GEOSPATIAL ANALYSIS OF DENITRIFYING SEPTIC SYSTEMS AND GROUNDWATER NITRATE CONCENTRATIONS IN JAMESTOWN SHORES, RHODE ISLAND

High-density housing areas with onsite wastewater treatment systems (OWTS) and domestic drinking water wells are susceptible to groundwater contamination from nitrate-nitrogen (NO3-N). One solution is to install denitrifying OWTS, which are designed to reduce the effluent nitrogen by approximately half that of conventional OWTS. Geostatistical methods were used to analyze groundwater NO3-N data from the Jamestown Shores neighborhood of Jamestown, Rhode Island to determine if denitrifying OWTS have had an affect on water quality. Temporal trends were analyzed using NO3-N concentrations from sample events approximately 15 years apart, 1996-1997 and 2010-2011, between which a number of denitrifying OWTS were installed. Spatial trends in the effects of housing density, percentage of denitrifying OWTS, and select confounding variables on groundwater NO3-N concentrations were analyzed using directional buffers for groundwater flow and fracture orientation as well as circle buffers. Regionally, groundwater NO3-N did not decrease from 1996-1997 to 2010-2011, likely because of a net increase in houses and septic systems during this time period. Although the statistical significance of each buffer-type varied, groundwater NO3-N was generally found to increase with housing density in the immediate surrounding area and along the fracture orientation and decrease with at least one denitrifying OWTS in the area. Well depth and relative soil permeability (both normalized by housing density) did not have a statistically significant affect on NO3-N concentrations; the sample size for the low permeability soils was too small to statistically analyze, but the NO3-N concentrations were considerably less than for the other soils. Expected groundwater NO3-N concentrations were determined using estimated nitrogen loading from area OWTS, compared with measured concentrations, and a prediction model developed for the effects of increasing percentage of denitrifying OWTS. The expected vs. measured comparison model showed there are some sites with low NO3-N concentrations that do not appear to be affected by the high density of OWTS and some with high NO3-N concentrations above the level predicted by the density of OWTS alone. The prediction model showed that the percentage of denitrifying OWTS needed in the surrounding 400-foot radius to achieve NO3-N concentrations below the action level (5 mg/L) is at least 75% in the highest density areas (3.1 houses/acre) and at least 25% in the average density areas (1.7 houses/acre). The town can use these two models for planning purposes to determine where denitrifying OWTS may be most effective or where confounding variables may have a more significant influence on NO3-N concentrations.


LIST OF TABLES
found to be effective at reducing groundwater NO 3 -N, they can be a viable community-planning tool for groundwater restoration and long-term protection.
This study used geostatistical methods to analyze the effects of denitrifying OWTS on groundwater NO 3 -N concentrations in a high-density housing area of Jamestown, Rhode Island called "Jamestown Shores". Temporal analysis was conducted on groundwater NO 3 -N samples taken approximately 15 years apart, between which a number of denitrifying OWTS were installed. Spatial analysis was performed to determine the relationship between groundwater NO 3 -N concentrations and the following parameters: housing density, percent denitrifying OWTS, and select confounding variables. This was accomplished by creating and utilizing buffers around each sample site, including wedge-shaped buffers encompassing the dominant groundwater flow direction and fracture orientation, in addition to circle buffers.

GROUNDWATER NO 3 -N CONTAMINATION FROM OWTS
Nitrate is an inorganic compound that is naturally present in groundwater at low concentrations, generally less than 1 mg/L NO 3 -N (Nolan and Hitt, 2003;USGS, 1999). Groundwater concentrations greater than this relative background level are primarily from septic systems, animal waste, and/or fertilizers (Canter, 1997). These higher levels can be detrimental to human and ecological health: a potential contributor to methemoglobinemia ("blue baby syndrome") in infants (Comly, 1945;Fewtrell, 2004;Knobeloch, et al., 2000;U.S. EPA, 2002;WHO, 2008) and to eutrophication of surface water bodies (EPA, 2002). Because the presence of NO 3 -N above background levels is evidence of a pathway from the surface/subsurface to the groundwater, it can also be an indicator of pathogens and other harmful chemicals such as those in pharmaceuticals and personal care products (PPCPs) and pesticides.
OWTS are often a major source of elevated NO 3 -N in groundwater because conventional OWTS are not designed to remove nitrogen (Canter and Knox, 1985;URI, 2005;U.S. EPA, 2002), nitrogen can travel long distances in aquifers due to insufficient dispersion and denitrification (Canter and Knox, 1985;Robertson and Cherry, 1991), and OWTS density is often too high to allow adequate dilution before reaching wells (Bicki and Brown, 1991;Drake and Bauder, 2005;Horn and Harter, 2011;Lowe, et al., 2000;Persky, 1986;Veeger et al., 1997;Yates, 1985). A potential solution applied by many communities is the installation of denitrifying OWTS.   , 1997 andMichaud, 1998).
from OWTS (also Joubert, 2003); and there is a statistically significant increase in NO 3 -N concentrations with decreasing lot size, specifically, smaller than one acre (0.4 hectare). Although the majority of the groundwater NO 3 -N was determined to be from OWTS, lawn fertilizer and pet waste may also be contributors. Agriculture is not a significant land use in this area other than a few small areas of "pasture" ( Addressing the ground water contamination problem by extending the public water supply is not an option because the existing public supply system is at capacity; neither is sewering the area because the OWTS recharge the aquifer and hence help prevent salt-water intrusion (Veeger, et al., 1997). Denitrifying OWTS may therefore be the best long-term solution if they are effective at reducing the groundwater NO 3 -N at a given level of implementation. ability of the aquifer, comparable to that of silty sand (Freeze and Cherry, 1979). The water table on Conanicut Island is generally within the till layer, shallow during the wet season, from 0 to 13 feet (0 to 4 meters) below land surface, declining as low as 30 feet (8.3 meters) in late summer (GZA, 1986 andVeeger et al., 1997). Saltwater intrusion is a risk in some areas of the island because the freshwater/saltwater interface is shallow; under non-pumping conditions it is estimated to range from more than 500 feet in the center of the island to just a few tens of feet near the coast (Veeger et al.,

WATER QUALITY DATA
Existing water quality data were used for this study. Groundwater samples were

GIS ANALYSIS
ESRI's ArcGIS 10.1 was used to compile and analyze data. Buffers were created around the NO 3 -N sample sites to analyze the effects of housing density, percent denitrifying OWTS, and confounding variables on the groundwater NO 3 -N result. The types of buffers are described in Table 2. The circle buffers were created for all of Northern Jamestown, in addition to Jamestown Shores, to compare with Sandorf's 1999 study by using housing density instead of lot size. The wedge buffers were created with the point at, and including, the sample site. The 400-foot (122meter) radius was chosen because this is a minimum setback distance requirement for public drinking water wells (RIDEM, 2012) and is considered to cover the area of highest risk to water quality. Although the wells in the Jamestown Shores area are private, not public, the density is high which could cause interference, increasing the size of the capture zones. The 600-foot (183-meter) radius accounts for the fact that groundwater NO 3 -N has been found to travel long distances (Canter and Knox, 1985;Robertson and Cherry, 1991), and that Donohue (2013) found similar results for both buffer sizes in the coastal town of Charlestown, RI. Buffer radii larger than 600 feet (183 meters) covered significantly more area outside of Jamestown Shores and were therefore not as representative of the area of interest.  -1997-2011-1997Circle (600 ft/183 m) 1996-1997-2011-1997

1) H d = #houses/Area 2) DOWTS(%) = #DOWTS/TOWTS (where Hd is housing density, DOWTS is denitrifying OWTS and TOWTS is total OWTS)
For housing density, the 1996-1997 dataset was analyzed using only the houses with built dates prior to 1997, and the 2010-2011 dataset included all of the houses.

STATISTICAL ANALYSIS
Groundwater NO 3 -N concentrations in Jamestown Shores were statistically analyzed temporally and spatially with Microsoft Excel 2008 and SAS 9.2.
Groundwater NO 3 -N results that were "non-detect" (ND) were changed to 0.5 mg/L for statistical analysis, as were any results less than 0.5 mg/L, because this is the highest minimum detection limit of all sample events. NO 3 -N results from sites sampled in both 1996 and 1997 or both 2010 and 2011 were averaged. All statistical hypothesis tests were conducted at the 95% confidence level; therefore, the term "significant" refers to a p-value less than 0.05, except where explicitly stated. The statistical tests and analysis parameters are listed in Tables 3 and 4 below. Because the  data did not appear to be well described by a normal distribution, and some of the data were "censored" (below the lab minimum detection limit), nonparametric statistics were used for most analyses (Helsel and Hirsch, 2002). In all statistical analyses except for housing density, NO3-N was normalized (divided by) housing density because the amount of nitrogen loading from homes in each buffer is expected to be the largest driver of groundwater NO3-N concentration in wells.

Temporal Trend Analysis
Before evaluating trends between the two temporal datasets (1996-1997 vs. 2010-2011), a statistical test was conducted to determine seasonal or short-term variability, i.e. if there is a significant relationship between the repeat groundwater NO 3 -N results in each pair of consecutive years (1996 vs. 1997 and 2010 vs. 2011). This repeatability was graphically analyzed using Bland-Altman Plots, which allowed a visual evaluation of variation against a horizontal zero-line (Bland and Altman, 1986). The temporal trends were then analyzed with the statistical tests listed in Table 3.

Spatial Trend Analysis -Housing Density and Percent Denitrifying OWTS
The groundwater NO 3 -N concentration of each sample site was compared with the housing density in its corresponding buffer using the statistical tests listed in Table   3. The 1996-1997 and 2010-2011 datasets were analyzed for housing density, but only the 2010-2011 dataset was analyzed for percent denitrifying OWTS because the denitrifying OWTS did not exist in 1996-1997.

Spatial Trend Analysis -Relative Soil Permeability and Well Depth
The relative soil permeability of high, moderate, and low was designated based   dataset, and the difference between the two mapped in GIS.

RESULTS AND DISCUSSION
The descriptive statistics for groundwater NO 3 -N concentrations in Jamestown Shores are shown in Table 5, including the percentage of results above the thresholds   2010-2011 repeats varied significantly, some by an order of magnitude, which is possibly due to sampling/analysis error or an influx of NO 3 -N due to precipitation/irrigation conditions and/or fertilizer application. The 1996 and 1997 datasets were both sampled summer through fall; however, 2010 was sampled in the fall and 2011 in the spring, which could capture seasonal changes. Despite some variability, the Pearson Correlation Coefficient was found to be significant for both consecutive-year datasets (Appendix B), which indicates seasonal or short-term variability is not statistically significant area-wide and we can have more confidence any water quality changes in results 15 years apart are due to anthropogenic modifications and not because of seasonal effects. This is consistent with a study performed in Oregon, which found considerable intra-well variability due to recharge events over the course of 15 months, but no statistically significant temporal variability area-wide, and attributed these findings to spatial heterogeneity in the subsurface as well as in land use (Mutti, 2007).

Temporal Trend Between the 15-Year Datasets
The

Travel Time
The response of well water quality to changes in land use is dependent on the age of the groundwater and the timing of the land use change (McMahon, et al., 2008;Rupert, 2008). The age of the well water in the Jamestown Shores wells is unknown, and the land use change was gradual over the 15 years. Figure 6 shows the annual percentage of the denitrifying OWTS installations between the two sampling periods.
Travel time calculations are often used to determine how long it would take for a contaminant to move through the aquifer; however, the pumping of almost 1,000 wells in this area means much of the groundwater could be cycling through the domestic system instead of traveling away from the area. This, along with the fractured nature of the aquifer, and heterogeneous nature of the subsurface, makes an estimate of travel time difficult. It can be argued that a reduction in groundwater NO 3 -N concentration is not seen because not enough time has passed for the denitrifying OWTS to have an effect; however, a discussion on how confounding variables can also affect this result is included later in the report.

Housing Density
The buffers used for spatial analysis are shown in Figure 7. In Jamestown Shores there is a more statistically significant increase in groundwater NO 3 -N concentrations with housing density in the 400-foot (122-meter) radius circle and fracture buffers than in the 600-foot (183-meter) radius circle buffer, and no statistical significance in the up-groundwater gradient buffer (Figures 8 and 9 for a trend that is statistically significant and a trend that is not statistically significant, also Table 6 and Appendix B). Also, these trends are more significant in the 1996-1997 dataset than the 2010-2011 dataset, possibly because of the variability of the data between the 2010 and 2011 datasets.
As shown on Table 6, the significance occurs both at less than vs. greater than 1 house/acre (1 house/0.4 hectares) and less than vs. greater than 2 houses/acre (2 houses/0.4 hectares) for some buffers and at one or the other for other buffers. This agrees with the groundwater NO 3 -N analysis performed previously in this area (Sandorf, 1999), which determined a significant difference in NO 3 -N between lot sizes less than 1 acre (0.4 hectares) and those greater than 1 acre (0.4 hectares), although Sandorf's study was more general in its analysis of lot sizes, whereas this study looked   in more detail at housing density surrounding each sample site. A recent study in Charlestown, RI found that lot sizes greater than 0.67 acres (0.27 hectares) (less than 1.5 houses/acre or 1.5 houses/0.4 hectares) were needed for NO 3 -N concentrations below the 5 mg/L action level (Donohue, 2013). The geology in Charlestown is different however, with wells terminating in either unconsolidated overburden or fractured bedrock, whereas all the wells in this study terminate in fractured bedrock.
Groundwater generally flows perpendicular to water table contours; therefore, the lack of statistical significance with the up-groundwater gradient buffer was an unexpected finding. One reason for the insignificance could be that many wells and fractures in a small area could interfere and change local flow directions. The fracture orientation may be a more dominant driver of groundwater flow in this setting than the water table gradient. In addition, localized zones of contribution may be established around some wells if the pumping rate of the well is greater than the rate of groundwater flow (Ceric and Haitjema, 2005), as suggested by the significance of the 400-ft (122-meter) circle buffers.
In Jamestown Shores, the 600-ft (193-meter) circle buffers were only found to be significant for less than or greater than 2 houses/acre (2 houses/0.4 hectares) in 2010-2011 and not at all in 1996-1997, although there is significance at the 90% confidence level. Given the significance of the 400-ft (122-meter) circle buffer, this is likely because a larger buffer is more likely to cover a variety of housing density patterns, including low-density areas along the edges of Jamestown Shores. This indicates the groundwater NO 3 -N at an individual well is controlled by more local conditions. Normalized groundwater NO 3 -N was found, by two out of three statistical tests, to improve (have a significant decreasing trend) with percent denitrifying OWTS in the up-groundwater gradient buffer (Figures 11 and 12 for a trend that is statistically significant and a trend that is not statistically significant, also    denitrifying systems within the buffers in the north to south fracture direction. As with housing density, normalized NO 3 -N was not found to have a significant trend with percent denitrifying OWTS in 600-ft (183-meter) circle buffer. These results imply that although the trend is not as significant as housing density, having at least one denitrifying septic system in the surrounding area within 400 feet (122 meters) positively correlates with improved groundwater NO 3 -N concentrations.

Percent Denitrifying OWTS
For both the housing density and percent denitrifying OWTS Kruskal-Wallis tests, all bins meet the minimum recommended sample size of n = 5; however in many cases there was great variability in sample sizes between the bins. For example, although a statistically significant improvement in normalized NO 3 -N was found between less than and greater than 50% denitrifying OWTS within the north to south fracture buffers, there were 139 buffers with less than 50% denitrifying OWTS and only six with greater than 50%. The sample sizes are included on the box plots in the report and in Appendix B.

Measured vs. Expected NO 3 -N Concentrations
To  Figure 13. The Pearson's Correlation was found to be statistically significant ( Figure 14A), although the Bland-Altman Plot ( Figure 14B) shows a trend wherein the expected values over-estimate concentrations in low housing-density areas and underestimate concentrations in high housing-density areas. Despite the significant correlation overall, there are many sites with large differences which implies confounding variables play a greater role in these areas or, in the case of a low measured concentration compared to expected concentration, the groundwater is not strongly influenced by anthropogenic activity.  replaced with denitrifying OWTS, the measured concentration would drop from above the action level to below, but at 25% to 50% it would still be over. Figure 15 shows what should be expected at the highest, median, and lowest density areas: at the most dense buffer area, the expected NO 3 -N concentration will only decrease less than the action level if 75% of houses had denitrifying OWTS. The median density areas can achieve this with 25% denitrifying OWTS and the lowest density should be below background concentrations at all times. The distribution of density and the percentage of denitrifying OWTS required to achieve groundwater NO 3 -N concentrations below the action level is shown in Figure 16. Figure 15. Change in expected NO 3 -N concentration with percent denitrifying OWTS for the maximum, median, and minimum housing densities within the 400-ft radius circle buffers in Jamestown Shores, Jamestown, RI. Given a buffer area of 11.54 acres, 3.1 houses/acre is 36 houses, 1.7 houses/acre is 20 houses, and 0.3 houses/acre is 3 houses. No other sources of NO 3 -N are considered. Figure 16. Percentage of denitrifying OWTS required within a 400-foot radius to achieve groundwater NO 3 -N concentrations less than the action level of 5 mg/L. Zero to 1.4 houses/acre require no denitrifying OWTS, 1.5 to 1.9 require 25%, 2.0 to 2.3 require 50%, and 2.4 to 3.1 require 75%. Also, the lightest areas, <0.3 houses/acre, are those where groundwater NO 3 -N should be less than the background concentration (1 mg/L) even without denitrifying OWTS.

Soil Permeability
Soil is the first line of defense for nitrogen degradation once it leaves the OWTS. Lower soil permeability increases the residence time and improves the potential for denitrifying conditions under anoxic conditions; therefore groundwater NO 3 -N concentrations in relatively low permeability soils are expected to be lower than in relatively high permeability soils (Nolan, 2001). In addition, the presence of a restrictive (low permeability) layer near the surface could prevent recharge to groundwater, instead creating runoff to other areas or surface water bodies (Art Gold, personal communication). The assignment of relative soil permeability in Jamestown Shores is categorized in Appendix A and the distribution is shown in Fig. 17. Few residences are located in the hydric (low permeability) soils because a shallow water normally not granted in these areas. At first glance, groundwater NO 3 -N was found to have a significant increasing trend from low to high permeability in both datasets ( Figure 18). Although the Low bin is too small to statistically analyze (n=4), there is still a statistically significant trend when combining the Low and Moderate bins and comparing them with the High bin. After normalizing groundwater NO 3 -N by housing density, however, a significant relationship was not found (Figure 19). Although not considered statistically significant because of the small sample size of the Low bin, it is worth noting that the NO 3 -N concentrations in the wells underlying the Low soils were considerably lower than for the other soils; however, the difference between the original and normalized analysis implies that housing density is a stronger driver of NO 3 -N concentration than relative soil permeability. These results are similar to the findings of the 1999 USGS study of public-supply wells in Rhode Island (USGS, 1999), which found that, within the wellhead protection areas, soils with high permeability or high leaching potential do not significantly correlate with elevated nitrate concentrations.

Well Depth
A relatively deep well is expected to have lower groundwater NO 3 -N concentrations than a shallower well because the longer travel time could provide more opportunity for dilution and degradation (Katz et al., 2011;Lichtenberg and Shapiro, 1997;Rupert, 2008). Estimated well depths were provided by homeowners for 55% of the sample sites in 1996-1997 and 51% of the sample sites in 2010-2011.
The distribution of these depths can be seen in Appendix C. As shown in Figure 20 and 21, groundwater NO 3 -N appears to have a decreasing trend with well depth, but it is not significant for total or normalized NO 3 -N concentrations. This lack of significance is also seen when comparing groundwater NO 3 -N with a continuous well depth variable; the p-values for the Spearman's Correlations are greater than 0.05 and the confidence limits for the Kendall-Theil Robust Lines include a slope of zero (Appendix B). This is reasonable given the amount of variability in the results, which can be seen in the long error bars and the difference between the mean and median symbols. An explanation may be that well depth in uncased, fractured bedrock wells is often not a strong proxy for the depth of the actual water-bearing zone. Well drillers generally drill significantly beyond the water-bearing fractures to provide storage capacity in the well bore, particularly for low-yield wells. In addition, well depths were provided by homeowners, without source (e.g. well log) documentation. These findings are also similar to the 1999 USGS study of public-supply wells in Rhode Island (USGS, 1999), which found that well depth does not significantly correlate with elevated NO 3 -N concentrations.

CHAPTER 4 CONCLUSIONS AND RECOMMENDATIONS
The groundwater NO 3 -N concentrations in Jamestown Shores, Jamestown, RI have not decreased area-wide between 1997 and 2011 despite the installation of 112 denitrifying OWTS. The temporal trend shows a small increase, which is statistically significant at the 90% confident level. This is likely because there was a net increase in OWTS during this time period.
Spatially, the groundwater NO 3 -N positively correlates with housing density within a 400-foot radius as well as along the dominant fracture orientation within 400 feet; but not along the dominant groundwater flow direction (400 feet up-gradient) or within a 600-foot radius. NO 3 -N concentrations increased significantly from a housing density of less than 1 house/acre (1-acre lot) to 1 houses/acre or greater and again from less than 2 houses/acre (0.5-acre lots) to 2 houses/acre or greater. In addition, having at least one denitrifying OTWS within a 400-foot radius decreases the groundwater NO 3 -N concentration when normalized by housing density. Relative soil permeability showed no statistical significance with NO 3 -N normalized by housing density; the sample size for the low permeability soils was too small to statistically analyze, but the NO 3 -N concentrations were considerably less than for the other soils. Well depth showed no statistical significance with NO 3 -N normalized by housing density. Other confounding variables could contribute to elevated or lower NO 3 -N concentrations, but were not analyzed due to insufficient data.
Based on OWTS nitrogen loading calculations, the expected NO 3 -N concentration for Jamestown Shores area-wide is 3.9 mg/L, which is close to the observed median concentration of 4.0 mg/L. The expected concentration at each well did not differ significantly from the measured concentration on a regional level, but there were several local differences.