Industries / Energy, water and waste / Environmental services / Environmental remediation
Monitoring
Monitoring samples wells, soil gas and indoor air again and again, during and after a cleanup, to show whether pollution is shrinking, holding steady or moving toward people.
6
Milestones achieved, from 1996 to 2015.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1996
Pump the well slowly
Before a groundwater sample is taken, the water standing in the well is pumped out. EPA's Robert Puls and Michael Barcelona of the University of Michigan wrote in April 1996 that the usual practice, bailing or fast-pumping three to five well volumes, stirred up particles that inflated results for metals and some organic chemicals. They set out low-flow sampling: pumping at about 0.1 to 0.5 liters a minute so the water level barely drops, and sampling once readings such as pH, conductivity and turbidity settle.
LessonThey listed practitioners' "resistance to change" and "concern that new data will indicate a change in conditions and trigger an action" among the method's drawbacks.
Credit
Robert W. Puls and Michael J. Barcelona, for EPA's Regional Superfund Ground Water Forum
1999
Nature's cleanup, under watch
Monitored natural attenuation (MNA) means relying on natural processes, such as bacteria breaking down fuel, to lower pollution over time, while measuring that it happens. EPA's directive of 21 April 1999 set the terms for using it at Superfund, hazardous waste and leaking underground tank sites: it is not a "presumptive" or "default" remedy, not a "walk-away" approach, and source control and long-term performance monitoring are "fundamental components" of it.
LessonEPA expected MNA alone to be appropriate "at relatively few contaminated sites," and for chlorinated solvents to reach cleanup goals in a reasonable time "only in limited circumstances.".
Credit
EPA Office of Solid Waste and Emergency Response
2001
Cleanup data made public
California requires those responsible for a cleanup to send their laboratory results electronically, along with their reports, to the state's GeoTracker database, and the public can reach the whole database of results. A 2022 review counted about 285 million analytical records for more than 50,000 current and former contaminated sites, some dating back to 2001, and said GeoTracker had been the main data source for a number of data-mining studies in the previous ten years.
Credit
California State Water Resources Control Board; review by Beckley, McMasters, Cohen, Cordano, Rauch and McHugh in Groundwater Monitoring & Remediation
2002
Vapors found in homes
Chemicals in groundwater can evaporate, rise through the soil and seep into buildings above, which is called vapor intrusion. In Endicott, New York, solvents were found under IBM's plant in 1979 and pumped out of the ground from the early 1980s, but indoor air in buildings above the plumes was first sampled in the winter of 2002 to 2003, under state oversight. Vapor intrusion was found in buildings on and off the site, and as of March 2024 about 500 ventilation systems had been installed at 452 properties.
LessonWhen testing began there was no national or New York standard for indoor air in homes, so the state offered ventilation wherever TCE showed up in soil vapor.
Credit
IBM, overseen by the New York State Department of Environmental Conservation and Department of Health
2006
Counting the useful bacteria
Some strains of Dehalococcoides bacteria break chlorinated solvents down into harmless products. Xiaoxia Lu, John Wilson and Donald Kampbell, at EPA's lab in Ada, Oklahoma, counted the bacteria's DNA in water from monitoring wells at six solvent plumes using quantitative PCR, a lab method that counts copies of a gene. At three sites where the DNA was found, two breakdown products, cis-dichloroethylene and vinyl chloride, declined at "generally useful" rates of at least 0.3 per year; at two sites where it was not found, they barely declined.
LessonComparing their results with another field study, the authors found that monitoring wells did not sample the bacteria in the aquifer efficiently.
Credit
Xiaoxia Lu (National Research Council fellow), John T. Wilson and Donald H. Kampbell, EPA
2015
A national vapor guide
In June 2015 EPA issued its technical guide for assessing and reducing vapor intrusion, 13 years after a November 2002 draft that drew hundreds of public comments. It asks investigators to collect and weigh several independent lines of evidence and to consider future buildings and uses, not only current ones. On 9 January 2017 EPA added a subsurface intrusion component to its Hazard Ranking System, the scoring used to put sites on the Superfund list, so that vapors entering occupied buildings can now be scored directly.
LessonThe guide stresses multiple lines of evidence most for "no further action" decisions, to reduce the chance of wrongly concluding that vapors pose no risk.
Credit
EPA Office of Solid Waste and Emergency Response
Watch for
Sensors instead of well sampling
Savannah River National Laboratory says its ALTEMIS system, which uses AI and sensors that read water level, pH, electrical conductivity and other basic properties to predict how a groundwater plume moves, has reduced monitoring at the Savannah River Site in South Carolina from quarterly sampling across 2,000 wells to a single annual verification sample. The laboratory's own fact sheet says sensors "will not eliminate groundwater sampling," because EPA and other rules set a minimum number of wells.
Achieved whenregulators approve the reduced sampling in the site's permits, or an independent review publishes the side-by-side comparison with conventional sampling.
Claimed by
Eric Pierce, Savannah River National Laboratory, at the SCSP AI+Expo, as reported by GovCIO Media in June 2026
Open
01
Back-test cleanup forecasts
A test of whether trend models can forecast when groundwater will meet its cleanup goal. Fit each model on a site's monitoring data up to a cutoff year, forecast which wells will be below the goal by a later year, and score the forecasts against what the data later showed, across hundreds of sites.
The briefCredit · unclaimed02
Check the monitoring reports
An agent that takes a site's lab results and water levels, drafts the tables, trend statements and plume conclusions of its routine monitoring report, and is scored against the reports that consultants actually filed.
The briefCredit · unclaimed
Open 01 · the brief
Back-test cleanup forecasts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of whether trend models can forecast when groundwater will meet its cleanup goal. Fit each model on a site's monitoring data up to a cutoff year, forecast which wells will be below the goal by a later year, and score the forecasts against what the data later showed, across hundreds of sites.
Why it is still open
A 2022 Colorado State University thesis set out to estimate time to site closure from historical well data, found no compelling correlations in the real data, "likely due to poor data quality from integrated well samples," and turned to a simulated plume, where monthly rather than quarterly sampling made closure times predictable. A 2024 preprint forecasting the time to reach standards at one site, the Savannah River Site's F-Area, was withdrawn by its author to fix its models.
Built with what exists
GeoTracker's public lab results; Python with statsmodels (which includes the damped Holt method the thesis used), Prophet, and PyLEnM, the open-source long-term monitoring package the preprint extended; a public notebook.
First step
Pick 20 closed fuel-leak sites in GeoTracker with at least ten years of quarterly data and fit a straight-line trend to their first five years.
Done looks like
Forecasts for at least 500 wells at 100 or more sites, with cutoff and test years fixed in advance, and published hit rates for "below goal by the test year" compared with a straight-line trend on log concentrations.
Open 02 · the brief
Check the monitoring reports
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An agent that takes a site's lab results and water levels, drafts the tables, trend statements and plume conclusions of its routine monitoring report, and is scored against the reports that consultants actually filed.
Why it is still open
Software such as Groundwater Assist and ChemGraph has long turned lab files into tables, graphs and contour maps. A March 2026 post by Amber Vault AI claims that AI cuts the time to produce monitoring reports by 40% to 70%, without naming a product or a test. No published test of AI-drafted monitoring reports against filed ones was found.
Built with what exists
GeoTracker, where California sites file both their lab data and their reports; a model built for long documents, such as Muse Ember (see the Note "The jobs agents repeat"); a model that checks its own work, such as Inkling 2 Small (see the Note "Checked without being asked"); an agent that cites the data row behind each statement (see the Note "Findings that cite notes").
First step
Download one site's lab data and its last four filed reports from GeoTracker and rebuild one report's main concentration table with a script.
Done looks like
Drafts for 50 filed reports in which at least 99% of table values match the filed report, and the trend call for each well (rising, falling or stable) agrees with the filed report or a reviewer's check for at least 90% of wells, with every disagreement listed.