Endodrift Upground Reservoir Hypothesis

Scientific Paper
Title	Endodrift Upground Reservoir Hypothesis
Read in full	Link to paper
Author(s)	Jeffrey N Cook
Keywords	cancer cluster, pesticides, endodrift, upground reservoir, Clyde Ohio, probability
Published	2011
No. of pages	30

Read the full paper here

Abstract

The author only heard that Clyde, Ohio was classified a "Cancer Cluster" after Christmas of 2010, brought to his attention by his wife, after reading an article on it from the Internet. The point of this case that concerned him, as with most cancer cluster cases, is that the cause of the dramatic and sudden rise in cancer cases in Clyde was still unknown, leaving the community still unsettled. The small town of just over six thousand people is less than an hour's drive from the author's home, so on New Years Day, he performed a cursory survey of the area. While the author is by no means a noted expert on cancer, nor a noted expert in environmental science, having only taken one university course in Environmental Science and one in Biology, his career has had notable moments of discovering general scientific causes and/or effects that have eluded others. However, having never been to Clyde, Ohio, he did not make any agreement or disagreement with the assumptions added to the news article almost seemingly ad hoc. Some of these articled suspicions were 1) local factories, 2) an environmental waste dump a half-mile underground and 3) the water supply itself. In general, the rise in cases did not appear to be of natural, long term environmental causes, as there has been a growing population in Clyde for decades, and the rise in numbers of these cases began to surface around 2000, peaking in 2005 and 2006.

Overview

The paper's full title in the manuscript is "Endodrift-Up Reservoir Hypothesis: a Case for Clyde", written by Jeffrey N Cook on 3 January 2011 and last updated in October 2012. It is not a physics paper but an amateur environmental-epidemiological investigation, prompted by news coverage of the childhood cancer cluster identified in Clyde, Ohio — twenty children diagnosed between 2001 and 2009 in a town of about 6,000. Cook drove to Clyde on New Year's Day 2011, photographed the town, its park, its reservoir and the farmland to the south, and then constructed a probabilistic model intended to identify the most likely source. He is explicit about his standing: one university course each in environmental science and biology, and a career in quality assurance. The document is illustrated throughout with sixteen of his own photographs.

His conclusion is a two-part hypothesis. First, that agricultural pesticides reach Clyde's water supply by endodrift — seepage into the ground and into drainage ditches after rain, as opposed to ecodrift, airborne drift from aerial spraying. Second, and this is the part he treats as novel, that the town's upground reservoir — a basin raised above grade and filled by pumping from Raccoon Creek, rather than a dammed impoundment with continuous outflow — concentrates rather than dilutes whatever the creek carries. His departure from the official account is procedural rather than theoretical: he argues that the EPA's negative water test was not merely unlucky but methodologically "inappropriate", because it was taken in January, months after any spraying and any subsequent rain, when by his own probability model contamination would be least detectable.

The argument

Ruling out the common culprits

Cook works through four standard suspects. Radon and uranium in granite he dismisses on timescale grounds: uranium's long half-life means a natural deposit present in 2011 would have been present decades earlier, whereas the Clyde cases rose from 2001 and peaked in 2005–2006. He notes that the largest uranium mines are in Australian ranges long regarded by Aboriginal people as cursed or sacred because of unexplained illness, and takes this as evidence that such exposures are "centuries in the making, not years". High-voltage power lines he rules out by inspection — as the fourth generation in a family of electricians he judged Clyde's electrical construction "admirable", with nothing objectionable near schools, churches or parks. Natural toxic emissions he treats as possible but untestable: the sulphurous springs at Green Springs four miles away show that gas emission occurs in the region, but a one-off emission could not be caught unless the air happened to be sampled at that moment, and accepting such an explanation would tempt the community to stop looking. That leaves pollution, and specifically the water supply.

His reason for favouring a waterborne agent is one of the paper's better arguments. Inhaled or chewed carcinogens tend to produce cancer at the site of first contact — lung cancer in smokers, mouth cancer in tobacco chewers — whereas "when a substance is suspended in a fluid and then ingested, the fluid can carry the substance throughout the body". Since the Clyde cancers were of varied types rather than concentrated in one organ, he infers a circulating rather than a contact exposure. He also dismisses the strontium finding: strontium-90 requires nuclear fallout, which Clyde has not experienced, and the common isotopes behave chemically like calcium and magnesium.

He also rejects the two suspects the news coverage favoured. The factories, including the Whirlpool plant, he places downstream of the water treatment intake, so that they "could not at all be leaking water back up-stream". The deep-injection waste half a mile below the town he dismisses because such waste is mostly fertiliser-like, because Raccoon Creek is shallow and non-cavernous, and because half a mile is a large depth relative to a town whose maximum elevation is not much above 700 feet.

Reading the wildlife

A short methodological digression argues that the thriving mallard population on Raccoon Creek is not evidence of clean water. Waterfowl digestive systems, he says, are far more tolerant of toxins than human ones — birds routinely eat naturally toxic seeds and swallow clay and grit to neutralise them — so "if mallards were representative of human digestion, scientists would be using lab ducks instead of lab rats". Better indicators would be declines in frogs, turtles, snakes and small mammals. He illustrates the point with photographs from the polluted Christopher Creek near his own home, where ducks flourish but a dead squirrel lies on the bank, and notes that in Clyde he saw no wildlife at all apart from the mallards, and none in the reservoir.

Topography and the endodrift scenario

Clyde and the land north toward Lake Erie are flat, but a mile or so south and west the terrain rises sharply to 600–700 feet, "creating a funnel-like drain leading right back down toward the town". Raccoon Creek rises in that farmland and runs down through Clyde to Lake Erie, with essentially nothing but rolling hills and fields along its upstream course. Cook argues that farmers on sloped ground tend to apply pesticide in larger quantities because runoff reduces its effectiveness on a hillside, so the topography raises both the dose applied and the fraction delivered. He lays out a twelve-step chain: spraying, rain, runoff into field ditches, ditches to small creeks, creeks to Raccoon Creek, intake by the pump — which operates only when water levels are high — into the upground reservoir, treatment, storage in the water tower for drinking water and the adjacent water park, with the unpumped remainder flowing through the town's parks and into Lake Erie.

The tabletop reservoir experiment

To test the second half of the hypothesis Cook ran a kitchen-scale analogue: two equal containers of water, one with a drain plug left continuously and slowly draining (the dammed reservoir) and one sealed (the upground reservoir), with food colouring introduced by eyedropper to represent pesticide and water withdrawn by a second eyedropper to represent the supply draw. Over equal times, "far more food coloring became concentrated in the container without the plug", and decanting to a further container (the water tower) did not help. He also reports a subtler observation he considered significant: in the draining container the dye migrated toward the outlet by suction, whereas in the sealed container it drifted slightly toward the intake dropper itself. He connects this to a demographic pattern — that upground reservoirs have proliferated in central and southern Ohio and Indiana in recent decades while comparable hilly farm regions further west rely on dams and natural lakes — which he offers as a possible reason why Clyde's problem, and that of another cluster town an hour south, might be locally distinctive.

The probability apparatus

The second half of the paper builds a chain of defined factors. The Random Spray Factor starts from a Bernoulli trial Q(ρ,d) = 1 if ρ < 1/d, else 0, summed over farmers and averaged over months, and is used to argue that unregulated spray dates cluster — some days carry several sprays, others none. Restricting spraying to three months of the year drops the daily figure for thirty farms from 100% to 25%; restricting each farm to one controlled spray per month drops it to 3.33%, and to 0.83% over a full year.

The Farmer Drift Control Factor (FDCF) is then given a per-farm form, the Drift Factor DF = S/EDF, where S is the field's rise over run and EDF is the farm-to-creek distance divided by the creek-to-pump distance. Two worked examples contrast a farm with a 100-foot rise over 3,000 feet (DF ≈ 417, which he glosses as roughly a 1-in-5 chance of endodrift) against a nearly flat farm with a 2-foot rise (DF ≈ 0.83, a 1-in-1,000 chance). From these he draws practical conclusions: farms further upstream contribute more because creeks are narrower there and concentrations accumulate along the length; the single most effective mitigation is unsprayed buffer land between field and ditch, because "the filtering by means of natural earth itself is the most effective means; the thicker the filter, the cleaner the water"; and pushing the FDCF below about 1 in 1,000 yields diminishing returns, so further burden should fall on the water plant rather than the farmer.

The Rain Factor RF and a set of pump-intake quantities (creek width, depth and area at the pump, intake area, gallons drawn versus gallons passing) combine into a Clean Water Potential and finally a reservoir Pesticide Factor PF. Applied to a large catchment with a billion gallons of rain and 90% reaching the creek, he obtains PF ≈ 0.00138, and argues that a hazard at that per-day rate would reach all of a group of ten people within about three years.

Back-calculating from the case count

Finally, using the twenty childhood cancers from 2001 to 2009, a 24.3% under-eighteen share of Sandusky County population and a town population of 6,064 (about 1,473 children), Cook fits his model and finds that only d ≈ 200,000 reproduces the observed count — a Cancer Probability CP = 0.000005 per child per month, or about 0.208 diagnoses per month. He then posits CP = (EFCP + FTP + PF)/3, where EFCP is the chance ingestion causes cancer and FTP the chance a free toxin passes the filter, and inverts it as PF = |3CP − EFCP − FTP|. With generous assumed values (EFCP = 10⁻⁷, FTP = 10⁻³) he gets PF ≈ 0.1%, which he calls "an enormously high probability". His conclusion: pesticides should have been tested for in the creek and the reservoir, not only the treated supply, and in the days and weeks after the first rains following spraying, "not one time, months later".

Assessment

Two things in this paper are genuinely worth having. The first is the criticism of sampling design, which stands independently of everything else: a single January grab sample from a treated supply is a weak test for a seasonal agricultural contaminant, and Cook's recommendation to sample the creek and the raw reservoir in the days after the first rains following application is sound environmental practice, not an exotic proposal. The second is the observation about reservoir hydraulics. An upground reservoir filled by intermittent pumping and drawn down only for supply genuinely has a longer residence time and less flushing than a run-of-river impoundment with continuous outflow, and residence time really does govern how a pulse input is diluted or retained. His food-colouring demonstration is crude but it is a fair analogue of that specific point, and he is right that Ohio's upground reservoirs are a comparatively recent design. The framing question — why here and not in comparable farm country elsewhere — is the right question to ask of any cluster.

Against that, the paper's quantitative core does not hold up, and the failures are of a kind Cook could have caught himself. Most seriously, equation (5), CP = (EFCP + FTP + PF)/3, is not a valid combination of probabilities. Independent conditions that must all occur — the toxin being present, passing the filter, and initiating disease — multiply; averaging them produces a quantity dominated by whichever term is largest and bearing no relation to the joint risk. Because the whole numerical conclusion is obtained by inverting this equation, the headline result rests on it entirely. The arithmetic then compounds the problem: PF ≈ 0.1% is one chance in a thousand, but Cook calls it "a 1 out of 10 chance" in the same paragraph, a hundredfold misstatement in his own final number. Elsewhere the same conflation recurs, with expected counts read directly as probabilities — thirty sprays expected across thirty farms in thirty days is reported as "a 100% probability" of a spray on any given day, which does not follow. Equation (4), g = x, is likewise not a result but an identity: the mean of x Bernoulli(1/d) variables summed over d days is x by construction, so the "convergence" demonstrated in Graphs 2 and 3 confirms only the arithmetic of averaging. The symbol Q is also used for two different things, the 0/1 spray indicator and the quantity in "PF = 1 − Q". And the parameters that actually drive the answer — EFCP and FTP — are, in Cook's own words, "reasonable ad hoc values"; changing FTP by one order of magnitude changes PF by roughly the same factor, so the model returns whatever is put into it.

Several factual claims are also wrong in ways that matter to the eliminations. Cook attributes cancer risk near high-voltage lines to "ionizing gamma waves"; power lines emit extremely-low-frequency electric and magnetic fields, which are non-ionizing and carry photon energies many orders of magnitude below the threshold for breaking chemical bonds. Whatever the epidemiological status of ELF exposure, the stated mechanism is not the one at issue. His account of Los Angeles building aqueducts "to replace the city's local water supply" because of pollution misdescribes what were supply-capacity projects. And the elimination of the industrial sources rests on a single afternoon's impression of which facilities lie downstream of the intake — a judgement that requires plant records and groundwater data, not a drive through town. That elimination has not aged well: PCB contamination was subsequently identified in soil at the former Whirlpool Park property near Clyde and became the focus of regulatory attention and litigation, an exposure pathway Cook's water-only framing does not accommodate.

There is a deeper methodological difficulty. The paper reasons backward from a presumed cause to a probability, having eliminated alternatives by inspection rather than by measurement, and then treats the fitted parameter as evidence that the cause is real. Cook's closing claim — "it would be highly unlikely that the Clyde Cancer Cluster case could be due to any other reason than pesticides in the drinking water" — is far stronger than anything his calculation supports, since no pesticide was ever identified, none was ever measured, and the model would have fitted equally well for any agent whatsoever. He is honest about this in places, calling the survey "cursory" and the values ad hoc, and it should be said that the paper is written in good faith about a real community problem, without the grandiosity common in the genre. Read as a citizen's argument for better sampling and for a second look at how upground reservoirs are managed, it has merit. Read as a probabilistic identification of a carcinogen, it does not.

See also

• Jeffrey N Cook