California went looking for oil in highway runoff and could not find enough of it to treat. Twenty-two maintenance stations were screened for an oil-water separator retrofit, ten were carried forward as the likeliest candidates, four were monitored, and only one looked promising enough to build. Even there, the concentrations of free oil measured during the study were too low for the device to work. The final report states the outcome for the whole set: none of the 25 maintenance stations evaluated across the Los Angeles and San Diego districts had free oil concentrations high enough for successful implementation of the technology.
That result, published in 2004 in the Caltrans BMP Retrofit Pilot Program final report, is a useful correction to the mental image of road pollution. The visible film on wet pavement is not where the load sits. What comes off a highway in a storm is mostly fine particles, the metals attached to them, and a set of dissolved constituents that behave nothing like each other.
What the runoff was actually carrying
The Caltrans program was built under a litigation settlement with the Natural Resources Defense Council, Santa Monica BayKeeper, San Diego BayKeeper and the Environmental Protection Agency, and it retrofitted a wide range of treatment devices into an existing freeway system rather than designing them into new construction. Because it monitored influent as well as effluent, it published something most performance studies do not: the average untreated quality of the runoff reaching its sites. For highway and maintenance station catchments during the study, total suspended solids averaged 114 milligrams per liter, total phosphorus 0.38 milligrams per liter and total zinc 355 micrograms per liter.
Zinc is the number to sit with, and the report treats it that way, carrying total zinc alongside solids and phosphorus through every device comparison it makes. Part of that zinc arrives dissolved rather than attached to a particle, which is the harder half of the problem, and the program’s most interesting chemistry finding sits there. Dissolved copper, which sand filters were not expected to reduce substantially, was reduced significantly once influent concentrations exceeded about 15 micrograms per liter. The authors attribute that either to adsorption on sand grains and accumulated sediment or to the dissolved and particulate phases not yet being in equilibrium when runoff enters the sedimentation basin, since rainfall arrives at very low pH and is neutralized rapidly during runoff.
Chloride is the constituent that does not wash out
Deicing salt behaves in a way no treatment device on a roadside addresses. A U.S. Geological Survey analysis published in 2015 examined 30 monitoring sites on 19 streams with chloride and flow records of 18 to 49 years, and found that average concentrations approximately doubled from 1990 to 2011, outpacing the rate of urbanization in the northern United States. Concentrations rose in all seasons at 13 of the 19 streams and in winter at 16 of them.
The seasonal detail is the mechanism. Increases during non-deicing periods led the authors to conclude that chloride was being stored in hydrologic reservoirs such as the shallow groundwater system during winter and released slowly in baseflow through the rest of the year. Salt applied in January is still arriving in the stream in July.
For regulatory consequence, the paper reports that 29 percent of the sites studied exceeded the Environmental Protection Agency chronic water quality criterion of 230 milligrams per liter, a four-day average value, by an average of more than 100 individual days per year during 2006 to 2011. The acute criterion, a one-hour average, is 860 milligrams per liter. The paper also traces the driver through the commodity data: road salt sales rose from an annual average of 9.6 million metric tons in the 1980s to 19.5 million metric tons in the last five years reported, ending in 2011.
The tire chemical that reordered the research agenda
6PPD-quinone is a tire-rubber-derived chemical, and its identification in 2021 as the cause of acute toxicity in coho salmon changed what a stormwater program has to worry about. The concentration that kills half an exposed group of coho sits below a thousandth of the zinc concentration measured in that untreated Caltrans runoff.
Its toxicity values need stating carefully, because they were revised. As reported in a 2022 paper in Environmental Toxicology and Chemistry, the original 2021 identification gave a median lethal concentration for coho salmon of 0.79 micrograms per liter; that value “was slightly later found to be substantially lower (0.095 µg/L) as a result of improved analytical methods and the availability of commercial reference standards of 6PPD-quinone.” Two LC50 values for coho are therefore in circulation, an order of magnitude apart, and a citation that names neither is unusable.
The same paper is the reason a highway agency cannot simply generalize the risk. Its authors exposed alevins of Atlantic salmon and brown trout in eight treatments at initial concentrations from 0.095 to 12.16 micrograms per liter for 48 hours and recorded no mortalities and no substantial behavioral changes in either species. A compound lethal to one salmonid at sub-microgram concentrations left two close relatives unaffected across a 128-fold exposure range. That pattern, closely related species running from extreme sensitivity to no measurable effect, makes a single national treatment target close to meaningless and pushes the question down to which receiving water a particular outfall discharges into.
The permit that makes a highway department a municipality
Highway runoff is regulated under the same instrument as city storm drains. An MS4, in the Environmental Protection Agency’s definition, is a conveyance or system of conveyances owned by a public entity, discharging to waters of the United States, designed or used to collect or convey stormwater, that is neither a combined sewer nor part of a treatment works. The 1990 Phase I rule brought in medium and large cities and certain counties with populations of 100,000 or more, and there are approximately 855 Phase I systems covered by 250 individual permits. The 1999 Phase II rule reached small systems in Census-defined urbanized areas, and, in the agency’s words, “also includes non-traditional MS4s such as public universities, departments of transportation, hospitals and prisons.” Most of the 6,695 Phase II systems are covered by statewide general permits.
A statewide general permit covering a highway department is doing something no municipal permit has to do. It applies one set of obligations to a linear network that crosses hundreds of receiving waters with different impairments, different criteria and, since 2021, different sensitivity to a compound nobody was measuring when the permit was written. No single design storm and no single treatment standard satisfies that set, which is why the conveyance and inlet sizing covered by how highway drainage systems work now sits inside the same argument as treatment performance. It is also why so much of the roads and the environment coverage here turns on which particular water a given outfall reaches.
Which devices earned their place
The most useful output of the Caltrans program is its ranking of treatment devices, and the methodological correction behind that ranking matters as much as the ranking itself. Reporting performance as a percent reduction in influent concentration misrepresented the relative merit of the devices, because influent quality differed between sites. The clearest case was the multi-chambered treatment train, installed at park-and-ride lots where untreated runoff was already clean, which scored a low removal efficiency while producing effluent as good as the best device in the study. The team therefore used regression to predict what each device would produce if all of them received identical influent.
On that basis, and against the influent averages above, the Austin sand filter produced expected effluent of 7.8 milligrams per liter of total suspended solids and 50 micrograms per liter of total zinc. The biofiltration strip produced 27.6 and 79. The unlined extended detention basin produced 36.1 and 139. Infiltration basins and trenches were scored at zero for both, because they discharge nothing to surface water. At the other end, one proprietary media filter produced 78.4 milligrams per liter of solids and 333 micrograms per liter of zinc against a 355 microgram influent, which is to say it did very little to the metal at all.
Biofiltration swales and strips were among the least expensive devices in the study and among the best performers for sediment and heavy metals, and they need no specialized maintenance. Their weakness was nutrients: phosphorus removal was worse than expected because the salt grass planted at the pilot sites leached phosphorus, a condition the team did not know about at the start. The lesson the report draws is specific and practical, which is to specify a mixture of drought tolerant native grasses rather than a monoculture, and to time establishment to the wet season.
One operational finding outranks all of the water chemistry for anyone who has to live with these assets. Mosquito breeding was observed at every site where standing water persisted for at least 72 hours, which covers not only the devices designed with a permanent pool but also stilling basins, riprap used for energy dissipation, flow spreaders and some outlet structures. Many of the pilot devices had to be modified during the study to drain.
The asset nobody designed as treatment
Infiltration devices scored best on surface water and left the study’s most important question open. Substantial load reduction in the unlined basins, swales and strips came from runoff soaking into the ground, and the part of the program meant to establish whether that threatens groundwater quality was unsuccessful, so the authors called for further investigation rather than declaring the practice safe. The same unresolved tradeoff runs through permeable pavement, and it is the strongest argument for treating infiltration as a site-specific question rather than a policy preference.
Federal practice guidance on this subject was overdue enough that Congress legislated its renewal. Section 11521 of Public Law 117-58 named two documents, the 1997 report Stormwater Best Management Practices in an Ultra-Urban Setting and a 2014 report on data collection and performance measurement, and directed FHWA to update and reissue each within a year and at least every five years thereafter. The finding buried in the Caltrans conclusions is the one still worth chasing. The pilot swales resembled the vegetated shoulders and conveyance channels already lining highways across the state, and the authors suggest those existing, unengineered surfaces could be expected to offer comparable water quality benefit, with more research needed to establish it. Two decades later, the largest treatment system on the network may be the one that was built for drainage and has never been credited for anything else.