversal to a dietary shift from cod to Atlantic herring. That is a good reminder that contaminant trends in wildlife are often mediated through food-web changes rather than simple changes in emissions.
A systematic review on plastic debris as a vector for bacterial disease pointed out how little work has actually been done on plastic fibers in this context. One amusing detail was how the literature surged after late 2019. It is hard not to suspect that the pandemic helped researchers across fields discover new ways to combine existing topics into suddenly fundable questions.
One paper argued that electricity generation in China has reduced freshwater biodiversity. The problem is that biodiversity was not measured directly but inferred from water consumption and thermal pollution through a modeling framework. When a model defines biodiversity loss from the outset, the conclusion can end up looking stronger than the evidence really is.
A review on selective removal of toxic organic pollutants from complex waters framed the topic as precision removal, though much of it still came down to molecular imprinting and surface modification. Many of the highlighted technologies are not especially selective in practice. More broadly, this period seemed full of reviews and perspectives, enough to raise the cynical question of whether journals were trying to pad their influence metrics.
A paper on chronic environmental disturbance and microbial community assembly was interesting enough, but one visual element stood out for the wrong reason: a figure masquerading as a table, and not even convincingly.
Using NMR and FT-ICR MS, another study showed that marine dissolved organic matter can share thousands of molecular formulae across major water masses while differing structurally. That logic is highly relevant to nontarget analysis too: stop obsessing over the exact identity of single molecules and look instead for broader structural patterns.
March
Policy support, cyclone mortality, 5G zebrafish, and source tracking without identity worship
A policy-oriented paper reviewed the continuing need for scientific support under the Stockholm Convention on Persistent Organic Pollutants. Effective and timely action on chemical pollution depends not only on natural scientists but also on expertise from the social sciences.
Another article argued that hazardous-chemical management remains too narrow and inefficient, and that climate governance offers lessons for addressing the global chemical pollution crisis. The main target was the kind of institutional “lock-in” that keeps systems stuck with weak and fragmented chemical regulation.
A JAMA study estimated excess mortality after tropical cyclones in the United States over a 30-year period. Using excess deaths rather than direct death counts, it found mortality increases as high as 33.4%. That approach is often more informative than official direct death tallies and is especially valuable in a climate context where indirect disaster impacts matter enormously.
There was also a paper combining automated literature searching and natural language processing to identify mechanistic evidence relevant to cancer risk assessment. It is not replacing experts anytime soon. The limitations were discussed openly, and the main challenge is obvious: biological concepts that humans understand intuitively are often difficult to formalize for automated systems. Still, this feels like the sort of limitation that machine learning will eventually grind down.
A highly newsworthy study tested developmental exposure of zebrafish to 3.5 GHz radiofrequency radiation associated with 5G and reported transcriptomic and long-term behavioral effects. It is easy to imagine this being used as ammunition against 5G more broadly. But with 5G already deployed extensively in many cities, epidemiological data should in principle be obtainable; without population-level support, small-effect animal studies are easy to overinterpret. The visualization choices and p-value usage in the paper also invited criticism.
Another paper on air pollution, traffic, biomarkers, and asthma had the feel of a statistician-led environmental health study: strong on macro indicators, weaker on mechanism. Biomarkers were used mainly as instrumental variables, with little biological discussion. This keeps happening because the people who understand the epidemiological models are often not the ones discussing the mechanisms, and the people who know the biology may not follow the statistical framework. The result is parallel work on the same scientific problem with surprisingly little cross-talk.
A source-tracking paper used machine learning on quantitative chemical fingerprints from nontarget data instead of insisting on full qualitative identification first. That is a sensible direction. Similar logic has long been used in food origin studies. If the real goal is source attribution, exhaustive compound identification may be much less essential than chemists often feel comfortable admitting.
Another study noted an interesting pandemic-era traffic effect: in 2020, there were fewer vehicles on the road in some places, but they moved faster. When traffic is dense, stop-and-go driving dominates, and that matters for CO2 emissions. It is a useful perspective, though the growing share of electric and hybrid vehicles means those models will need updating.
A warming experiment on a globally important diatom tracked both short-term and long-term responses, including proteomic changes, over 700 generations. Metabolism was disrupted in the short term, but the population eventually adapted to higher temperatures. Climate change does not mean Earth itself needs rescuing; it means organisms and societies adapted to current conditions may find themselves in serious trouble.
April
Wastewater sildenafil, metaverse biodiversity, human VOCs, and nanoscale plastic hype
A wastewater-based epidemiology study covering 33 cities found the highest sildenafil concentrations in wastewater from northwestern China. Sildenafil levels were positively associated with tobacco and alcohol use, as well as regional age structure, economic level, marriage proportion, and employment rate. Whatever else one says about it, it was a very memorable WBE paper.
A daily PM2.5 personal exposure assessment paper used random forests in a way that managed to capture two recurring habits outside statistics and computer science: treating random forests as an impressive black box anyone can use by calling a package, and then evaluating a nonlinear model with linear-model reflexes like R². Since random forests can produce negative R² values, that combination can become awkward fast. The study ended up inheriting both poor interpretability and a somewhat doctrinaire attachment to R².
A perspective on the metaverse and biodiversity discussed opportunities and risks, but in practice much of the appeal came down to immersive virtual reality experiences that might help people engage with species diversity. The argument was unusual enough to be admirable on sheer angle alone.
A study of Chinese lakes linked dissolved organic matter composition with carbon emissions and pollutant emissions to disentangle natural and anthropogenic drivers of lake chemodiversity. That broader idea is genuinely stimulating: pollutant emissions may leave detectable signatures not just in contaminants themselves but in shifts to the organic-matter matrix of environmental media.
Human VOC emission rates were measured in relation to ozone, clothing, and temperature. The experimental design had a slightly psychological-lab flavor, but the central finding was straightforward: what people emit chemically depends in part on the microenvironment around them.
Another indoor study modeled clothing as a transport vector for airborne particles and pathogens, including SARS-CoV-2 on cotton fabrics. It concluded that secondary transmission by inhalation of particles resuspended from clothing was likely low, with viral particles largely inactivated after about a day.
Flight trajectory data were used to build a high-resolution emission inventory for China’s aviation sector. This is exactly the kind of thing modern sensor and tracking infrastructure makes possible, and environmental science still has plenty of room to do more with such real-world high-resolution data streams.
A concise review on environmental DNA laid out different physical states of eDNA in aquatic systems and examined what is actually known about persistence. Much of the literature, it turned out, does not discuss these parameters carefully enough. Short paper, high information density.
Then there was the headline-grabbing study showing that common single-use plastic products release trillions of sub-100 nm particles per liter into water during normal use. The title practically writes itself. One could joke that if we are defining nanoparticles broadly enough, water molecules are also doing pretty well in the under-100 nm category.
May
Causal inference, Gulf War genetics, asbestos in one page, and PFAS labels that tell on themselves
One article made a strong case for bringing genetic data into exposomics for causal inference. That is exactly the right direction. Environmental analysis cannot ignore genetic background, and methods developed in GWAS can inform EWAS-style research on how much exposures contribute to disease.
A study of Gulf War illness revisited gene–environment interaction three decades after the conflict, focusing on PON1 and low-level nerve agent exposure. This research area has always been difficult because only some individuals appeared to show strong associations, prompting the hunt for relevant genetic loci. It also became a classic case for discussing analytic problems such as recall bias. A major strength of this paper was that much of its effort went into addressing those longstanding criticisms.
One of the sharpest brief papers of the period examined asbestos through a remarkably compact presentation—essentially a page, a figure, and a table. The policy message was clear: banning asbestos is typically cheaper in the long run than trying to control exposure indefinitely, assuming adequate substitutes exist. In an aging world where public health costs can become bottomless, prevention usually beats chronic management.
A study on valved N95 respirators tested how well they filtered exhaled particles during speech. Two models performed about as well as surgical masks, and one barely filtered exhaled particles at all. That makes sense: valved respirators are designed primarily for wearer comfort and protection from incoming pollution. If the wearer is the source, the valve can become the whole problem.
A darkly funny paper looked at labels on children’s and adolescents’ products—especially “waterproof” and “green” claims—and compared them with the presence of PFAS. Waterproof labels were strongly associated with PFAS, while green labels told you almost nothing. So if your goal is to reduce PFAS exposure, one practical rule is to avoid waterproof claims. The irony, of course, is that parents often choose waterproof textiles, furniture, or carpets precisely because children use them and they are easier to clean.
Another paper described crack patterns on environmental plastic fragments and how those patterns vary with composition. For now it is largely observational, but if the morphological signal is real, image-based classification with a convolutional neural network would be a natural next step.
A U.S. cohort paper from the ECHO program measured contemporary and emerging chemicals among pregnant women. The bigger lesson is not just the chemical data but the infrastructure behind it: long-term, high-quality cohort platforms are exactly what environmental health research needs more of. Such datasets are expensive and messy to build, but they can be reused repeatedly to test new analysis methods and mine new questions. Comparable examples like UK Biobank show how valuable this kind of national-scale data resource can be.
A dietary intervention study linked hot soup packaged in plastic bags with phthalate body burden and inflammatory mRNA expression. The self-controlled design was a reasonable choice, though the data analysis then drifted into differences between two universities in a way that felt oddly beside the point.
Finally, a paper proposing that a geomagnetic pole reversal 42,000 years ago triggered a global environmental crisis quickly attracted two comments challenging it, followed by a response from the authors. The exchange was unusually direct and would make an excellent teaching case for graduate students learning how scientific argument actually works.
June
PFAS and mortality, comments worth reading, Indian EVs, yeast for lead, and the politics of negative results
A population-based cohort study using NHANES 1999–2014 provided direct evidence linking PFAS exposure with all-cause mortality in U.S. adults. Methodologically, it used unsupervised k-means clustering with the elbow method to define low, medium, and high mixed-exposure groups, then calculated group mean concentrations for each PFAS and proceeded with standard epidemiologic survival analyses. That is a clever workaround for the persistent difficulty of defining “overall” exposure levels in mixtures, and the strategy could be extended to other pollutant mixtures as well. More broadly, the paper was also a reminder that when public data are truly open, researchers anywhere will use them—and the country that shares the data often gets the largest policy benefit from the resulting evidence.
A response paper tied to the Gulf War gene–environment debate followed criticism centered on ancestry. For students who rarely get to review manuscripts, reading comments and replies may be the next best thing. Conclusions alone are engineering; the back-and-forth is where scientific judgment gets sharpened.
A life-cycle assessment of vehicle electrification in India found that a shift toward battery electric four-wheel vehicles could actually increase greenhouse gas emissions and sulfur dioxide emissions under current conditions, largely because of the country’s present energy mix and infrastructure constraints. Electrification is not a magic word; context matters.
A Communications Earth & Environment paper showed that inactive yeast cells can remove trace concentrations of lead from water. One charming detail was how carefully the methods began with bottle washing. Sometimes craftsmanship is the method.
Another paper argued that reducing beef and increasing whole grains in U.S. school lunches could greatly reduce environmental impacts. At first glance that may sound oddly specific, but the context is important: school lunch waste in the United States is substantial, so shifting menus in a more environmentally favorable direction is hardly unreasonable.
A randomized controlled trial on portable HEPA air cleaners during pregnancy reported that prenatal exposure reduction may affect children’s cognitive performance at age four. Whenever authors rely heavily on “may,” it is often wise to check the p-values, and in this case the results were indeed not conventionally significant. Writing up negative or borderline findings is an art form of its own.
Another study inferred from hydrocarbon tracers that methane emissions from the oil and gas sector occur mainly before gas processing rather than during later production stages, and that petroleum plays are a significant source. That has obvious implications for where mitigation efforts should focus.
A paper using the updated WHO air quality guidelines highlighted something almost absurd but still important: in some places, non-anthropogenic PM2.5 alone can exceed the guideline level. Put differently, if standards become strict enough, parts of Earth start failing them even without human emissions. That is not an argument against ambitious standards, but it does show how guideline interpretation can get complicated once natural sources dominate.
Finally, a study on phytoplankton responses to ocean warming suggested that after adaptation, their nutritional value may decline. That kind of shift feels remote until one remembers how much of the marine food web rests on plankton quality as well as quantity. The consequences could travel far beyond the organisms being measured.