1. Introduction

Since the 1990s, biologists in the Puget Sound basin have documented a recurring phenomenon in which adult coho salmon die within hours of entering urban streams to spawn, often before completing redd construction, and frequently exhibiting no external signs of disease or injury. This pattern, termed Urban Runoff Mortality Syndrome (URMS), was strongly associated with the first flush of autumn stormwater but resisted causal identification for nearly three decades despite extensive toxicological screening (Scholz et al., 2011; Feist et al., 2017).

In December 2020, a non-targeted analytical chemistry effort identified the causal agent: 6PPD-quinone (6PPD-q), an ozone-driven transformation product of N-(1,3-dimethylbutyl)-N'-phenyl-p-phenylenediamine (6PPD), an antiozonant compounded into essentially all vehicle and commercial vehicle tires to prevent rubber cracking (Tian et al., 2020). 6PPD-q is acutely lethal to coho salmon at concentrations in the sub-microgram-per-liter range, making it among the most potent toxicants ever characterized for a vertebrate species in an environmentally realistic exposure scenario.

Tire-wear particle generation, and by extension 6PPD-q loading, scales with vehicle mass, tire count, braking frequency, and cumulative vehicle-miles traveled — all of which are disproportionately elevated in corridors serving freight and drayage traffic relative to typical passenger vehicle corridors. This raises a direct and, for the logistics sector, commercially material question: do distribution and intermodal freight hubs represent point-source-scale contributors to 6PPD-q loading in adjacent salmon-bearing waters, and can engineered stormwater controls meaningfully reduce that loading?

This study was conducted along Harmon Creek, a fourth-order tributary in the Kent Valley (King County, Washington) that receives channelized stormwater from a segment of State Route 167 and from the Alpine Logistics Corp Kent Valley Intermodal Hub, a 41-hectare truck marshalling, cross-dock, and trailer storage facility handling an estimated 2,600 heavy-duty truck movements per day. Following a widely reported pre-spawn coho die-off at the Harmon Creek confluence in November 2019, the lead author began independent baseline field monitoring the following month as part of a master's thesis project on urban runoff mortality syndrome in Kent Valley tributaries, prior to and independent of any facility involvement. The Washington State Department of Ecology issued an administrative order to Alpine Logistics Corp in December 2019 citing unpermitted stormwater discharge; in January 2020, Alpine Logistics Corp's newly established Environmental Stewardship Initiative began funding an expansion of the lead author's ongoing monitoring program through the Cascadia Institute for Watershed Sciences. Alpine Logistics Corp subsequently installed a bioretention and media-filtration retrofit across its four permitted outfalls between April and September 2020, ahead of the second coho return cycle. This paper reports 6PPD-q concentrations and coho pre-spawn mortality observations across both the pre-retrofit (Wet Season 1, 2019–2020) and post-retrofit (Wet Season 2, 2020–2021) return cycles, and their relationship to freight traffic volume.

2. Background and Regulatory Context

Reported acute LC50 values for coho cluster in the range of 0.041–0.095 μg/L in laboratory exposures (Tian et al., 2020), though field-realistic mixtures containing co-occurring tire-wear leachate constituents may modify observed toxicity, and sensitivity is expected to vary across salmonid species and life stage.

Green stormwater infrastructure (GSI) — including bioretention cells, soil-media filtration, and engineered infiltration — has independently been shown to reduce acute toxicity of urban runoff to coho in controlled comparisons (McIntyre et al., 2018; Spromberg et al., 2016), primarily via sorption of organic contaminants to soil and biochar media rather than targeted removal of 6PPD-q specifically. At the time of this study, no state or federal water quality criterion for 6PPD-q existed in Washington; Alpine Logistics Corp's retrofit was undertaken as a best-management-practice commitment associated with its December 2019 administrative order, and this research was commissioned, in part, to independently evaluate that commitment.

3. Materials and Methods

3.1 Study Area and Site Selection. Nine stormwater and in-stream sampling sites were established along a 3.4 km reach of Harmon Creek and its principal tributary storm drains, spanning from approximately 400 m upstream of the Alpine Logistics Corp facility (reference sites HC-1 and HC-2) through the facility's four permitted outfalls (HC-3 through HC-6), a mid-reach integration point (HC-7), and two downstream sites above and below the SR-167 highway crossing (HC-8, HC-9).

3.2 Sample Collection. First-flush grab samples were collected within 90 minutes of the onset of qualifying rainfall events (≥2.5 mm following ≥72 hr antecedent dry period), per the protocol of Peter et al. (2018). Sampling was conducted across 11 qualifying storm events in Wet Season 1 (October 2019–March 2020, pre-retrofit) and 10 qualifying storm events in Wet Season 2 (October 2020–March 2021, post-retrofit). The lead author's Wet Season 1 sampling began in December 2019 as an unfunded thesis pilot effort; Alpine Logistics Corp funding, received in January 2020, supported the expanded nine-site network and chemical analysis from Wet Season 1 onward.

3.3 Chemical Analysis. Samples were filtered (0.7 μm glass fiber), extracted via solid-phase extraction, and analyzed for 6PPD-quinone by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) in multiple reaction monitoring mode, with quantification against an isotopically labeled internal standard (6PPD-quinone-d5). Method detection limit was 0.008 μg/L; recoveries in matrix spikes ranged 84–109%.

3.4 Biological Monitoring. Carcass surveys were conducted weekly along the Harmon Creek reach during the October–December coho return period by trained field technicians, following the standardized URMS field necropsy checklist (gill flaring, loss of equilibrium, surface orientation, absence of net-marking or predation injury) used to distinguish stormwater-associated mortality from other causes of death (Scholz et al., 2011; Spromberg et al., 2016).

3.5 Retrofit Installation. Alpine Logistics Corp's bioretention and media-filtration retrofit was constructed across outfalls HC-3 through HC-6 between April and September 2020, with final media placement completed at all four outfalls by mid-September 2020, approximately four to six weeks before the onset of Wet Season 2 first-flush sampling. Manufacturer specifications for the bioretention soil media indicate a recommended 90-day microbial establishment period prior to peak sorption performance.

3.6 Traffic and Loading Data. Average daily truck volume at the facility gate was obtained from Alpine Logistics Corp's internal gate-access logging system for the corresponding sampling dates.

3.7 Statistical Analysis. Site-level 6PPD-q concentrations were log-transformed prior to analysis. Pre/post-retrofit differences were evaluated using paired comparisons at each outfall site; the relationship between daily truck volume and 6PPD-q concentration was evaluated using ordinary least squares regression. All analyses were performed in R (v4.0.3).

4. Results

4.1 Pre-Retrofit 6PPD-Quinone Concentrations. During Wet Season 1, first-flush 6PPD-q concentrations at facility outfall sites (HC-3–HC-6) ranged from 0.94 to 4.92 μg/L, exceeding the upper literature LC50 estimate (0.095 μg/L) by one to two orders of magnitude. Reference sites upstream of the facility (HC-1, HC-2) ranged from 0.18 to 0.31 μg/L, itself above the acute threshold, reflecting the underlying contribution of SR-167 highway traffic independent of the facility.

^† Post-retrofit samples at HC-6 were excluded from analysis in 3 of 10 Wet Season 2 sampling events due to reported autosampler calibration faults; the site was omitted from the Wet Season 2 summary statistics reported in Section 4.3 pending equipment replacement.

4.2 Relationship to Truck Volume. Across facility outfall sites, log-transformed 6PPD-q concentration was positively associated with average daily truck volume at the corresponding gate count (R² = 0.71, p < 0.01, n = 33 site-events, HC-6 excluded per above). No statistically significant relationship was observed at upstream reference sites (R² = 0.06, p = 0.41).

4.3 Retrofit Performance and Mortality Outcomes. Across the five outfall and downstream sites with complete pre/post data, mean 6PPD-q reduction was 92.0% (range 86.8–92.0%; SD not reported in source dataset). Concurrent carcass surveys recorded pre-spawn mortality consistent with URMS symptomatology in 61% of recovered coho carcasses (n = 47) in Wet Season 1, declining to 9% (n = 52) in Wet Season 2.

5. Discussion

These results are consistent with a substantial contribution of freight-corridor tire-wear loading to 6PPD-q concentrations in Harmon Creek, and with a meaningful reduction in that loading following the biofiltration retrofit at four of five outfall sites with complete data. The magnitude and post-retrofit uniformity of the reduction (92.0% at three of four treated outfalls, to one decimal place) is larger, and considerably more consistent across sites, than reductions reported for comparable bioretention media treating mixed contaminant loads in prior work (McIntyre et al., 2018 report 58–79% reduction in aggregate toxic unit loading across heterogeneous sites), and may reflect a design specific to this facility, a narrower range of influent conditions than in prior comparative studies, or unresolved variability that a fuller dataset would reveal.

The missing Wet Season 2 dataset at HC-6 — the outfall situated immediately adjacent to the marshalling yard and the single pre-retrofit site with the highest recorded concentration (4.92 μg/L, more than double the next-highest site) — is a material gap. Repeated equipment fault at the same site across three of ten sampling events, without a corresponding fault at any other autosampler in the network, was not independently investigated as part of this study.

6. Limitations

This study is limited by a two-season monitoring window, reliance on facility-provided (rather than independently metered) truck volume data, the exclusion of the HC-6 site from post-retrofit summary statistics, and a carcass survey method that cannot fully exclude co-occurring stressors (temperature, dissolved oxygen, predation scavenging) as contributors to observed mortality. Wet Season 2 sampling also began only four to six weeks after final media placement at the retrofit outfalls, substantially short of the manufacturer's recommended 90-day microbial establishment period; whether the observed reduction reflects mature, steady-state media performance or an initial adsorptive capacity that may decline with saturation over subsequent seasons cannot be determined from a single post-retrofit wet season. The uniformity of percent-reduction values across independent outfalls (Section 4.3) should be interpreted with caution pending replicate sampling and full disclosure of underlying event-level data, which were not included in the dataset made available to the authors for statistical review.

7. Conclusion

Freight-corridor facilities can be significant, spatially concentrated sources of 6PPD-quinone loading to salmon-bearing urban streams, and engineered biofiltration retrofits appear capable of substantially reducing that loading at treated outfalls. However, the incomplete post-retrofit dataset at the facility's highest-loading site, and the unusually uniform reduction values reported elsewhere, warrant an independently supervised follow-up monitoring program — including third-party autosampler maintenance and unredacted event-level data release — before this retrofit is characterized as a fully validated model for freight-sector 6PPD-q mitigation.

1. Introduction

Laurent et al. (2021) reported that a bioretention retrofit at a freight intermodal facility reduced stormwater 6PPD-quinone concentrations by 92.0% and reduced observed coho pre-spawn mortality from 61% to 9% across a single paired comparison of wet seasons. The study has since been widely cited in industry sustainability reporting and, by the reporting company's own account, materially supported its growth and public reputation over the following several years. Given the single-institution funding arrangement, the modest number of independent monitoring sites (n = 9), and the absence of any published independent replication, a purely statistical audit of the reported summary tables — without reference to the underlying chemistry, which the author is not qualified to evaluate — seemed a reasonable and low-cost check before any party made a material financial decision on the strength of the paper's conclusions.

This note restricts itself to what can be assessed from Tables 1–3 of the published manuscript. No raw chromatography output, autosampler logs, or event-level measurements were available to the author; the original authors state that this material is "available upon request, subject to funder review." We present findings in order of evidentiary strength: Sections 2 and 3 identify internal inconsistencies that follow directly from arithmetic on the manuscript's own published figures and require no distributional assumption; Sections 4 and 5 present statistical arguments that do rely on stated modeling assumptions, which we make explicit; Section 6 reports a supplementary digit-distribution check that is under-powered at the available sample size and is not load-bearing for our conclusions.

2. Arithmetic Reconciliation of Reported Reduction Percentages

Table 4.1 of the original manuscript reports, for each site, a mean pre-retrofit concentration, a mean post-retrofit concentration, and a percent reduction. These three columns are not independent — the third is arithmetic on the first two — so they can be checked against one another without any statistical model at all. We recompute reduction = (pre − post) / pre for each site using the paper's own published means:

The two sites outside the retrofit's headline narrative — the upstream reference (HC-1) and the downstream site (HC-9) — reconcile exactly with their own published means, to the one decimal place at which the paper reports them. The three treated outfalls central to the 92.0% headline figure do not: correctly rounded, their own published pre/post means imply 91.9%, 92.1%, and 92.2% respectively — three different values spanning 0.22 percentage points — not the single identical value of 92.0% printed for all three. This is not a matter of statistical power or sample size; it is a direct check of the manuscript's own numbers against each other, and the published reduction column does not follow from the published concentration columns in the same rows.

We note further that the facility-wide headline of "92.0% reduction across treated outfalls" (Abstract; Sec. 4.3) is arithmetically just the mean of these same three already-identical printed values, and so provides no information beyond what is already in the row-level figures above — it is not an independent corroborating measurement, despite being presented as the paper's central quantitative result.

3. Reported Sample Completeness vs. the Published Table

Section 4.3 of the original manuscript states: "Across the five outfall and downstream sites with complete pre/post data, mean 6PPD-q reduction was 92.0% (range 86.8–92.0%)." The facility's outfall and downstream sites, as defined elsewhere in the same manuscript (Sec. 3.1), are HC-3, HC-4, HC-5, HC-6, and HC-9 — five sites. But the manuscript's own table marks HC-6's post-retrofit value as missing ("n/a"), leaving only four sites (HC-3, HC-4, HC-5, HC-9) with a non-missing post-retrofit reading. The stated range (86.8–92.0%) is likewise consistent only with these same four values, not five.

We can identify no reading of the manuscript's own definitions under which "the five outfall and downstream sites" both have "complete pre/post data" and are consistent with the table immediately preceding this sentence. Either a fifth site's data informed the reported mean and range without appearing anywhere in the published tables, or the sample size stated in the Results text is simply incorrect. We are not in a position to determine which from the published material alone, and flag it as a specific, narrow, and easily resolvable item for the original authors or the repository to clarify.

4. Statistical Implausibility of the Residual Clustering

Section 2 shows that the three treated outfalls' own recomputed reductions — 91.9%, 92.1%, 92.2% — are not in fact identical, but they are still tightly clustered: a spread of just 0.22 percentage points across three physically distinct outfalls with meaningfully different pre-retrofit loading (1.86, 2.41, and 1.53 μg/L respectively, a 58% relative spread in influent concentration). Bioretention performance depends on local soil composition, compaction, flow rate, and antecedent moisture, which vary between physically distinct outfalls even under a common design specification, and the original authors themselves cite comparable published work — McIntyre et al. (2018) — reporting a 58–79% reduction range across heterogeneous sites treating similar contaminant loads. Taking this 21-percentage-point published range as a working estimate of between-site variability and approximating it as roughly four standard deviations of spread (a standard rough conversion for a moderate-sized heterogeneous sample), we obtain σ ≈ (79 − 58)/4 ≈ 5.3 percentage points.

Under a normal model with this standard deviation, the probability that three independent draws fall within a range of w = 0.22 percentage points of one another can be approximated analytically (for w small relative to σ) as P(range ≤ w) ≈ 3w²∫φ(x)³dx = 0.276 × (w/σ)². With w/σ = 0.22/5.3 ≈ 0.0415, this gives P ≈ 0.276 × 0.0415² ≈ 4.8 × 10⁻⁴ — roughly 1 in 2,100. This is an approximation, not a simulation result, and it is sensitive to the assumed σ. Taking the most conservative direction for our own argument — halving σ to 2.65 points, i.e. assuming the true site-to-site variance is far smaller than the cited comparison study suggests, which makes tight clustering less surprising — still only raises the probability to roughly 1 in 530. Taking σ at face value from the cited range (5.3 points) or larger gives odds of 1 in 2,100 or longer. We present the derivation in full, including the conservative case, so the estimate can be checked or revised against a different variance assumption; we do not claim precision beyond the order of magnitude, but the finding is not an artifact of a favorably chosen σ.

5. Missingness at the Highest-Concentration Site

Site HC-6 recorded the highest pre-retrofit concentration in the study (4.92 μg/L, more than double the next-highest site) and is the only site with no reported post-retrofit value, attributed to autosampler calibration failure in 3 of 10 Wet Season 2 sampling events. The original manuscript states that this material was excluded rather than partially reported, and that event-level data are available only "subject to funder review" — we do not have access to it, and none of what follows assumes we do.

5.1 Exact probability of the failure pattern. The manuscript states that no other autosampler, at any of the other eight monitored sites, failed at any point across the same ten-event window. Treating the design as nine sites sampled at each of ten Wet Season 2 events (90 site-events total, per Sec. 3.1–3.2) and treating the three reported failures as occurring independently and with equal probability at any of the 90 site-events under a null "failures unrelated to site" model, the probability that all three land specifically within the ten site-events belonging to HC-6 — the single highest-concentration site — is exactly C(10,3)/C(90,3) = 120/117,480 ≈ 0.00102, or approximately 1 in 980. This is an exact combinatorial calculation, not a simulation, and depends only on the site/event counts and failure count the original manuscript itself reports.

5.2 A defensible bound on the excluded concentration, without the unavailable raw readings. We do not have HC-6's underlying post-retrofit measurements and do not claim otherwise. We can, however, construct a best-case bound using only published figures: even if HC-6 achieved the single best reduction rate recomputed for any treated outfall in Section 2 (92.2%, at HC-5), its implied post-retrofit concentration would be 4.92 × (1 − 0.922) ≈ 0.384 μg/L — still roughly four times the upper end of the acute LC50 range the original authors themselves report for coho (0.041–0.095 μg/L, Sec. 2). In other words, under the most favorable reduction rate documented anywhere in the retrofit's own published data, the excluded site would still plausibly remain in acutely toxic territory post-retrofit — the opposite of the retrofit's headline claim. This bound uses only figures already in the manuscript (HC-6's pre-retrofit concentration, the best documented per-site reduction rate, and the manuscript's own cited LC50 range) and requires no access to unpublished data.

5.3 Timing of the failures. The manuscript does not disclose which of the ten Wet Season 2 events the three HC-6 failures occurred during, so we cannot test whether they cluster in the highest-flow portion of the season as opposed to being scattered — this would be a natural, specific, and easily answerable follow-up question for the original authors or the repository, but we flag it as an open question rather than a finding, since we have no basis in the published material to characterize the timing either way.

6. Supplementary Digit-Distribution Analysis

Naturally occurring measurements spanning several orders of magnitude — as stormwater contaminant concentrations typically do — tend to follow Benford's Law: the leading digit 1 occurs roughly 30% of the time, decreasing monotonically through leading digit 9 (Benford, 1938; Nigrini, 2012). Applying this check to the concentration values reported across the manuscript's tables (n = 21, including the event-level entries in Tables 1–2 that are not reproduced in full above) is, at this sample size, under-powered: several Benford bins have expected counts below 5, and a standard χ² goodness-of-fit test is not strictly valid here. Using a Kolmogorov–Smirnov statistic against the Benford CDF instead (reference distribution obtained by bootstrap resampling, appropriate for small n) gives D = 0.31, a nominal p = 0.008, though we would not treat this as a stand-alone finding given the sample size. We similarly note the terminal (final) digit distribution of the same 21 values shows some clustering on 0 and 5 relative to the uniform expectation (Preece, 1981), consistent with rounded or estimated rather than raw instrument values, but subject to the same small-sample caveat.

We report this section for completeness because it was one of the first checks we ran, but we want to be explicit that it is not load-bearing: unlike Sections 2, 3, and 5.1, which are exact or near-exact given the manuscript's own disclosed figures, this digit-distribution check would benefit from a larger sample than is verifiably available to us, and a skeptical reader should weight it accordingly.

7. Discussion

Sections 2 and 3 do not depend on any statistical model, sampling assumption, or estimate of variance — they are direct checks of the manuscript's own published numbers against each other, and in both cases the numbers do not reconcile. Section 4 shows that even the true, recomputed values are implausibly tightly clustered under variance assumptions drawn from the original authors' own cited comparison study. Section 5.1 gives an exact, one-in-several-hundred probability that the reported equipment failures would, by chance, concentrate entirely at the single highest-concentration site. Each of these could, individually, have an innocent explanation — a transcription slip, a rounding convention applied inconsistently, an unlucky equipment fault, unusually consistent engineering performance. What is harder to explain innocently is that they all point in the same direction, all favor the more optimistic reading of the retrofit's performance, and none of them favor a less optimistic reading. Resolution ultimately requires the material the original authors describe as available only "subject to funder review": event-level chromatography output, unredacted autosampler maintenance logs, and — ideally — independent field resampling at HC-6 by a party with no funding relationship to the facility operator. We are not in a position to obtain that material and take no position on whether the irregularities identified here reflect data manipulation, uncorrected error, or some combination; we simply note that the published record, on its own internal arithmetic, does not support the confidence with which its conclusions have been publicly represented.

8. Combining the Evidence

We do not combine the findings above into a single omnibus p-value: Sections 2 and 3 are deterministic inconsistencies, not probabilistic tests, and combining them with the probabilistic findings in Sections 4–6 via a method such as Fisher's would overstate precision it does not have. What we can say is that the findings are drawn from largely independent features of the published data — row-level arithmetic, stated versus tabulated sample size, cross-site variance, failure-site concentration, and digit distribution — and would not be expected to co-occur under any single innocent process, while the two strongest findings (Sections 2 and 3) require no probabilistic argument at all to establish. An innocent-explanation defense now has to hold simultaneously across all of them, not just the weakest one.