CHECKED AGAINST FILED REPORTS
How to read the statistics appendix in a landfill groundwater report
13 November 2026 · about an eight minute read
Many semiannual groundwater reports for California landfills carry an appendix of prediction limit tables printed by Sanitas, a Windows statistics program widely used for this work. Each row compares one new result at one well with a limit computed from background, the earlier results that represent water the landfill has not affected. This article explains every column and every method code in those tables, using real rows from two reports anyone can download.
It covers municipal solid waste landfills under RCRA Subtitle D, the part of the Resource Conservation and Recovery Act whose groundwater rules are in 40 CFR Part 258, Subpart E. Both example reports are from California, where state rules and each facility's waste discharge requirements add to the federal rule. Your facility's program may differ, and its permit decides the retest rule.
The two reports are:
- Badlands Sanitary Landfill, Riverside County. Report filed on GeoTracker on 30 April 2026, Appendix L: intrawell limits for the February 2026 sampling event. Intrawell means each well is compared with its own history.
- Visalia Landfill, Tulare County. First Semi-Annual 2026 report, Appendix C-5, printed by Sanitas 9.6.37. It runs in three stages, described below.
The columns
- Constituent and Well. What was measured, and where. One row is one constituent at one well.
- Upper Lim. The prediction limit: the highest value a new result can take and still be consistent with background, at the chosen false positive rate. At Badlands, boron at well BH-25 has an upper limit of 0.595 mg/L.
- Lower Lim. Filled in only where a drop matters as much as a rise, which in practice means pH. At Badlands, pH at BH-11 has an upper limit of 8.56 and a lower limit of 7.14. Leachate can push pH either way, so both directions are tested.
- Date and Observ. The sampling event and the result being tested. A value such as
1NDis a nondetect: the laboratory did not detect the constituent, and 1 is the reporting limit. - Sig. Yes if the observation is above the upper limit or below the lower one. A Yes is a potential increase, which the retest rule then confirms or clears.
- Bg N. How many background readings the limit was computed from. This number governs almost everything else in the row.
- Bg Mean and Std. Dev. The average and spread of the background readings. When the Transform column names a transformation, these are computed on the transformed values, so do not compare them directly with the limit.
- %NDs. The share of background readings that were nondetects.
- ND Adj. How nondetects in background were handled before the arithmetic, for example by substituting a value, or not at all.
- Transform. Any transformation applied so the background looks closer to a normal distribution: natural log, square root, cube root, or none. The limit is computed on the transformed scale and converted back to the reported units. At Badlands, nitrate as N at BH-22 uses a square root, with 32 background readings, and a limit of 1.25 mg/L.
- Alpha. The false positive rate for that row: the chance the test flags a well that has not changed.
- Method. Which kind of limit was computed. The codes follow.
The method codes
Every code says two things: parametric or nonparametric, and intrawell or interwell. Interwell means the well is compared with background pooled from other wells, usually upgradient ones.
- Param Intra and Param Inter. A parametric limit: the background mean plus a multiple of the standard deviation, computed on the assumption that background, perhaps after a transformation, follows a normal distribution. EPA's Unified Guidance gives the basic form in chapter 18 (equation 18.3) and the multiples for retest plans in chapter 19. Badlands boron at BH-25 is Param Intra with only 11 background readings.
- NP Intra (normality) and NP Inter (normality). A nonparametric limit, chosen because background failed the test for normality even after the transformations were tried. The limit is one of the background readings themselves, usually the highest. Badlands chloride at BL-03 is NP Intra (normality), with 84 background readings and a limit of 163 mg/L.
- NP Intra (NDs) and NP Inter (NDs). A nonparametric limit chosen because too much of the background is nondetect to estimate a mean and standard deviation. Chapter 15 of the guidance recommends nonparametric methods when nondetects are a large share of the data. Badlands nitrate as N at BL-03 has 55.42% nondetects in 83 readings, a limit of 0.300 mg/L, and the code NP Intra (NDs).
- Param Intra Deseas. A parametric intrawell limit with the seasonal pattern taken out. When a constituent rises and falls with the seasons, the pattern is estimated from background and removed, and the new observation is adjusted for the season in the same way before it is compared. Chapter 14 of the guidance covers this. That is why the Observ. column can show a value that is not in the laboratory report.
Why a nonparametric alpha is not 0.01
For a parametric limit the alpha is chosen, and the multiple is computed to match it. For a nonparametric limit it is the other way round. The limit is a background reading, so the only thing that can change the false positive rate is how many readings there are and which one is used. Section 18.3 of the guidance explains this: the confidence of a nonparametric limit follows from n and the retest plan, and cannot be set freely.
The Badlands rows show it. If the limit is the highest of n background readings, the chance that one new reading from the same water lands above it is 1 in n + 1:
- Chloride at BL-03, 84 background readings, with a 1-of-2 retest plan: alpha 0.01176, which is 1/85.
- Nitrate at BH-11, 32 background readings: alpha 0.0303, which is 1/33. The limit is 32.0 mg/L, and the code is NP Intra (normality).
The 1-of-2 plan means a first exceedance is followed by one resample, and an increase is confirmed only if the resample also exceeds. The resample is what keeps the overall false positive rate acceptable when one reading alone carries 3 percent. A small n gives a large alpha, which is one reason the guidance asks for more background over time.
The false positive rate line
Some appendices add a line under the table giving the false positive rate across all of its rows. One well and one constituent is a single test, but a site runs dozens or hundreds of them each event, and some will exceed by chance. The guidance recommends a site-wide false positive rate of 10 percent a year across all tests, and chapter 19 explains how retesting reaches that target while each test keeps enough power to catch a real release. The line tells a reviewer what rate the per-row alphas were chosen to add up to.
Reading the Visalia stages
The Visalia appendix runs the same data through three stages, and the same well can appear in each.
- Stage 1, interwell. Background pooled from 11 wells since July 2015, 51 readings. Bicarbonate at M-2B has a limit of 380 mg/L and an observation of 640, so Sig. is Yes. The code is NP Inter (normality), and the alpha is 0.01837, again fixed by the background count and the plan rather than chosen. Carbonate in the same stage has a limit of 1 and an observation of 1ND, with 100% nondetects and the code NP Inter (NDs). When every background reading is a nondetect, the limit is the reporting limit, and a new nondetect cannot exceed it.
- Stage 2, all background history pooled. Bicarbonate at M-2B now has a limit of 848.7, and the observation adjusted for the season is 638.7, so it no longer exceeds.
- Stage 3, intrawell. Potassium at M-12 uses a cube root and the code Param Intra Deseas. The limit is 12.06. The laboratory reported 4.8, and the observation adjusted for the season, as the table prints it, is 4.208.
How Koyna handles these tables
Koyna prints prediction limit tables with the same columns and the same method codes, so an engineer or reviewer who knows the Sanitas appendix can read Koyna's without learning a new layout. We ran Koyna on two filed reports, and its numbers agree with the Sanitas tables: all 31 of 31 limits in the Badlands appendix and all 79 of 79 rows in the Visalia appendix. The comparison is written up in Koyna against filed reports.
If something here does not match how you read or prepare these appendices, or a column means something different in your reports, we would like to hear it. Write to support@koynasystems.com.
Sources
- US EPA, Statistical Analysis of Groundwater Monitoring Data at RCRA Facilities: Unified Guidance, EPA 530-R-09-007, March 2009. Chapter 14 on seasonality; chapter 15 on nondetects; chapter 18 on prediction limits, including equation 18.3 for the parametric limit and section 18.3 for nonparametric limits; chapter 19 on retesting and the site-wide false positive rate. PDF
- 40 CFR § 258.53, which lists the statistical methods allowed, prediction intervals among them, and the standards they must meet. eCFR
- Badlands Sanitary Landfill, Riverside County, groundwater monitoring report filed 30 April 2026, Appendix L; and Visalia Landfill, Tulare County, First Semi-Annual 2026 report, Appendix C-5. Both are public on the State Water Resources Control Board's GeoTracker. GeoTracker
Sanitas is a trademark of its owner. Koyna Systems is not affiliated with Sanitas or its owner, and nothing here is endorsed by them.