In 2022, the U.S. Geological Survey (USGS) National Water Quality Laboratory (NWQL) experienced a large sample backlog that resulted in many nutrient, alkalinity, and acid-neutralizing capacity (ANC) samples not being analyzed within their holding times as required by the NWQL. This issue persisted through May 2024. Chemical properties and constituents, otherwise referred to as observed properties, for which a large number of holding-time exceedances were observed, are shown in table 1. Nearly 80,000 individual analyses were completed on samples after the recognized NWQL method holding time. This affected numerous USGS studies. Many of these specific observed properties are used in the computation of total nitrogen in water and are often associated with analyses done for regulatory purposes.

The U.S. Environmental Protection Agency (EPA) defines sample holding time as the period of time a sample may be stored before analysis (EPA, 2002). According to the EPA, although exceeding the holding time does not necessarily negate the veracity of analytical results, exceeding the holding time requires any data not meeting all the specified acceptance criteria be qualified or flagged. The EPA sets holding-time and preservation requirements for nonpotable waters to ensure that analyte concentrations do not change significantly between sampling and the time of analysis. Specific sample holding-time and sample preservation requirements are provided under 40 CFR part 136 as authorized by section 304(h) of the Clean Water Act (33 U.S.C. 1251 et seq.). The effects of sample storage duration on sample integrity and resulting data could vary and depend on many variables, including inherent microbial activity, sample matrix (for example, saline compared to freshwater), whether the samples were filtered or unfiltered, sample preservation (Degobbis, 1973; Klingaman and Nelson, 1976; Gardolinski and others, 2001), and possible redox conditions (Tate and others, 1995).

As the issue of samples being analyzed after their elapsed holding times came into focus, USGS scientists from around the country formed a study group and met to discuss what was needed to evaluate these sample results. The meetings began in the spring of 2023 to discuss ways to evaluate data from samples analyzed after NWQL method holding times. To understand the effects more fully on data quality from samples being analyzed after their listed holding times, a holding-time study (HTS) was designed to evaluate the potential changes in sample quality from a variety of water matrices and environmental conditions.

This report presents the design and results of the HTS done in 2024. The USGS collected the samples and analyzed the data. The results of previous investigations are summarized in this report to provide background and context for understanding the HTS results. This report also provides an evaluation of the effects of extended holding times on nutrient concentrations in environmental water samples and observations for consideration when using data pertaining to such samples in interpretive studies.

The results of the HTS pertain specifically to samples collected by the USGS and analyzed at the NWQL. Custom NWQL laboratory schedule 3077 was used for the study. The evaluated observed properties are listed in table 1.

Wide application of the HTS results outside of the USGS should be done with an understanding of the study’s limitations. The results may not necessarily be applicable to samples collected, handled, and analyzed by methods that differ from those of the NWQL. Although the HTS was designed to replicate actual field and laboratory conditions imposed upon the affected environmental samples collected, the HTS samples were not necessarily handled the same as routine environmental samples. This was because the HTS samples were part of a large study and required some handling logistics that varied from that of routine environmental samples.

The HTS was designed to evaluate how the results of specific observed properties from a water sample varied during a period of approximately 6 months after the time of collection. The observed properties selected for the HTS were based on the high frequency of holding-time exceedances for those parameters. These observed properties and the associated NWQL methods are shown in table 1.

The sampling sites selected for the HTS were chosen to replicate sample types (matrices), as closely as possible, to the sample types that were analyzed outside the NWQL-method holding time at the NWQL. The HTS included sites that were distributed geographically, had long-term monitoring records, had data affected by the holding-time exceedances, and included groundwater and surface-water sites. Also, the study group recommended inclusion of sites that were part of ongoing monitoring networks so that samples could be collected as part of ongoing sampling activities. Potential sampling locations were identified after reviewing historical concentrations of nutrients and alkalinity (the observed properties affected by the holding-time exceedances) as well as dissolved oxygen, dissolved organic carbon, specific conductance, and other characteristics that possibly could affect sample stability. In addition, seasonality was considered, but time constraints prevented incorporating this variable into the study. Based on available resources, ultimately six USGS sites from a wide geographic distribution were selected and sampled in November and December 2023. The locations of the sites sampled for this study are shown in figure 1. Average concentrations and their subjective categorical classifications (low, moderate, and high) for the six sites are shown in table 2 (U.S. Geological Survey, 2024).

Seven time intervals were initially chosen for analysis of the HTS test samples. Time intervals include three intervals within the NWQL holding time to look for early sample instability, and four intervals outside the NWQL method holding time to investigate long-term sample stability. The time intervals for the study were 48 hours (2 days; referred to as “day 2” in the report), 2 weeks (14 days), 4 weeks (30 days), 8 weeks (60 days), 12 weeks (90 days), 16 weeks (120 days), and 20 weeks (150 days) after sample collection. Due to laboratory logistics, the intervals were not always exactly met. An additional eighth time interval at 26 weeks (180 days) was added by using an additional set of sample bottles submitted for most sites. No spare sample bottles were submitted for site number 01193050 (Connecticut River at Middle Haddam, Conn.), so no data were collected for this site for the 180-day interval.

Surface-water and groundwater samples were collected and processed for filtered and unfiltered water analysis in accordance with the USGS National Field Manual for the Collection of Water-Quality Data, which includes standard USGS sample collection methods (U.S. Geological Survey, variously dated). Field measurements of water temperature, specific conductance, pH, and dissolved oxygen were collected and recorded following standard USGS protocols (U.S. Geological Survey, variously dated). In addition, field titrations were performed for unfiltered ANC and filtered alkalinity at each site. Field replicates for ANC and alkalinity measurements were performed at some, but not all, sites. Information for each sample is provided in table 3.

A large sample volume was collected to ensure enough sample water for all required analyses in the HTS. The overall sample volume needed for all the sample bottle rinses, filter conditioning, sample splits, field titrations, and analyses was at least 12 liters (L). For samples to be as uniform as possible, surface-water samples were collected and composited in a large churn (14.0 L) where possible. Groundwater samples were collected from a pumped well. The volume of sample for each bottle was approximately 100–125 milliliters (mL).

A minimum of 21 sets of bottles (4 bottles per set), plus an additional 3 sets of spare bottles (96 bottles total), were submitted to the NWQL for each site to accommodate three replicate analyses for each of the seven time intervals. Separate bottles for each time interval and each replicate were collected so that bottles could be opened immediately prior to analysis instead of reusing samples that had been previously opened and exposed to the atmosphere for various amounts of time. The use of three separate bottles for each time interval allowed for comparing bottle-to-bottle sample stability as well as stability through time.

This effort required the following bottles to be collected as a bottle set: 125 mL filtered, chilled, high-density polyethylene, amber, unacidified (FCC); 125 mL unfiltered, chilled, high-density polyethylene, acidified (WCA); 250 mL (125 mL minimum volume) filtered, high-density polyethylene, unacidified (FU); 250 mL (125 mL minimum volume) unfiltered, high-density polyethylene, unacidified (RU; table 1).

All samples were shipped overnight to the NWQL. Upon sample receipt at the NWQL, the bottles were checked for proper labelling. The first set of bottles (FCC, WCA, FU, and RU) for each site was logged into the NWQL Laboratory Information Management System (LIMS) as a routine sample and received an NWQL Analytical Services Request (ASR) number. No other bottles were logged into the NWQL LIMS for that site for that date and time, but all bottles from that site had the NWQL ASR number on the label for tracking purposes. Bottle labels indicated the following: the proposed analysis date after collection (eight time intervals total); bottle number (1, 2, or 3 for each of the selected time intervals); container type; and analytical tests required for each container type (table 4).

The specific requirements of the HTS created a challenge to the NWQL’s routine procedure for one-time analysis of LIMS-tracked environmental samples. To ensure that samples could be analyzed within 48 hours of arriving at the NWQL, sample labelling and distribution to the NWQL analysts were managed separately from the routine NWQL log-in procedure and uniquely designed for this particular study to ensure that all analytical time periods would be met as needed. Samples treated as FCCs and WCAs were shipped to and stored at the NWQL as chilled samples, whereas the FU and RU samples were stored at room temperature.

Sample stability was evaluated by using three approaches, two of which accounted for the variability of the analytical methods: (1) by evaluating sample results with respect to blind-blank quality-control data for the analytical methods; (2) by evaluating sample results with respect to natural matrix water or spiked reagent water quality-control data for the analytical methods; and (3) by comparing day-2 HTS analytical results to subsequent measurements within a criterion of ±10 percent difference from the day-2 results. These approaches are described in this section and explained in further detail in the “Interpretation of Analytical Results” section of this report. Two additional approaches of evaluating sample stability, from the American Society for Testing and Materials (ASTM, 2002) and from Gardolinski and others (2001), were investigated but were found to be too sensitive for the HTS results. These approaches are also described in this section.

The American Society for Testing and Materials (ASTM) provides a standard protocol for determining holding times that incorporates the range of variation among replicate measurements at the start of the holding time period. The ASTM allowable range of variation is described by equation 1:d = ±(t×SD)/n^1/2(1)where

A high percentage of sample results from the HTS would be considered instable when evaluated using the ASTM approach and day-2 replicate (triplicate) measurements to represent initial concentration in the samples at the start of the holding time period. The mean results for some observed properties from time intervals well after two days were within ±2 percent of the original mean value but would be considered instable according to the ASTM approach. Because the allowable range of variation calculated from the initial triplicate values determined on day 2 indicated unusually low variability, there was an unrealistically small allowable range of variation by the ASTM approach. For example, for the observed property ANC as calcium carbonate (CaCO₃), water, unfiltered, the allowable range that indicates stability for sample 3 based on day-2 results is 158.7 mg/L±1.4 mg/L. This acceptance range is less than ±1 percent from the initial mean value. Using the range allowed by the ASTM approach results for days 60, 90, 150, and 180 are all flagged as instable (fig. 2). Therefore, the ASTM approach was considered not applicable for the sample stability evaluation.

An approach similar to the ASTM approach, by Gardolinski and others (2001), was also explored. The Gardolinski approach provided a more generous allowable range of variation because the variability from the initial triplicate values determined on day 2 and the variability associated with subsequent triplicate values were considered. Although this approach does consider more than single-day, single-analyst variability, it does not take into account long-term method variability, and this approach proved to be oversensitive for this study.

We believe that the oversensitivity of both the ASTM and Gardolinski approaches is due to the fact that single-operator variability on a single day and single instrument is significantly less than that of long-term (multiday and possibly multi-instrument and [or] operator) variability (ASTM, 2002). Because all original day-2 results are based on single-operator variability, it is reasonable to conclude that many results after day 2 that failed to fall within the allowable range of variation for the ASTM and Gardolinski approaches were from a stable sample, but the results did not meet the stringent criteria of the ASTM and Gardolinski approaches.

To determine when subsequent results varied significantly from the day-2 values, long-term method variability had to be considered. By properly characterizing actual method variability, it was possible to create limits (referred to as “stability limits” in this report) around the day-2 results within which subsequent results would fall if the sample was stable. Samples with results falling outside of those limits, or outside of expected method variability, could be considered instable.

Method variability depends on concentration. As concentrations decrease, absolute variability decreases, and as the concentration approaches the detection level, the variability stabilizes at a level whereby the noise starts to outweigh the signal. This is illustrated for the variation of the SD with mean concentration in figure 3 (Mueller and others, 2015). One of the challenges with the six samples in this study is that they all had different concentrations. Some analyte results are less than the detection level; some are more than 40 times the detection level. Method variability at all concentration ranges had to be considered to calibrate appropriate expectations.

Three separate approaches were used to calculate method and concentration-specific allowable ranges of variation (from the initial day-2 concentrations) for each observed property. The largest computed range of the three approaches is used in this study as the criterion to distinguish between (that is, identify) stable and instable analytical results. The three approaches are defined in the next three sections of this report.

A summary of field measurements and information on conditions at each sampling site for each set of samples used in the HTS are shown in table 6. The data in table 6 are intended to characterize the general status of the samples at the time of collection. Time series data for each set of triplicate results obtained for each sample are shown in appendix 1, figure 1.1 and 1.2. All the data shown in appendix 1, figure 1.1 and 1.2, are available in the data release that accompanies this report (Struzeski, 2025).

For the 12 observed properties studied, the goal was to determine how long after sampling would changes in initial property concentrations be first detected (table 1). Initial concentrations are represented by the mean concentration of the triplicates analyzed within 48 hours of sampling, otherwise referred to as day-2 measurements. Observed properties were evaluated by reviewing percent recovery results and concentration results through time (app. 1, fig. 1.1 and 1.2). Changes from initial day-2 concentrations were considered relevant when differences in subsequently determined mean concentrations were greater than the expected method variability adjusted for concentration or greater than ±10 percent, whichever was larger. Mean concentrations determined after day 2 were binned into the following intervals, in days: 2, 14, 30, 60, 90, 120, 150, and 180. Actual days varied within each bin. Percent recoveries were calculated using equation 8:

The time series charts shown in figure 1.1A–H (app. 1) include parameters and triplicate sample results evaluated with allowable range-of-variation limits set at the day 2 ±10 percent approach. The time series shown in figure 1.2A–H (app. 1) include parameters and triplicate sample results evaluated using stability limits from method precision based on the blind-blank data method (Approach 1) described in the “Approach 1: Stability Limits Determined Using Blank Water” section of this report. Ultimately, limits using the natural matrix and spiked reagent water data (Approach 2) were not used to evaluate stability for this study because there were no cases in which those limits were the largest of the three approaches investigated.

Mean concentrations for time intervals after day 2 falling within the selected limits were assigned an “S” (stable). Mean concentrations falling outside the limits were assigned an “I” (instable). Note that while the largest of the three limits calculated was used for this analysis, individual projects should consider their data quality objectives and use a more conservative approach when warranted. A summary of S and I ratings used to determine data usability are listed by initial mean concentrations in table 7.

Observed property stability was assessed based on mean triplicate concentrations after day 2 relative to upper and lower stability limits. Some observed properties, such as orthophosphate as P, water, filtered, exhibited little change in concentrations through time (fig. 1.1H), and mean sample concentration results were between the upper and lower stability limits for all samples studied. However, the assessment of stability was more difficult when analyte recoveries started to show large changes in concentrations through time, such as changes more than ±20 percent of day-2 results for total ammonia as N, filtered (fig. 5). Therefore, it is important to note that sample instability is specific to each analyte and sample matrix combination.

Differences in analytical results among the triplicate sample bottles for each single-time interval and site could be due to several sources, which include analytical method bias and variability, variation in field processing, and within-bottle reaction kinetics.

Based on the limited number of six sample sources used in this study, some patterns emerge for the 12 observed properties studied (table 7). Five observed properties generally indicate stability (S rating) for as many as 180 days (total nitrogen as N, both filtered and unfiltered; orthophosphate as P, filtered; and phosphorus as P, both filtered and unfiltered.) Other observed properties indicate stability for as many as 180 days for some matrices but not for others. Finally, some observed properties indicate stability at 180 days but some instability (I rating) before that. General determinations of analyte stability are listed below by observed property.

The conclusions of this study are based on the following:

Based on the limited number of six sample sources used in this study, the general determinations of analyte stability for each observed property are as follows:

Sample instability varied by observed property and matrix. However, it is not possible to give general guidance for sample stability duration based on matrices alone. No correlations between field measurements of sample characteristics and sample instability were observed. No conclusions can be drawn pertaining to biological causes of sample degradation. Although observations for some properties indicate sample stability well beyond the recognized NWQL method holding times, this is not necessarily the case for matrices and seasonal characteristics that were not investigated.

The effect of sample stability on the outcomes of water-quality studies is linked to study-specific data-quality objectives. For this report, upper and lower stability limits were set at plus or minus 10 percent for Approach 3. Water-quality studies with more flexible data-quality requirements could use a wider stability interval.