Our grant from the Water Foundation gave me an excuse to dive deep into a big statewide dataset collected by the Iowa DNR. I presented some of that analysis (along with general tips for data analysis) at the Iowa Water Summit in October. Since my slides rarely make sense without the narration, I have also written up a series of set of three case studies about trend monitoring.
I started learning R, a computer programming language for data science and statistics, in 2020. There was a steep learning curve, but now I get to coast downhill. If I can make an interactive map to display which chloride and phosphate readings from a volunteer event fall into the “good” category, I can reuse a lot of that code to make another interactive map that shows which lakes meet the recreation standard for E. coli. If I can figure out how to overlay maps of watersheds and animal feeding operations to calculate livestock density for 60 monitoring sites, it’s not hard to adapt that to calculate a similar metric for wastewater treatment plants. If I can make a good graph for one site, it’s not hard to scale that up to 50 sites.
Behold! My finest graph yet! On a single page, you can see 20 years of monthly nitrate data for the 48 sites in Iowa’s Ambient Stream Monitoring network with the longest record. The dots are color-coded to show streamflow at the nearest gage on the day the water samples were collected, yellow when water levels were at their lowest and grading to purple when water levels were at their highest. The black line is a 3-year moving average (or at least a close approximation with LOESS). You’ll notice that nitrate levels in most streams shot up sometime around 2014 and have been declining since. But how much of that is a precipitation-related trend tied to the El Niño-Southern Oscillation and how much is due to conservation efforts in the watershed?
Here comes the magic trick! Watch closely, for when I place into my hat all those samples collected when streamflow was lower than average (yellow and green dots) some of these apparent nitrate trends will…. disappear!
Darn it, that didn’t work at all! For the third of three case studies I’ll link to below, I take a closer look at two of these sites: Black Hawk Creek, where I’d expect a moderate improvement based on cover crop acreage and the East Fork of the Des Moines River, where I’d expect very little improvement. The East Fork actually has a much bigger nitrate trend than Black Hawk Creek and I can’t account for it. Maybe I should give up and leave trend monitoring to the experts at Iowa State University and Iowa DNR, they seem to know what they’re doing (see Danalatos et al. 2022).
However, the trick did work when I compared a watershed impacted by mostly point sources of pollution (i.e. sewage treatment plants) to a watershed impacted by mostly non-point source pollution (i.e. agricultural runoff). This is the first of the three case studies. Phosphorus concentrations have increased in the North Raccoon River and decreased in the South Raccoon has decreased, but that’s entirely due to recent drought. The second case study introduces a new metric for identifying rivers that a strongly influenced by point source pollution, and how to track improvement.
Iowa has a lot of hogs, poultry and cattle raised in concentrated animal feeding operations (CAFOs). They produce a lot of manure. However, CAFOs are not evenly distributed across the state, and it’s rarely practical to haul the manure long distances. Do rivers with more CAFOs in the watershed have worse water quality? I was curious and recently completed a big data analysis project to find out. I’ve omitted some of the technical details in the interests of making this article easier to read, but hope to eventually submit this research to a scientific journal for peer review. Get ready for a nuanced, data-driven look at the elephant in the room!
Livestock Density by Watershed
The water quality data for this study comes from 60 sites in Iowa DNR’s ambient stream monitoring network. (Two sites were later dropped because of incomplete data). For each monitoring site, I delineated a watershed (the land draining to that point) and overlaid databases of animal feeding operations. CAFO density in these watersheds varies greatly: from 12 animal units per square mile in Cedar Creek near Bussey, to 883 animal units per square mile in the Floyd River near Sioux City.
Animal units are a way of standardizing herd size across ages and species. For regulatory purposes, one 1000 pound steer is equivalent to 10 pigs under 55 pounds, 2.5 pigs over 55 pounds, 55 turkeys, or 82 layer hens. Feedlots with at least 300 animal units are tracked in Iowa’s database. Feedlots with 500 animal units require a manure management plan, and feedlots with 1000 animal units require a construction permit. The Iowa Environmental Council continues to follow and raise concerns about these rules.
Initial Findings and Complications
In the article that inspired this project, “The Fair, the Marginal, and the Ugly”, Chris Jones used this same dataset to rank water quality in Iowa’s rivers and noted that the river with the worst water quality has the most CAFOs. The Floyd River had the highest nitrogen and total phosphorus, the third highest turbidity, and the sixth worst E. coli. Sticking with the same time period (2016-2020) and similar metrics, I plotted water quality against livestock density for 58 sites to see if the Floyd River is part of a larger pattern. For nitrate, yes; for total phosphorus, maybe; for turbidity and E. coli, no. The relationship with turbidity is weakly negative; rivers with muddier water actually tend to have fewer CAFOs in the watershed.
The best explanation for this is that there is a third factor influencing both water quality and CAFO density: terrain. CAFOs are most common in flatter parts of the state where construction permits are more likely to be approved and there is plenty of cropland nearby to spread the manure. The notable exception to the pattern is Bloody Run, a trout stream in northeastern Iowa. In 2021, the Iowa DNR approved the construction of a 11,600 head cattle feedlot in this watershed, despite the steep terrain and abundant sinkholes. Given the timing, I am excluding this site from analysis and hope we do not have to find out what happens to water quality when this much manure is added to an environmentally sensitive area.
Primary drivers of water quality
To better understand the interactions of multiple variables without a lot of statistics, I like to color-code one of them (in this case, CAFO density) and then focus on a narrow range (in this case, watersheds with less than 160 animal units/square mile). You’ll see this technique several times in this article. This shows how slope and cropland in the watershed influence water quality, independent of CAFOs.
Slope: As you’d expect, turbidity in rivers is strongly correlated with the average slope of land in the watershed. Steep hills are more susceptible to runoff and erosion. Phosphorus and E. coli are also attached to sediment and carried by runoff, so are moderately correlated with turbidity, and weakly correlated with slope.
Cropland: Nitrate in rivers is strongly correlated with corn and soybean acres in the watershed. Long-term nitrate trends can also be explained by changes in cropping patterns (a replacement of hay and small grains with corn and soybeans). I’ve heard corn and soybeans described as a leaky system, and want to echo that. Whether the nitrogen comes from manure, ammonia, or soil organic matter, if you don’t have something green and growing in the early spring, you’re going to lose a lot of it.
Manure and Bacteria in the Water
E. coli is a bacteria found in the guts of birds and mammals, an easy-to-measure proxy for poop in the water and the pathogens that might come with it. For many environmentalists, the reason for Iowa’s long list of impaired waters seems frustratingly obvious. Hogs, poultry, and cattle outnumber humans, dogs, geese, raccoons, and deer, so they must be the main source of E. coli. Here’s an example of that kind of thinking from a report by the Environmental Integrity Project.
“Iowa is America’s hog capital – and also one of the most unhealthy areas in America to swim in rivers and streams. That’s in part because of the vast amounts of hog waste and farm runoff polluting the state’s waterways.”
The same logic showed up in the watershed management plan for Ioway Creek (and some others like it), which assessed likely bacteria sources based on the population of various kinds of animals and the amount of manure they excrete per day. While the consultants were careful not to say that hog confinements in Hamilton County were the main reason for chronically high E. coli in the creek, I sure got that impression from reading the maps and tables.
Looking at livestock populations turns out to be an unreliable way to guess which rivers will have bacterial impairments. Statewide, there is no correlation between E. coli in the river and livestock density in the watershed. The three worst rivers for E. coli in this dataset (the Soldier River near Pisgah, Maquoketa River near Maquoketa, and W. Nodaway River near Shambaugh) have less than 320 animal units per square mile, on the low side for Iowa.
More sophisticated models take into account the fraction of manure that reaches streams, how long it takes to get there, and how much of the bacteria dies off in the meantime. Unsewered communities, geese on the beach, raccoons in the storm sewer, and cows wading in the creek produce much less manure than animals in CAFOs, but a larger fraction of the manure is delivered directly to the water when it’s still fresh. That’s not to say that manure from CAFOs have no influence on E. coli in rivers. Once rivers with slopes steeper than 4% were excluded, the remaining sites had a moderate correlation between E. coli and livestock density.
Manure Nutrients in the Water
If manure is applied to fields that are not too steep and set back from streams, during appropriate weather conditions, and especially if the manure is knifed into the soil, very little of the solids, E. coli bacteria and pathogens in the manure should reach streams. The same is not true of the nutrients in the manure. Nutrients cycle between different forms, and the more readily dissolved forms (nitrate and orthophosphate) can easily leak out of the root zone during periods when crops aren’t growing, and make their way to streams.
Watersheds with a high density of CAFOs tend to have much higher nitrate concentrations, but most of that is because those watersheds also have a large proportion of the land in row crops. However, focusing on sites with at least 80% of the watershed in row crop production, there is still a positive correlation between livestock density and both nitrate and total phosphorus.
In the science assessment for the Iowa Nutrient Reduction Strategy, manure was not treated as a challenge for nutrient reduction, it was treated as a best management practice. It makes a certain amount of sense: manure is a slow-release fertilizer that adds organic matter to the soil. Compared to plots fertilized with commercial fertilizer, plots fertilized with swine manure had 4% less nitrate loss by 46% less phosphorus loss, mainly due to soil improvements that reduced the amount of runoff. However, those agronomic trials must have used a different set of application rates than usually occurs in practice. If you look at both commercial fertilizer sales and manure availability, counties with many CAFOs apply nitrogen and phosphorus at higher rates, with consequences for water quality. Here’s one study from Minnesota and another from Iowa that document this.
Closing thoughts
Prairie Rivers of Iowa has worked with some large swine and cattle producers who were early adopters of cover crops and who are very careful about how they manage manure. We salute their efforts to improve soil health and protect water quality. A study like this can only address the impacts of the industry as a whole.
This project was funded in part by a research grant from the Raccoon River Watershed Association, which has been monitoring water quality in Greene County. Last summer, the group watched with alarm as hog manure leaking from an earthen storage basin turned the water in a creek brown and caused the dissolved oxygen in the water to drop to zero. These kinds of incidents happen way too often, but usually affect a small stretch of stream for a short period of time, so don’t show up in monthly water quality datasets.
The correlations between water quality and livestock density disappeared entirely when I looked at two drought years (2021-2022). During dry periods, runoff and tile drainage from farmland is minimal, but effluent from sewage treatment plants and industry (including meatpacking plants) can have a bigger influence on water quality. Manure from CAFOs definitely impacts water quality in Iowa, but if we’re too quick to blame them in every situation, we may miss what’s really going on.
We know that weather influences water quality in Iowa’s rivers. Last year, there was a drought and nitrate was lower than usual. This spring, it’s been wetter and nitrate is higher than usual. If you monitor for 10 years and the first 5 are a little wetter or drier than the last five, you’ll a water quality trend to go with it. Boring!
What we really want to know is how people are influencing water quality. We can get a lot closer to that answer by peeling away the obvious weather-related patterns to reveal underlying trends.
In statistics, it’s called a covariate or an explanatory variable. If there’s a relationship between your water quality metric and some other thing you’re not really interested in (i.e. streamflow), you can model that relationship to account for part of a water quality trend over time. What’s left over might be the things you’re really interested in (i.e. how water quality has been affected by changes in crop rotations, conservation practices, sewage treatment, manure management, and drainage). It’s common enough in the scientific literature (Robert Hirsch’s Weighted Regression on Time, Discharge, and Season is a good example), but should be used more often for progress tracking at the watershed scale.
To illustrate this general approach, I downloaded daily nitrate data from three stations maintained by the US Geologic Survey. The sensors at the Turkey River at Garber and the Cedar River near Palo (north of Cedar Rapids) were installed in late 2012; the sensor Raccoon River near Jefferson was installed in 2008. I wanted a high frequency dataset (to minimize sampling error) that included the episodes of “weather whiplash” in 2013 and 2022.
“Residuals” are the difference between what we predict and what we measured. In the first panel, that’s the difference between a measurement and the long-term average. In the second and third panels, we see how nitrate measurements differ from what we’d expect given flow in the stream today, and flow in the stream last year. Gray dots – daily measurements. Red dots- yearly averages. Blue dotted line – trend. If I did this right, some of the dots should get closer to the middle.
Nitrate concentrations in rivers increase as the weather gets wetter and streamflow increases… up to a point. When rivers are running very high, there’s a dilution effect and nitrate concentrations fall. Based on that relationship, we can explain high nitrate levels in the Cedar River in 2016 (a wet year) and low nitrate levels in 2021 (a dry year).
Nitrate concentrations tends to be highest on wet spring days following a dry summer and fall, as nitrate that accumulated in the soil during the drought is flushed into drainage systems or washed off the land surface and into rivers. Here I’ve calculated a moving average of flow over the previous 365 days, and compared that to nitrate concentrations during high flow or low flow conditions. Based on that relationship, we can explain high nitrate in the Cedar River on wet days in the spring of 2013 and 2022 (following a dry year) and low nitrate on wet days in the spring of 2019 (following a wet year).
After making these adjustments, the downward trend in the Cedar River looks much smaller (0.53 mg/L per year, adjusted to 0.25) and is overtaken by the Turkey River (0.37 mg/L, adjusted to 0.28). The adjusted trends are statistically significant and could be attributed to conservation efforts in those watersheds.
How did I do this? For technical details, read here.
However, there’s still some weather-related patterns we haven’t accounted for. The Raccoon River near Jefferson also had a steep decline in nitrate since 2013 (1.42 mg/L per year, adjusted to 0.77 mg/L per year) but if you look at the entire record (going back to 2008), it’s part of an up-and-down cycle. I’ve seen that same pattern in the South Skunk River. The model explains some of those swings but doesn’t fully explain high nitrate in fall of 2014, spring of 2015 and spring of 2016. Perhaps the nitrogen that accumulated in the soil during the drought of 2012 took several years to flush out.
In addition to streamflow and last year’s weather (antecedent moisture is the technical term), nitrate can be explained by season, soybean acreage, and baseflow. If it’s not enough to know that water quality is improving or getting worse, and you’d also like to know why, then let’s peel that onion!
Later this month, we are releasing a report with the findings from Story County’s 2021 water monitoring season.
In some ways, 2021 was an unlucky year to launch a water quality monitoring program. Story County was in drought conditions for much of the year, and smaller streams were frequently dry when we did our monitoring routes.
In some ways, it was an ideal year to launch a monitoring program, because weather always has an influence of water quality and the challenging conditions in 2021 forced us to better account for it.
For the report, this means asking a simple question: “was there enough water to float a canoe on the day you sampled?”
When the South Skunk River is too low for paddling:
Not much water (and not much nitrogen and phosphorus) reaches the Gulf
“Hot spots” for nitrogen and phosphorus are below wastewater treatment plants
When the South Skunk River rises high enough for paddling:
“Hot spots” for nitrogen are in the Headwaters of the South Skunk River Watershed upstream of Ames, as shown in the graph
E. coli levels upstream of Ames (and Ioway Creek) get worse but still meet the standard
E. coli levels downstream of Ames (and Ioway Creek) get better, but still exceed the standard
If I had less curiosity and more sense, I would have written a short report: “great job everyone! We collected a lot of data. Here it is! It’s possible that drought had an influence on water quality.” This was more work, but I hope you get more out of it.
Before state wastewater standards went into effect in the 1960s, raw sewage could flow directly to a stream without treatment. Despite the standards, this continues in many areas today. In areas called “unsewered communities,” outdated or poorly functioning septic tanks still allow untreated wastewater into our waters. The Iowa DNR works with these communities to find funding sources and alternative treatment systems and to allow adequate time to upgrade the systems.
The Governor has announced that additional funding through the infrastructure bill that will be available to help unsewered communities upgrade their systems. Could this make a big difference for water quality in Iowa? Statewide, I’m not sure, but I’ve taken a closer look at the Iowa River Basin upstream of Marshalltown, where we know of 11 unsewered communities. Based on my first look at the data, it appears that these communities have little influence on E. coli in the Iowa River itself, but could make a difference for water quality in tributary streams like Beaver Creek in Hardin County.
There are 11 unsewered communities in the upper part of the Iowa River Basin, marked here with yellow circles with an X.
A Water Quality Improvement Plan for E. coli bacteria in the Iowa River Basin was released by Iowa DNR in 2017. As required by the Clean Water Act, these kinds of plans include a Total Maximum Daily Load (TMDL) of pollutants that a water body could handle and still meet water quality standards. Author James Hallmark compares this pollution budget to a family budget: regulated point sources are your fixed bills, non-point sources are your variable expenses, and the margin of safety is your emergency fund. I like this analogy and would add that without some understanding of where your discretionary spending is going, and a realistic strategy to reign it in, you’re probably not going to achieve your goals.
The Water Quality Improvement Plan includes a comprehensive list of E. coli sources but doesn’t single any of them out as being particularly important. It includes a list of potential solutions, but it doesn’t identify which of those would make the most difference. That’s a job for a Watershed Management Plan written with stakeholder input, apparently. However, the document is chock-full of load-duration curves, which I wrote about previously. We can use the information in these charts and tables to take the next step and begin to narrow down where and when the pollution is most serious!
In this article, I won’t pay much attention to “High Flows” and “Low Flows” because there wouldn’t be much recreational use under these conditions. I also don’t look at “mid-range” flows because there’s a bigger mix of sources influencing water quality in these conditions. A closer look at the other two categories is revealing.
If houses are discharging raw sewage directly into a stream, we’d expect to see the highest E. coli concentrations when the stream is running lower than normal, and there’s less dilution. This is indeed what we see in Beaver Creek in Hardin County, which is downstream from the unsewered community of Owasa. Beaver Creek would need a 79% reduction in E. coli load to meet the primary contact recreation standard during “Dry Conditions” and a 38% reduction during “Wet Conditions”.
If not fully treated, sewage could be a major contributor to E. coli in some tributaries of the Iowa River.
Treated sewage also has the biggest influence when streams are lower than usual. The upper reaches of the South Fork receive effluent from the small towns of Williams and Alden, which have waste stabilization lagoons. It’s likely that some bacteria makes it through the treatment process, and this would explain why E. coli is higher during “Dry Conditions” (needing a 73% reduction) than during “Wet Conditions” (needing a 30% reduction). When their permits come up for renewal, Iowa DNR could require a UV disinfection system to ensure that E. coli in effluent is no greater than 126 colonies/100mL.
The blue line is the wasteload allocation–the regulated part of the pollution budget. Even with the best available treatment, wastewater from two towns has a big influence on the South Fork during dry conditions.
In a watershed with few people and many hogs, we’d expect to see the highest E. coli concentrations when the streams are running high and runoff from fields that receive manure application is more likely. This is indeed what we see in Tipton Creek in Hardin County, a watershed containing 47(!) CAFOs, but the levels are not especially high compared to other sites in the Iowa River basin. The recreation standard is met during “Dry Conditions” and would need a 36% reduction during “Wet Conditions.” Handled correctly (applied to flat ground at the right time, and preferably incorporated into the soil), manure and the microbes it contains can be kept out of streams. Preventing loss of the nutrients in manure is a more difficult challenge—nitrate concentrations in Tipton Creek often exceed 20 mg/L!
Despite there being a lot of hogs in the Tipton Creek watershed, E. coli levels are not especially high, relative to downstream locations.
It’s not clear to me whether primary contact recreational use of these streams is a relevant or attainable goal, or whether we should be calibrating our level of concern to the secondary contact recreation criteria. Unless there’s a permit holder affected, IDNR doesn’t investigate whether there’s enough water for kayaking in Tipton Creek, or whether children play in Beaver Creek, so the designated use is presumptive and tells me nothing.
E. coli and recreation on the Iowa River is not as big a concern at Crystal Lake as it is at Steamboat Rock. Photo Credits: Ryan Adams, photojournalist
To protect fishing, paddling, and children’s play on the Iowa River itself, where and when should we focus? The Iowa River at Marshalltown needs a 60% reduction in bacteria load to meet the recreation standard during “Wet Conditions” (10-40% flow exceedance). However, it actually meets the primary contact recreation standard during “Dry Conditions” (60-90% flow exceedance). Focusing on unsewered communities in the watershed would NOT be an effective way to address this impairment.
Beaver Creek (left) has worse E. coli when it’s dry. The Iowa River near Marshalltown (right) has worse E. coli when it’s wet. If the green line is above the red line, that indicates that the E. coli geometric mean for that range of flows exceeds the standard.
Galls Creek in Hancock County has some of the worst E. coli levels measured in the basin, and would have a larger per-acre benefit to the Iowa River if standards could be met. Galls Creek has no unsewered communities but at least 20 farmsteads located along the creek that could have issues with septic systems overflowing under wet weather. The watershed has little woodland and no pasture, so land application of manure from the several CAFOs in the watershed would be most likely animal source of E. coli.
Table by Prairie Rivers of Iowa, using information from the Water Quality Improvement Plan for the Iowa River Basin
This is just a partial review of one of three HUC8s in the Iowa River Basin. There is much more to learn from further discussion with people who know the area well, or from on-site investigation. However, I hope I’ve demonstrated how we might squeeze some more insight out of the data we have, in order to make smart investments in water quality.