For twenty years I have been saying some version of the same thing about leafy greens and cattle. Water runs downhill. Cows poop. Lettuce grows at the bottom of the hill.
Now FDA’s own scientists have put real numbers on it, and they deserve credit for doing it.
A new paper in the Journal of Food Protection, posted online November 6, 2026, by Mehran Niazi and colleagues in FDA’s Human Foods Program, builds a model that follows rainwater off cattle pastures and an animal feeding operation in the Salinas Valley and traces which leafy-green farms that runoff reaches on its way to the sea.
This is exactly the kind of work I have hoped to see for years. It is careful, transparent and honest about its limits. And it is the closest thing I have seen to a map of the plumbing that keeps sending E. coli O157:H7 into our salad bowls.
We Keep Finding Cows at the Scene
If you have read this blog for any length of time, you know the pattern. Here is a short refresher.
In 2006, bagged spinach sickened about 200 people nationwide and killed three. I represented many of those families. Investigators found the outbreak strain in cattle feces, feral swine, surface water and sediment on a San Benito County ranch, roughly a mile from an implicated spinach field. The final state and federal report could not say definitively how the spinach was contaminated, but it flagged irrigation wells sitting near surface waterways exposed to cattle and wildlife feces.
In spring 2018, romaine from the Yuma growing region sickened 210 people in 36 states, hospitalized 96, caused 27 cases of kidney failure and killed five. FDA found the outbreak strain in irrigation canal waterrunning alongside a massive concentrated animal feeding operation.
In fall 2018, romaine from California’s Central Coast sickened 62 people in the United States. FDA found the outbreak strain in sediment from an on-farm water reservoir in Santa Barbara County.
In fall 2019, three separate outbreaks were traced to Salinas Valley romaine. The largest sickened 167 people in 27 states. In its investigation report, FDA concluded that adjacent and nearby cattle grazing appeared to play a critical role, and listed water runoff, wind, animals and vehicles as possible routes from the cattle to the lettuce.
In fall 2020, it happened again: 40 sick in 19 states, 20 hospitalized, four with hemolytic uremic syndrome. This time FDA found the outbreak strain in a cattle feces sample on a roadside about 1.3 miles upslope from a produce farm with multiple fields linked to the outbreak.
And an FDA and CDC review of farm investigations tied to leafy-green STEC outbreaks from 2009 through 2021 found the same themes again and again. One of that paper’s authors, Kurt Nolte, is also an author on this new study.
So nobody serious is arguing anymore about whether cattle matter. The harder questions have been how, when and which farms. This paper takes those questions on directly, and that is a real step forward.
What the FDA Team Actually Did
Strip away the jargon and the method is pretty intuitive.
The team drew a study area out of 14 small watersheds in the Salinas Valley, about 1,239 square kilometers, or roughly 480 square miles. They mapped the leafy-green fields, the grazing pastures (each tagged with an estimated cattle capacity) and one animal feeding operation.
Then they pulled 48 rainstorms from 2018 through 2021 out of NOAA data. For each storm, they used the standard USDA Curve Number method to estimate how much rain would run off a pasture versus a compacted feedlot surface instead of soaking into the ground.
Next came the plumbing. Using elevation data, a computer figured out which direction water flows from every patch of ground and built a drainage network. Each pasture, the feeding operation and each farm were “snapped” onto that network, so the model could tell whether a given pasture actually sits upstream of a given farm, and how far the water travels between them.
Finally, each pasture’s influence was scaled by its cattle capacity and shrunk with distance. At their baseline setting, a pasture’s weight drops by about 63 percent over 10 kilometers of channel. Multiply by the storm runoff and you get a relative “runoff-exposure” score for every farm in every storm.
What They Found
The results are genuinely useful. The routing turned up 1,818 pasture-to-farm connections, involving 58 pastures and 277 downstream farms. The single animal feeding operation was hydrologically connected to 45 farms.
These are not short trips. The median distance along the channel from a pasture to a connected farm was about 16.8 kilometers, a little over 10 miles. The longest was 82.6 kilometers, more than 50 miles.
Most farms, 86.3 percent, had no modeled upstream pasture contribution at all under the baseline assumptions. That is good news for those growers, and it tells regulators where not to waste their time.
Here is the finding I like best: being next to cows is not the same as being downstream of them.
One pasture with an estimated capacity of 1,170 cattle physically overlapped 13 farm fields. Only two of those 13 were actually downstream of it. Yet that same pasture was hydrologically connected to 56 farms. A second pasture, about 722 cattle, overlapped eight farms, but only one was downstream, and it was connected to 36 farms in all.
Think about what that means for a buffer-distance rule. A setback measured in feet from the fence line protects the farm next door that might not be at risk, and does nothing for the farm ten miles down the creek that is.
For one farm sitting just 0.6 kilometers downstream of that 1,170-head pasture, the model kept about 94 percent of the pasture’s weight. During the biggest storm in the dataset, that one pasture accounted for a large share of the farm’s highest score.
And speaking of the biggest storm: every farm with a nonzero score hit its peak during the January 27 to 29, 2021 storm, which dropped an area-wide average of 147.3 millimeters of rain, almost 6 inches. Big storms do not just add runoff; under the Curve Number math, they add it disproportionately.
The team also stress-tested its assumptions. Changing the runoff parameters did not change the farm rankings at all. Changing how fast influence fades with distance left the overall rankings almost untouched, but it did reshuffle the very top. Compared with a simple count of upstream pastures, the top 25 farms overlapped by zero at the steepest decay setting, eight at baseline and 24 at the gentlest. In plain English: the model is solid at sorting farms into “worry” and “don’t worry,” but you should not bet everything on which farm is number one.
Honest About Its Limits, and That Is a Strength
The authors are admirably clear about what this model is not. It does not measure pathogen presence, concentration, contamination probability or food-safety risk. It does not model groundwater, floods, overbank flow, buffers, wetlands, die-off, sediment or irrigation practices. It says nothing about wind, dust, wildlife or feral pigs, all of which have shown up in past investigations. And the animal feeding operation is treated as a simple yes-or-no flag, because nobody had data on how it handles manure and stormwater.
That candor is a strength, not a weakness. The authors built a screening tool, said plainly that it is a screening tool, and published the data and sensitivity tests so others can build on it. A tool that tells investigators where to look is worth a great deal, and it is a big step beyond reports that could only point to “adjacent land use.”
The paper also points naturally to the next questions.
First, timing. These outbreaks tend to hit in the fall, while the biggest runoff in the Salinas Valley comes in winter. Connecting the two is the logical next piece of the puzzle. Maybe it is pathogens settling into reservoir and canal sediment, as FDA found in 2018. Maybe it is cattle droppings on roadsides and dust blowing across fields at harvest. This study gives a strong foundation for answering that, and I am hopeful FDA’s California Longitudinal Study, which the authors say this work is meant to inform, gets us there.
Second, sharing. The paper’s map of farm scores leaves out exact livestock locations so as not to identify individual operations. That was a reasonable choice for a published study. Going forward, the real payoff will come when growers at the bottom of the hill can see who is at the top of it, ideally through cooperation between neighbors. Getting the right information to the right people is the same goal behind my outbreak firm disclosure petition to FDA.
Building on This Work
First, put the results to work. FDA can share them with the farms that score high, confidentially if need be, and the industry’s food safety programs should build hydrologic connectivity into preseason and preplant risk assessments. FDA already told Central Coast growers after the 2020 outbreak to treat the recurring outbreak strain as a reasonably foreseeable hazard. This model tells them where that hazard is most likely to come from.
Second, sample after big storms. The authors suggest targeting upstream water and sediment during and after runoff-producing rain. That is inexpensive compared with a recall, and far better than another outbreak.
Third, pair the map with access. A model can point to a pasture or a feedlot; investigators still need to be able to sample there. That is why I am drafting a citizen petition asking FDA for the ability to sample at animal feeding operations during outbreak investigations. This study makes the case for it better than I could have. Combine that authority with this kind of map and investigators would know exactly where to start.
Fourth, expand it. Yuma, Imperial and Santa Maria are obvious candidates. The authors rightly note that Salinas results cannot simply be copied to other regions, but the framework is built to be adapted, and that is one of its best features. The 2020 Leafy Greens STEC Action Plan already calls for looking beyond farm boundaries. This is what that looks like.
Fifth, bring livestock owners to the table. FDA asked them in its fall 2020 outbreak report to consider monitoring their herds for human-pathogenic E. coli and sharing results with neighboring growers. A map showing which pastures drain toward which farms makes that conversation a lot more concrete.
Water runs downhill. We have known that since the spinach outbreak. Now, thanks to Mehran Niazi, Francisco Garces-Vega, Kurt Nolte, Michael Mahovic and Jane Van Doren, we have a map that shows which way it runs. That is good science and good public health. Let’s use it.
Sources
Niazi M, Garces-Vega F, Nolte K, Mahovic M, Van Doren JM. Event-based surface-water connectivity modeling of leafy-green farm exposure to livestock-associated runoff. Journal of Food Protection (2026), in press. https://doi.org/10.1016/j.jfp.2026.100959
Jay MT, et al. Escherichia coli O157:H7 in feral swine near spinach fields and cattle, central California coast. Emerging Infectious Diseases 13:1908–1911 (2007). https://pmc.ncbi.nlm.nih.gov/articles/PMC2876768
NBC News/Reuters. Source of 2006 toxic U.S. spinach uncertain, study says (March 2007). https://www.nbcnews.com/id/wbna17757136
CDC. Multistate outbreak of E. coli O157:H7 infections linked to romaine lettuce (2018). https://www.cdc.gov/ecoli/2018/o157h7-04-18/index.html
FDA. FDA investigated multistate outbreak of E. coli O157:H7 infections linked to romaine lettuce from Yuma growing region (2018). https://www.fda.gov/food/outbreaks-foodborne-illness/fda-investigated-multistate-outbreak-e-coli-O157:H7-infections-linked-romaine-lettuce-yuma-growing
FDA. Investigation summary: Factors potentially contributing to the contamination of romaine lettuce implicated in the Fall 2018 multistate outbreak of E. coli O157:H7 (2019). https://www.fda.gov/food/outbreaks-foodborne-illness/investigation-summary-factors-potentially-contributing-contamination-romaine-lettuce-implicated-fall
CDC. Outbreak of E. coli infections linked to romaine lettuce (2019). https://www.cdc.gov/ecoli/2019/o157h7-11-19/index.html
FDA. Factors potentially contributing to the contamination of romaine lettuce implicated in the three outbreaks of E. coli O157:H7 during the fall of 2019 (2020). https://www.fda.gov/food/outbreaks-foodborne-illness/factors-potentially-contributing-contamination-romaine-lettuce-implicated-three-outbreaks-e-coli
FDA. Factors potentially contributing to the contamination of leafy greens implicated in the fall 2020 outbreak of E. coli O157:H7 (2021). https://www.fda.gov/food/outbreaks-foodborne-illness/factors-potentially-contributing-contamination-leafy-greens-implicated-fall-2020-outbreak-e-coli
Viazis S, et al. An overview of farm investigation findings associated with outbreaks of Shiga toxin-producing Escherichia coli infections linked to leafy greens: 2009–2021. Journal of Food Protection 88:100542 (2025). https://doi.org/10.1016/j.jfp.2025.100542
FDA. 2020 Leafy Greens STEC Action Plan. https://www.fda.gov/media/135855/download
FDA. California Longitudinal Study (2020–present). https://www.fda.gov/food/environmental-studies/california-longitudinal-study-2020-present
Recent Comments