The Stormwater Blogs

How hard would it be to learn if river phosphorus concentrations are dropping in response to a new city ordinance restricting the sale and use of lawn fertilizers containing phosphorus? That is the question posed to me about a year ago by the environmental coordinator of the city of Ann Arbor, Michigan. Like several other communities in Michigan, as well as in New Jersey, Wisconsin, Florida, and entire states including Maine and Minnesota, the city council had voted to take a step that they thought would be good for the environment. They hoped to limit the amount of phosphate runoff from residential properties and maybe curtail eutrophication of the scenic Huron River.

It so happened that a student, Julie Ferris, and I were putting the final touches on a scientific study that we were planning to publish in Lake and Reservoir Management, the journal of the North American Lake Management Society (NALMS). As part of that work we had been examining the statistical properties of a water quality dataset that I collected for the Huron River from 2003 to 2005. I used the data to calculate nutrient loads for an EPA-sponsored research project on some downstream lakes, but we had become curious about the magnitude of the “ordinary variability” that we could expect from year to year and month to month. The data had an unusual high degree of temporal resolution, with measurements at weekly and sub-weekly scales. And we were focusing on a subset of the data that included the city of Ann Arbor and points upstream of it (Figure 1).

Figure 1. The Huron River of southeastern Michigan. The study region is enclosed in a rectangle.

“How big a change were you expecting to get from this ordinance?” I asked. “About 22%,” was the reply. Now this was a question that actually has an answer, or many answers depending on how you construct the statistical model. The trick is in how you balance Type I and Type II error.

Recall that Type I error occurs when you mistakenly accept a hypothesis as true when it is really false. Type II error happens if you reject a valid conclusion because you think it is false. There are big sample size and effort problems with trying to minimize both types of error simultaneously. Besides, in this case the historical data were what they were, and we could not go back in time and collect more. So we set Type II error to 75% (chance of detecting a decrease if it is real) and Type I to 10% (chance of thinking the effect is real when it is not).

Three operational measures of phosphorus (P) were at issue. The first is total phosphorus (TP), the total mass of P in all forms: dissolved, colloidal, and particulate. The second is total “dissolved” phosphorus (DP), defined as the phosphorus in filtrate that has been passed in our case through a filter with 0.45 micrometer aperture size. The final is “soluble” reactive phosphorus (SRP), or the amount of P in the filtrate that can be measured by reacting it with molybdate ion in an acid solution, but without chemically digesting all the organic compounds present.

Across the various sample sites in our historical data set we discovered that SRP was more variable than DP, which in turn was more variable than TP. Our model told us that the median time it would take to detect a 25% change, collecting weekly data from May to September, was eight years for SRP, two to three years for DP, and one to two years for TP. We published the prediction (December 2008 issue of LRM).

A Test of the Prediction
While the paper waited in its publication queue, the city asked to fund a student to conduct a study under my supervision, since we were measuring nutrients all the time in the course of our lake research. I had to confess some skepticism, however. The city had reasonable control over what the lawn care businesses were applying and what was for sale in local stores, but nobody had a real clue about the extent of compliance, and that seemed a weak link. It was a chance to teach a student some useful analytical and statistical methods, though, and Doug Bell was enthusiastic about the opportunity.

We decided to approach the problem as a field experiment. Experiments need “controls.” We selected two kinds of controls: a control site and control variables. The control site was the station labeled as “1” in Figure 1. It lies several miles upstream from the city limit of Ann Arbor and outside the jurisdiction of the city ordinance. Our experimental sites are labeled 5 and 6 in Figure 1. The first has about 29 square kilometers of drainage attributable to Ann Arbor, and the second has about 94 square kilometers. We call these two sites A and B, respectively. Control variables were chemical properties that had nothing to do with phosphorus. We selected nitrate, silica, and colored dissolved organic matter (CDOM), a measure of humic acids in the water.

Thus there were three control variables and three response variables (SRP, DP, and TP); one control site; and two experimental sites. The statistical test was a simple t-test contrasting the 2008 data with the reference data stratified by month. All six of the water chemistry variables had lognormal frequency distributions, so they were log-transformed before analysis. The months tested were May to September.

Control Variables
There were no statistically significant differences in silica concentrations at any site for any month. CDOM was higher in 2008 than the reference period at both experimental sites only in the month of July. Otherwise, there were no differences. Nitrate was significantly different for two months each at all three of the sampling sites, one time higher, and one time lower. In short, the control variables had basically the same values in 2008 that they had from 2003 to 2005.

Phosphorus Variables
SRP behaved much like the control variables. There was no statistical evidence of altered concentrations at any site in any month. But given the variability of the reference data, we had predicted it would take eight years to see an effect of 25% magnitude. The prospects were better for DP and especially TP. Figures 2 and 3 show the findings for these response variables. First of all, there are no significant decreases in DP or TP at the control site for any month. For TP, however, there were statistically significant decreases at site B in four months out of the five, and there was a trend of decreasing concentrations at both sites for every month but September for A. DP also exhibited a trend of decreasing concentrations at site B every month, but the differences exceeded the level of statistical significance for only one month at each site.

Figure 2. Average concentrations of TP measured in 2008 expressed as percent of 2003-2005 values. * signifies that the reduction is statistically significant.

Site B, you recall, receives runoff from three times the city drainage area as site A. The odds of the DP levels at site B being less than the reference period for five months in a row are the same as those for flipping a coin and getting five heads in a row.

These results seemed worth sharing with professionals who may be contemplating the possible value of an ordinance like Ann Arbor’s, coupled with environmental education efforts, in their own communities. For the six statistically significant TP reductions flagged by asterisks in Figure 2, the average decrease was 31%.

Figure 3. The same as Figure 2, but showing results for DP.

It is possible to state objectively with a considerable degree of confidence that phosphorus concentrations were lower in 2008 at experimental sites compared with the reference period (2003 to 2005) and that the reductions were coincident with a city ordinance restricting use of lawn fertilizers containing phosphorus. It would be tempting to conclude that the phosphorus reductions were caused by implementation of the ordinance, and that may indeed be the case. However, we must bear in mind that the ordinance was enacted in the context of public education efforts that encourage citizens to be more mindful of yard waste discharges into storm drains, to exert more diligence regarding buffer strips of vegetation along stream banks, and to exhibit more environmental awareness in general. These multifaceted efforts make it difficult to isolate a single cause for the changes, but the changes appear to be real.