CHECKED AGAINST FILED REPORTS
What it takes to rebuild a landfill's prediction limits from public data
13 November 2026 · about a six minute read
If background is a new term, start with what "background" means in landfill groundwater monitoring.
A filed groundwater monitoring report is a public record, and so, in California, is much of the laboratory data behind it. That raises a fair question for anyone checking a statistics program: can you take the public data, run the same method the report describes, and get the same numbers? We tried it on one landfill's report. The first attempt matched 3 limits out of 25. The second matched all 31. This article explains what changed between them, because the reason applies to almost any site.
The landfill is a municipal solid waste landfill regulated under RCRA Subtitle D, the part of the Resource Conservation and Recovery Act that governs those landfills, with the federal groundwater rules in 40 CFR Part 258, Subpart E. California adds its own requirements, and each facility's waste discharge requirements set details such as the retest rule. Another facility's program may differ, so treat the method below as this site's, not as a rule.
The report we checked
Riverside County Department of Waste Resources filed the semi-annual groundwater report for Badlands Sanitary Landfill on 30 April 2026. It is on GeoTracker, the State Water Resources Control Board's public database. Appendix L of the report holds 31 intrawell prediction limits for the sampling event of 5 and 10 February 2026. A prediction limit is the highest value a new result can take before it counts as a potential increase. Intrawell means each well is compared with its own earlier results, rather than with upgradient wells.
The limits were made with Sanitas, a widely used groundwater statistics program. They cover boron, chloride, nitrate as N, pH (an upper and a lower limit), potassium, sulfate and total dissolved solids, at five wells: BL-03, BH-11, BD-04, BH-22 and BH-25. The report states the method:
- Each well is compared with its own history.
- A normality test decides whether a limit is parametric, which assumes the data follow a normal distribution, possibly after a transformation such as a square root, or nonparametric, which makes no such assumption.
- Parametric limits use a significance level, alpha, of 0.01.
- A 1-of-2 retest applies: a result above its limit is confirmed only if a second sample is also above it.
EPA's Unified Guidance, the agency's guidance on the statistics these rules require, describes this family of methods in chapters 18 and 19.
First attempt: the public lab data
GeoTracker publishes laboratory results county by county as downloadable files. We loaded the file for this landfill's county and took every reading before the February 2026 event as background, the history each new result is compared with.
Only 3 of 25 comparable limits matched. The six pH limits could not be compared at all: pH is measured in the field, and GeoTracker's file holds laboratory results. In almost every other row, the number of background readings was different from the number the county printed.
Why the public data did not match
Nothing was wrong with the county's limits. The public file simply does not hold the same background the county used.
- It starts later. GeoTracker's data for this site begins in 2005. The county's background for some pairs of well and substance reaches back to 1988 to 1990, and some backgrounds hold 71 to 84 readings.
- Dates differ or are missing. Some sampling dates on GeoTracker are a day away from the county's, and some dates the county used are not in the file.
- The county screened its background. The report says the backgrounds were checked for outliers, trends and seasonal effects, and then fixed, last updated in the third quarter of 2024. Each pair has its own background window. Boron at BH-25, for example, starts on 5 May 2022.
- A screened-out reading is still in the public file. Nitrate as N at BH-22 was 21 mg/L on 8 July 2014. The county removed it from background. With it left in, the limit comes out at 21 rather than 1.25.
One reading, correctly set aside by the people who know the site, moves that limit from 1.25 to 21 mg/L. No method choice can recover that from the public data alone. It is a judgment recorded somewhere other than the lab file.
Second attempt: the background printed in the report
Appendix L prints the background readings behind each limit, 1,363 readings in all. We loaded those exactly as printed and ran the same method.
All 31 of 31 limits agree with the county's to the precision the report prints. Each one also has the same form, parametric or nonparametric, the same transformation (a square root for nitrate at BH-22, for example), and the same number of background readings. Where the report gives the share of background below the detection limit, that agrees too: 55.42% for nitrate at BL-03. Some examples, county first and Koyna second:
- Boron at BH-25: 0.595 and 0.5954
- Chloride at BH-22: 463 and 462.5
- Nitrate as N at BH-22: 1.25 and 1.247
- Total dissolved solids at BD-04: 476 and 475.9
The full table, row by row, is on our page comparing Koyna against filed reports.
What this says about any site
A filed limit can be rebuilt from public data only when its background can be rebuilt. The arithmetic of a prediction limit is short. The background is where the work, and the judgment, sits. The usual reasons the background cannot be rebuilt from a public lab file are:
- the public lab data starts later than the facility's own records;
- years are missing from the public file;
- background readings were screened or fixed years earlier, and the report does not print them;
- the limits were set in an earlier year and held fixed since;
- field measurements, such as pH and specific conductance, are not in the lab file at all.
So the best source for checking a limit is the facility's own data file. The next best is a report that prints its background readings, as this one does. A public lab file on its own is a good place to see recent results, and a poor place to reproduce a limit set from decades of screened history.
Reading GeoTracker's file as it is written
One practical note for anyone who tries this. Koyna reads GeoTracker's county file in the form GeoTracker writes it: values such as ".5" without a leading zero, the yes or no column that says whether a substance was not detected, and GeoTracker's substance codes, which Koyna turns into names using GeoTracker's own list of valid values.
What Koyna does today
Koyna loads a facility's whole history from a CSV or Excel file, with a step where you match the file's columns to what Koyna expects. It computes intrawell and interwell prediction limits, with the form chosen by a normality test, and prints method codes in the style Sanitas uses, such as "NP Intra (normality)". Each result keeps the background it was computed from, the method version and the software version, so it can be checked again later against the same inputs.
If you have rebuilt limits from public data and found something different, or if this does not match how background is kept at your facilities, we would like to hear it. Write to support@koynasystems.com.
Sources
- Riverside County Department of Waste Resources, semi-annual groundwater monitoring report for Badlands Sanitary Landfill, filed 30 April 2026, Appendix L. GeoTracker global id L10003146754. GeoTracker
- GeoTracker laboratory data downloads by county. GeoTracker
- 40 CFR Part 258, Subpart E, the groundwater monitoring rules for municipal solid waste landfills. eCFR
- US EPA, Statistical Analysis of Groundwater Monitoring Data at RCRA Facilities: Unified Guidance, EPA 530-R-09-007, March 2009. Chapter 5 on background, chapter 18 on prediction limits and chapter 19 on retesting. PDF
Sanitas is a trademark of its owner. Koyna Systems is not affiliated with Sanitas or its owner, and nothing here is endorsed by them.