Naturally occurring radium (Ra) in home drinking-water wells in the Sandhills region of South Carolina, USA: Can high concentrations be predicted?

A strong relationship exists between concentrations of several principal constituents of water chemistry and naturally occurring radium in water from home wells in the large Sandhills region of the inner coastal plain of South Carolina, United States, an area of common radium problems (75% (21 of 28 of randomly selected wells in the main study area were above the U.S. Environmental Protection Agency (USEPA) maximum contaminant level (MCL): 5 picocuries per liter for 226Ra plus 228Ra). Ingested radioactive radium is potentially carcinogenic, and water far above the MCL occurs in places. One focus here was to determine if elevated radium concentrations can reliably be predicted using more easily measured characteristics of water chemistry. The initial phase investigated (1) concentrations of radium in well water within a smaller (~10 km radius) test area of known common problems plus (2) the associated water chemistry. This revealed a correlation between elevated radium and raised electrical conductance, total dissolved solids, calcium, nitrate, magnesium, hardness, and lowered pH. All are potentially useful screening tools for the many other rural home wells of this region. Natural groundwater in these highly leached sands is of very low dissolved salt content of all types, and local human influence on geochemistry is thus indicated. Predictive utility was examined by two additional series of wells (1) of higher radium contamination or (2) of elevated calcium concentration, both identified from file data. Higher radium wells were retested for water chemistry and higher calcium wells for radium content. Overall, a similar association between radium and water chemistry was observed. ©2017. The Authors.

Introduction

An estimated 700,000 people in South Carolina, United States, utilize private residential wells for their home drinking‐water supply [South Carolina Department of Health and Environmental Control (SCDHEC), 2014]. Naturally occurring dissolved radium (here 226Ra plus 228Ra, termed “combined radium”) has been detected at elevated concentrations (radioactivities) in groundwater from several regions of South Carolina and is a known human carcinogen. Once ingested, radium can become deposited within the body where its radioactivity presents a risk to human health by cell damage, with an overall increased risk of cancer [U.S. Environmental Protection Agency (USEPA), 2000; SCDHEC, 2010]. Health officials cannot feasibly test each home well within South Carolina directly for radium; thus, an effective screening method is sought, one which would allow both a reliable estimate of radium concentrations more quickly than direct radium analysis and at a substantially lower cost. Ideally, an effective screening program would evaluate home wells using more easily measured parameters found to be associated with higher radium concentrations. Identified at‐risk wells could then be tested directly for radium.

Solubility and migration of radium in groundwater are driven by a multitude of complex geochemical processes, including ion exchange with the aquifer material, leaching from the soil, and alpha‐recoil during isotopic decay of the parent isotope where the radium particle may actually be ejected due to the release of energy from the solid aquifer material into the liquid phase [Cape and Grundl, 2006; King et al., 1982; Almeida et al., 2004; Dafauti et al., 2007; Banner et al., 2001; Harmon and Ivanovich, 1992]. The naturally occurring isotopes 226Ra, an alpha particle emitter with a half‐life of approximately 1600 years, and 228Ra, a beta particle emitter with a much shorter half‐life of about 5.7 years, are individual byproducts from the decay series of the primordial radionuclides uranium (238U) and thorium (232Th), respectively [U.S. Environmental Protection Agency (USEPA), 2009]. In general, the sediments composing the aquifer are the immediate sources of radium in groundwater in the South Carolina coastal plain. Uranium‐bearing sediments, eroded originally from piedmont crystalline rock farther inland, may disperse dissolved uranium much more widely as a result of the increased solubility of the oxidized uranyl ion (UO2 2+) [Michel and Moore, 1980]. In contrast, thorium is much less mobile due to low solubility. With its short half‐life and a relatively slow groundwater movement, 228Ra distribution away from its source parent will be minimal; thus, the immediate sources to wells must lie relatively close [Michel and Moore, 1980; Almeida et al., 2004].

Results from controlled experiments by others indicate strong associations between leaching of radium from a soil matrix and the presence of other divalent cations such as calcium and magnesium [Dafauti et al., 2007]. A New Jersey coastal plain water‐quality study by Barringer et al. [1995] reported elevated concentrations of dissolved radium in the groundwater and attributed radium mobility to an increased competition for a limited number of cation exchange sites on the aquifer material by an abundance of similar divalent cations (i.e., Ca2+ and Mg2+). That study area, as in South Carolina, had a common use of dolomite (CaMg [CO3]2) in agricultural practice to reduce soil acidity, and this was cited as the primary source of introduced divalent cations to the aquifer. In other midcontinental U.S. aquifers, 226Ra levels were reported by Banner et al. [2001] to be highly correlated with total dissolved solids (TDS; R 2 = 0.92) and calcium (R 2 = 0.80).

Problem Identification

The South Carolina coastal plain is bounded on the north and west by the fall line (transitional area between the piedmont and coastal plain) and is further subdivided (for purposes of this paper) into an inner coastal plain (Sandhills) in the middle region of the state and an outer coastal plain along the Atlantic coast, as depicted (Figure 1) by Comstock et al. [2002]. The South Carolina Sandhills region includes all or parts of 17 counties, crossing from the Georgia border to North Carolina border and beyond in those states. It features highly leached quartz sands and kaolinitic clays with acidic soils and sediments. This region is where most of the naturally occurring dissolved radium has been detected in public wells at levels (radioactivities) in excess of the USEPA MCL for drinking water. Historical time series data for wells serving the Town of Leesville, South Carolina showed relatively constant elevated levels of combined radium through 1976–1978 [Michel and Moore, 1980]. These findings are of concern, especially for populations in rural areas that rely on home wells for their drinking water. These home wells typically are not tested for radium, due to high cost commercially and limited health agency capacity. Vast numbers of wells are involved, too many to test them all for radium. Most public supply wells are tested. Radium is relatively easy to remove routinely even in a home (e.g., by reverse osmosis), if one knows this is needed.

Figure 1

South Carolina physiographic provinces.

Hydrogeologic and Geochemical Setting

The entire coastal plain aquifer system in South Carolina is composed of several distinctive confined aquifers, which combined form a wedge of sand, clay, and highly permeable limestone sediments that thickens from the central region of the state southeastward toward the Atlantic coast [Aucott and Speiran, 1985]. The entire system lies above a basement of metamorphic, igneous, and some sedimentary (basin) rock. In general, drinking water in the inner Sandhills region of South Carolina is drawn from unconsolidated sand aquifers and commonly exhibits very low dissolved solids, low pH, and low electrical conductivity. The Middendorf Aquifer is one such nonmarine, sandy sediment, high in ferric iron (Fe3+), low in calcite (CaCO3), and outcrops along the fall line region, which is the major recharge area for the aquifer [Lee and Strickland, 1988; Bethke et al., 2006].

Methods

The objectives of this study were (1) investigate combined radium concentrations in randomly selected private residential drinking‐water wells from a primary area of investigation located within the Sandhills region, (2) conduct an in‐depth water‐chemistry analysis of each of these same wells, (3) evaluate the results for potentially useful covarying associations between radium and chemistry, and, (4) if possible, determine (a) what caused the chemistry to vary and (b) how radium levels were influenced geochemically. This first phase identified several chemical species that varied strongly with radium and potentially were useful as surrogate indicators or screening tools for other wells. As the second and third phases, these where then tested as predictors. In a known radium problem area ~60 km to the southwest of the main study area, 10 radium‐contaminated wells were resampled and tested both for radium (retesting) and for selected water chemistry constituents. Additionally, a wide scattering of Sandhills wells with known elevated calcium concentrations (from file data on private‐well analyses) were resampled and tested for combined radium and selected water chemistry constituents (retesting). Both tactics tested whether the potentially useful screening relationships from the initial limited‐area study are more widely applicable in the Sandhills region.

Site Selection

Through the three phases, a total of 57 wells located across three counties were sampled (November 2007 to February 2009) (Figure 2). The main study area (~10 km radius) encompassed n = 28 (wells #1–28) randomly chosen wells of unknown radium concentration and in‐depth water chemistry. These wells were identified using several criteria that began with an initial review of aerial images in an attempt to identify homes that were unlikely connected to a public water supply. This most certainly eliminated from consideration homes along main highways where a water line connection was likely available. Once a potential target well was identified several other chance factors presented themselves, including owner permission, whether the well was operational and used regularly, and whether an outdoor sample tap (faucet) was available prior to any treatment and was located relatively close to the wellhead.

Figure 2

Location of wells sampled (shaded insert indicates Sandhills region).

Testing then was extended to a wider area that included wells of known location:

n = 10 (wells #29–38) with known elevated radium levels (archived private‐well analyses) located away from the main study area (~60 km to the southwest) were then sampled and tested both for radium (retesting) and for selected water chemistry constituents.

n = 19 (wells #39–57) of known elevated calcium concentrations (file data, as above) were then tested for combined radium and for selected water chemistry constituents.

Well location was documented at the site using a Global Positioning System (GPS) hand‐held unit and is accurate to within 7 m. Well depth information was obtained from information tags on the well casing, when available, or through homeowner interview. Total well depths ranged from 12.8016 to 141.732 m deep (Table 1). All wells were presumed screened in the unconsolidated sands (and possibly all in the Middendorf Aquifer system).

Table 1

Universal Transverse Mercator Geographic Well Coordinates and Depth for Main Study Area (Wells #1–28), Known Radium (Wells #29–38), and Known Calcium (Wells #39–57)

Well #	GPS Coordinate	Well Depth (m)	Well #	GPS Coordinate	Well Depth (m)
1	UTM NAD83 17 463257°E 3748390°N	45.72	29	UTM NAD83 17 435490°E 3721850°N	51.816
2	Calculated coordinate	44.196	30	UTM NAD83 17 435681°E 3722232°N	n/a
3	UTM NAD83 17 464649°E °N	54.864	31	UTM NAD83 17 435662°E 3722606°N	n/a
4	UTM NAD83 17 464424°E 3748547°N	53.9496	32	UTM NAD83 17 434964°E 3722407°N	n/a
5	UTM NAD83 17 464535°E 3748376°N	n/a	33	UTM NAD83 17 435085°E 3721136°N	n/a
6	UTM NAD83 17 463536°E 3748388°N	n/a	34	UTM vNAD83 17 429130°E 3725796°N	n/a
7	UTM NAD83 17 464228°E 3748397°N	n/a	35	UTM NAD83 17 423566°E 3726476°N	n/a
8	Calculated coordinate	n/a	36	UTM NAD83 17 425587°E 3729242°N	n/a
9	UTM NAD83 17 464398°E 3753640°N	21.336	37	UTM NAD83 17 425533°E 3729074°N	73.152
10	UTM NAD83 17 466163°E 3741279°N	48.768	38	UTM NAD83 17 425854°E 3729114°N	n/a
11	UTM NAD83 17 465717°E 3748061°N	24.384	39	UTM NAD83 17 422926°E 3731014°N	20.1168
12	UTM NAD83 17 460651°E 3748230°N	47.5488	40	UTM NAD83 17 426556°E 3725232°N	45.72
13	UTM NAD83 17 463162°E 3748197°N	48.768	41	UTM NAD83 17 434737°E 3733349°N	91.44
14	UTM NAD83 17 462993°E 3748464°N	n/a	42	UTM NAD83 17 432396°E 3740128°N	91.44
15	UTM NAD83 17 457631°E 3748885°N	n/a	43	UTM NAD83 17 436866°E 3741096°N	n/a
16	UTM NAD83 17 458612°E 3747309°N	42.3672	44	UTM NAD83 17 443659°E 3738502°N	121.92
17	UTM NAD83 17 465040°E 3749157°N	54.864	45	UTM NAD83 17 448342°E 3748547°N	91.44
18	UTM NAD83 17 464463°E 3749212°N	37.1856	46	UTM NAD83 17 449578°E 3745901°N	n/a
19	UTM NAD83 17 463722°E 3747043°N	40.2336	47	UTM NAD83 17 449350°E 3745833°N	128.016
20	UTM NAD83 17 466496°E 3743217°N	30.48	48	UTM NAD83 17 456387°E 3749316°N	n/a
21	UTM NAD83 17 466572°E 3743170°N	29.8704	49	UTM NAD83 17 460457°E 3749299°N	24.384
22	UTM NAD83 17 462819°E 3746805°N	30.48	50	UTM NAD83 17 436324°E 3733527°N	121.92
23	UTM NAD83 17 460827°E 3745974°N	53.34	51	UTM NAD83 17 435987°E 3708042°N	45.72
24	UTM NAD83 17 463257°E 3748390°N	12.8016	52	UTM NAD83 17 419415°E 3723935°N	91.44
25	UTM NAD83 17 460132°E 3750621°N	42.0624	53	UTM NAD83 17 485460°E 3737354°N	91.44
26	UTM NAD83 17 460058°E 3751831°N	33.528	54	UTM NAD83 17 487373°E 3751394°N	45.72
27	UTM NAD83 17 463746°E 3750739°N	n/a	55	UTM NAD83 17 476514°E 3754757°N	141.732
28	Calculated coordinate	n/a	56	UTM NAD83 17 466821°E 3759681°N	n/a
			57	UTM NAD83 17 466187°E 3756470°N	n/a

Groundwater sampling followed standard health agency procedures. All wells were purged for several minutes prior to sample collection to ensure that the groundwater being collected had not been stagnant in the plumbing and well. Once flushed and with the well pump running, specific conductivity, temperature, and pH measurements were collected using a calibrated YSI Environments hand‐held multimeter (Model 556, YSI Incorporated, Yellow Springs, OH). The multimeter was calibrated using a 1000 μS/cm potassium chloride solution at the beginning of every sampling day and once every 20 samples per the operation manual. Water samples were collected as close to the well source as possible and up flow of any home treatment system. Radium samples were collected in 3.8 L plastic bottles and were acidified to pH ≤2 with sulfuric acid within 24 h of collection. Samples for the remaining chemical analyses were collected simultaneously with the radium samples and stored at 0–4°C in transit and at the laboratory. Water chemistry samples were collected, bottled, preserved, and stored per USEPA protocols.

Chemical Analyses

Samples of water for radium analysis from all 57 wells were submitted within 5 days of collection to the SCDHEC radiological laboratory for both 226Ra and 228Ra analyses. USEPA‐approved standard collection, preservation, and laboratory procedures were followed (detailed in the supporting information).

All water‐chemistry analyses were performed by the SCDHEC analytical services laboratory using EPA‐approved standard procedures in collection, preservation, and analysis. For the 28 wells in the main study area, an extensive suite of geochemical constituents (results available in the supporting information) included metals and trace elements (Si, Hg, Ag, Al, B, Ba, Be, Co, Cr, Cu, Fe, Mn, Mo, Ni, Li, Sn, Sr, U, Zn, Cd, Pb, Sb, Se, and As), major cations (Na, K, Ca, and Mg), major anions (NO3 1−+NO2 1−─N, Cl1−, F1−, and SO4 2−), and general inorganic indicators (conductivity, pH, alkalinity, TDS, and hardness); pH and conductivity were also measured in the field at sampling using a calibrated quality field instrument. The remaining 29 wells located outside the main study area were analyzed for selected water chemistry constituents, which only included Ca, Mg, Ba, Cu, Fe, Mn, Zn, Pb, Cl1−, NO3 1−+NO2 1−─N, K, Si, Sr, alkalinity, and hardness, along with conductivity, pH, and TDS.

Water Quality Results

Water quality results for select parameters are given in Table 2. Simple scatterplots illustrate the relationships between combined radium and select (usefully associated) water chemistry constituents for each of the three study areas (Figures 3, 4, 5, 6, 7, 8, 9). Because ultimate concerns here were related to public health, well waters near, at, and above the USEPA drinking water limit (MCL) for radium were given closer attention.

Table 2

Water Quality Results for Select Parameters

Well #	226Ra (pCi/L)	228Ra (pCi/L)	226Ra + 228Ra (pCi/L)	Ca (mg/L)	Mg (mg/L)	Ba (mg/L)	Sr (mg/L)	K (mg/L)	Mn (mg/L)	Alkalinity (mg/L)	Hardness (mg/L)	Total Dissolved Solids (mg/L)	Sp. Cond. (μS/cm)	pH	Cl (mg/L)	NO3 + NO2 (mg/L)	Si (mg/L)	Cu (mg/L)	Fe (mg/L)	Pb (mg/L)	Zn (mg/L)
1	11.6	39.6	51.2	8.6	5.2	0.220	0.063	3.6	0.120	<1.0	43	130	127	4.5	7.0	14.0	5.1	0.017	0.022	<0.005	0.018
2	0.5	3.2	3.7	0.9	0.5	0.025	0.005	0.5	0.005	<1.0	5	45	20	5.2	1.2	1.8	5.3	<0.01	<0.02	<0.005	0.016
3	3.9	4.0	7.9	1.6	1.3	0.025	0.014	0.5	0.005	–	9	29	22	4.9	3.6	3.1	6.1	<0.01	<0.02	<0.005	0.016
4	0.9	5.5	6.4	1.9	1.4	0.025	0.012	1.2	0.042	–	10	26	36	5.2	3.1	1.9	4.9	<0.01	0.68	<0.005	0.27
5	1.3	30.0	31.3	8.3	7.3	0.110	0.069	3.4	0.064	0	51	140	179	4.4	14.0	14.0	5.5	0.036	<0.02	<0.005	0.016
6	7.6	21.4	29.0	8.6	5.3	0.220	0.063	3.6	0.120	<1.0	43	100	112	4.4	4.6	11.0	5.4	0.017	0.022	<0.005	0.018
7	0.4	1.7	2.0	0.4	0.2	0.025	0.005	0.5	0.005	<1.0	2	16	12	5.3	1.2	0.6	5.4	0.014	<0.02	<0.005	<0.010
8	5.7	29.2	34.9	4.2	3.3	0.130	0.022	2.4	0.038	<1.0	24	45	85	4.8	6.9	4.2	5.0	0.018	<0.02	<0.005	<0.010
9	1.5	9.2	10.7	1.0	0.6	0.025	0.005	1.1	0.011	<1.0	5	26	31	4.8	4.0	2.0	4.4	0.028	0.04	<0.005	0.021
10	1.6	4.3	5.9	0.4	0.6	0.025	0.005	0.5	0.005	<1.0	3	24	23	4.8	1.7	1.4	4.5	0.028	<0.02	<0.005	0.015
11	3.7	10.5	14.2	1.0	1.2	0.025	0.012	0.5	0.005	<1.0	7	28	51	4.8	3.6	3.2	5.5	<0.01	<0.02	<0.005	0.016
12	3.0	7.1	10.1	1.4	1.4	0.025	0.014	0.5	0.005	<1.0	9	29	40	4.7	2.3	3.0	5.3	<0.01	<0.02	<0.005	<0.010
13	3.6	9.9	13.5	3.6	2.0	0.061	0.028	0.5	0.026	–	17	50	71	4.6	2.7	5.8	5.6	<0.01	<0.02	<0.005	<0.010
14	0.9	2.8	3.7	1.2	0.5	0.025	0.005	0.5	0.005	–	5	22	26	5.1	1.9	2.0	5.6	<0.01	<0.02	<0.005	<0.010
15	2.1	16.5	18.6	13.0	7.4	0.025	0.110	0.5	0.005	<1.0	63	27	52	4.7	6.5	4.2	3.5	<0.01	<0.02	<0.005	0.013
16	3.6	9.8	13.4	1.4	1.2	0.025	0.013	0.5	0.005	<1.0	8	30	39	4.8	2.7	2.9	5.2	0.025	<0.02	0.0065	0.016
17	0.2	0.8	1.0	0.8	0.2	0.025	0.005	0.5	0.005	<1.0	3	21	13	5.3	1.2	0.6	6.4	<0.01	<0.02	<0.005	0.01
18	0.5	1.8	2.2	0.2	0.2	0.025	0.005	0.5	0.005	<1.0	1	45	11	5.1	1.4	0.2	4.9	0.021	<0.02	<0.005	<0.010
19	1.5	8.7	10.2	0.5	0.8	0.025	0.005	0.5	0.005	<1.0	5	32	32	4.7	2.8	2.0	4.4	<0.01	<0.02	<0.005	<0.010
20	13.6	24.5	38.1	3.0	2.8	0.220	0.021	4.7	0.091	–	19	65	100	4.4	5.5	9.5	4.9	0.012	<0.02	<0.005	<0.010
21	4.2	8.2	12.4	1.9	1.9	0.055	0.019	0.5	0.015	–	12	40	56	4.7	3.4	4.8	5.4	<0.01	<0.02	<0.005	<0.010
22	0.3	1.7	2.0	0.6	0.3	0.025	0.005	0.5	0.005	<1.0	3	45	18	5.4	1.3	1.0	6.6	<0.01	<0.02	<0.005	<0.010
23	0.3	1.1	1.5	0.6	0.3	0.025	0.005	0.5	0.005	<1.0	3	20	18	5.2	1.2	4.2	5.7	<0.01	<0.02	<0.005	<0.010
24	0.9	7.4	8.3	2.9	1.0	0.052	0.014	3.2	0.023	<1.0	11	45	57	4.8	1.7	2.0	3.6	0.014	<0.02	<0.005	<0.010
25	1.8	7.3	9.0	0.8	0.8	0.025	0.005	0.5	0.005	<1.0	6	45	29	4.9	1.8	4.2	4.8	0.016	<0.02	<0.005	0.032
26	3.4	18.2	21.6	1.2	0.8	0.025	0.010	1.1	0.014	<1.0	6	30	45	4.6	2.8	2.2	4.9	<0.01	<0.02	<0.005	0.013
27	0.8	4.9	5.7	0.6	0.6	0.025	0.005	0.5	0.005	<1.0	4	25	25	4.9	1.6	4.2	5.6	0.016	<0.02	<0.005	0.015
28	15.3	14.5	29.8	4.0	3.3	0.130	0.038	2.3	0.018	<1.0	24	58	96	4.7	8.5	8.3	5.3	0.036	<0.02	<0.005	<0.010
29	11.6	1.4	13.0	0.4	0.3	0.025	0.005	0.5	0.005	<1.0	2	14	17	4.8	1.3	0.8	6.2	0.028	<0.02	<0.005	0.027
30	14.2	2.4	16.6	0.3	0.4	0.025	0.005	0.5	0.005	<1.0	2	15	17	4.9	1.1	0.7	6.5	0.017	<0.02	0.026	0.023
31	9.0	1.6	10.7	0.3	0.4	0.025	0.005	0.5	0.005	<1.0	2	13	19	4.8	2.1	0.6	5.5	0.011	<0.02	<0.005	<0.010
32	2.9	2.0	4.9	0.6	0.7	0.025	0.005	0.5	0.005	<1.0	4	19	23	4.9	2.2	1.1	5.8	<0.01	<0.02	<0.005	<0.010
33	11.0	1.4	12.4	0.5	0.5	0.025	0.005	0.5	0.005	1.5	3	21	22	4.7	2.4	1.1	5.8	<0.01	<0.02	<0.005	<0.010
34	1.4	0.6	2.0	0.2	0.2	0.025	0.005	0.5	0.005	<1.0	2	10	11	5.0	1.4	0.4	5.1	<0.01	0.048	<0.005	<0.010
35	1.1	0.7	1.8	0.1	0.2	0.025	0.005	0.5	0.005	<1.0	<1.0	11	9	4.9	1.3	0.0	4.3	<0.01	<0.02	<0.005	<0.010
36	57.4	8.6	66.0	4.1	2.7	0.025	0.042	1.6	0.011	<1.0	21	61	91	4.3	12.0	4.9	5.1	0.017	<0.02	<0.005	0.019
37	40.7	8.0	48.7	3.0	2.3	0.025	0.036	1.4	0.015	<1.0	17	48	79	4.4	9.8	4.6	5.2	0.023	<0.02	<0.005	0.015
38	53.1	15.0	68.1	3.4	2.8	0.058	0.039	1.3	0.005	<1.0	20	55	93	4.3	10.0	6.1	5.3	0.026	<0.02	<0.005	0.019
39	8.1	5.5	13.6	4.8	2.0	0.067	0.051	2.2	0.005	<1.0	20	64	72	4.5	9.9	3.8	4.9	0.017	<0.02	<0.005	0.031
40	0.3	<0.737	<1.037	2.1	1.4	0.025	0.029	3.5	0.110	12	11	88	47	6.3	1.3	0.0	46.0	<0.01	3.2	<0.005	<0.010
41	0.4	<0.735	<1.135	5.7	1.8	0.067	0.050	2.8	0.170	23	22	97	71	6.5	1.1	0.0	49.0	<0.01	1.5	<0.005	0.012
42	<0.24	<0.691	<0.931	2.7	0.8	0.025	0.020	2.0	0.005	22	10	87	39	6.1	1.7	0.2	58.0	<0.01	<0.02	<0.005	0.51
43	0.7	<0.610	<1.31	11.0	1.5	0.092	0.051	2.0	0.120	36	34	93	87	6.8	1.5	0.0	43.0	<0.01	6.6	<0.005	0.059
44	3.2	<0.868	<4.068	7.1	2.0	0.074	0.047	4.2	0.170	25	26	100	74	6.3	1.2	0.0	46.0	<0.01	1.9	<0.005	0.019
45	<0.246	<0.484	<0.73	2.2	1.0	0.025	0.018	1.6	0.005	12	10	48	34	6.4	1.7	0.6	24.0	<0.01	<0.02	<0.005	0.048
46	0.5	0.6	1.1	1.4	0.6	0.025	0.017	2.2	0.005	7.3	6	49	35	5.7	2.9	1.4	22.0	0.022	<0.02	<0.005	0.057
47	1.0	1.7	2.7	8.3	1.9	0.053	0.044	2.6	0.140	44	28	110	88	6.4	2.6	0.0	57.0	<0.01	0.2	<0.005	0.01
48	<0.249	0.9	<1.149	5.8	1.0	0.025	0.065	3.0	0.014	23	18	68	64	5.8	6.3	1.3	23.0	<0.01	0.026	<0.005	<0.010
49	9.6	21.9	31.5	5.4	4.8	0.140	0.046	3.5	0.040	<1.0	33	79	115	4.3	5.3	13.0	4.4	<0.01	<0.02	<0.005	0.025
50	1.4	1.9	3.3	4.6	2.6	0.096	0.038	3.5	0.450	25	22	100	76	6.6	1.4	0.0	48.0	<0.01	3	<0.005	0.022
51	4.8	0.7	5.5	0.7	0.4	0.025	0.005	0.5	0.005	<1.0	3	21	22	5.1	2.6	1.5	6.4	0.026	<0.02	<0.005	0.015
52	0.3	<0.663	<0.963	2.8	2.3	0.025	0.033	0.5	0.018	27	16	82	56	7.0	1.5	0.1	43.0	<0.01	0.11	<0.005	0.024
53	13.2	6.5	19.7	1.0	1.2	0.025	0.013	0.5	0.005	<1.0	7	23	31	5.3	2.9	2.0	5.4	0.095	<0.02	<0.005	0.015
54	2.6	5.9	8.4	1.0	0.7	0.025	0.005	1.3	0.005	<1.0	5	37	55	4.9	6.5	3.0	5.2	0.01	<0.02	<0.005	0.03
55	<0.329	1.3	<1.629	5.1	1.4	0.025	0.036	1.1	0.005	32	18	77	72	7.1	1.3	0.0	32.0	<0.01	<0.02	<0.005	0.092
56	<0.389	<0.928	<1.317	4.8	5.2	0.025	0.052	2.0	0.028	47	33	81	81	7.1	1.2	0.0	32.0	<0.01	0.26	<0.005	0.021
57	29.4	63.7	93.1	8.4	5.7	0.170	0.005	4.8	0.028	<1.0	44	120	169	4.3	13.0	15.0	3.6	<0.01	<0.02	<0.005	<0.010

Figure 3

Comparison of combined radium versus TDS for each study site.

Figure 4

Comparison of combined radium versus pH for each study site.

Figure 5

Comparison of combined radium versus specific conductivity for each site.

Figure 6

Comparison of specific conductivity versus TDS for each site.

Figure 7

Comparison of combined radium versus nitrate plus nitrite for each site.

Figure 8

Comparison of combined radium versus calcium for each site.

Figure 9

Comparison of combined radium versus magnesium for each site.

In the main study area of known widespread radium problem, sampling wells that had not been previously analyzed and were randomly selected showed (when plotted) substantially higher elevated levels of combined radium with calcium, nitrate‐plus‐nitrite, conductivity, TDS, and with pH lowered below ~4.5. The almost mineral‐free nature of the natural background groundwater in this highly leached quartz‐sand and kaolinitic‐clay sedimentary geology helps make these associations starker and more easily detected: Even small additions are obvious. This is a desirable condition for obtaining surrogate indicators, especially any that is geochemically directly involved in the presence of higher levels of radium.

The second set of wells (known radium site)—those that had already shown distinctly elevated radium concentration—also exhibited the greatest increase in concentrations of combined radium when regressed against TDS (R = 0.97; p < 0.05) (Figure 3).

For the known calcium wells, where the majority were less acidic and lower in combined radium, the pH and combined radium correlation (R = 0.65; p < 0.05) was less prominent compared to the acidic groundwater associated with the main study wells (R = 0.76; p < 0.05) and the known radium wells (R = 0.96; p < 0.05). In all three series of wells investigated here, it appears that wells with a pH value >5.5 did not tend to exceed the MCL for combined radium (Figure 4). There also appears to be a distinct negative association between pH and combined radium. Figure 3 suggests that the known calcium wells did not exhibit the same positive association between TDS and combined radium (R = 0.09; p = 0.3), as did the main study wells (R = 0.85; p < 0.05).

Conversely, specific conductivity measurements did not vary significantly among the phase area groups; however, the known calcium wells, similar to the TDS observations, did not appear to exhibit the positive association with combined radium as the other sites (Figure 5). When TDS in groundwater is ionic in nature, TDS and specific conductivity are approximately related (Figure 6). The observed deviations as seen here can be explained by the presence of high silica concentrations in some samples. In most natural groundwaters (pH < 8), dissolved Si exists dominantly as silicic acid (H4Si04), a nonionic species, thereby adding to TDS but not to specific conductivity.

Nitrate‐plus‐nitrite showed good correlation with combined radium (R = 0.91, p < 0.05) for the main study site wells, and correlations were relatively consistent between the three areas (Figure 7) in that there appears to be a strong positive relationship between these two parameters. Wells associated with the main study site also exhibited some of the highest concentrations of calcium, whereas the known calcium site wells comprised a greater number of wells with high calcium concentrations (Figure 8). Measured radium concentrations were consistently less in the known calcium site wells compared to the other two groups. Similar to calcium, magnesium concentrations were somewhat consistent between all three areas (Figure 9) with radium concentrations trending lower in wells of the known calcium compared to the other two sites.

Modeling Effort

Simple and multiple linear regression modeling techniques were used to construct the most parsimonious models for estimating combined radium concentrations within the Sandhills region using water‐chemistry variables found to correlate significantly with combined radium. Pearson correlation coefficients derived from the water‐chemistry results of the main study site wells are presented for the potential predictor variables: TDS, pH, specific conductivity, nitrate (plus nitrite), calcium, magnesium, sodium, chloride, and hardness (Tables 3 and 4).

Table 3

Pearson's r Correlation Coefficient Matrix Illustrating log10 Transformed Data for Select Water Quality Variables for Wells Located Within the Main Study Area (n = 28)

N = 28	log10 Combined Ra	log10 226Ra	log10 228Ra	log10 Ca	log10 Na	log10 Mg	log10 Hardness	log10 Conductivity	log10 pH	log10 Cl	log10 TDS	log10 NO3 + NO2
log10 combined Ra	1
log10 Ra 226	0.90	1
log10 Ra 228	0.99	0.84	1
log10 Ca	0.76	0.62	0.76	1
log10 Na	0.38	0.41	0.36	0.43	1
log10 Mg	0.90	0.78	0.89	0.92	0.48	1
log10 hardness	0.85	0.72	0.84	0.98	0.47	0.98	1
log10 conductivity	0.92	0.78	0.92	0.86	0.30	0.92	0.91	1
log10 pH	−0.90	−0.81	−0.89	−0.67	−0.37	−0.82	−0.77	−0.86	1
log10 Cl	0.87	0.74	0.86	0.80	0.42	0.90	0.87	0.87	−0.76	1
log10 TDS	0.81	0.61	0.82	0.91	0.30	0.92	0.95	0.93	−0.81	0.85	1
log10 NO3 + NO2	0.91	0.81	0.88	0.90	0.45	0.97	0.96	0.96	−0.87	0.88	0.93	1

Table 4

Pearson's r Correlation Coefficient Matrix Illustrating Nontransformed Data for Select Water Quality Variables for Wells Located Within the Main Study Area of Concern (n = 28)

N = 28	Combined Ra	226Ra	228Ra	Ca	Na	Mg	Hardness	Conductivity	pH	Cl	TDS	NO3 + NO2
Combined Ra	1
226Ra	0.81	1
228Ra	0.97	0.64	1
Ca	0.67	0.38	0.71	1
Na	0.17	0.11	0.18	0.66	1
Mg	0.74	0.44	0.78	0.96	0.56	1
Hardness	0.71	0.42	0.76	0.99	0.62	0.99	1
Conductivity	0.87	0.61	0.88	0.71	0.08	0.83	0.78	1
pH	−0.76	−0.60	−0.74	−0.58	−0.25	−0.66	−0.63	−0.77	1
Cl	0.76	0.50	0.79	0.71	0.26	0.85	0.79	0.88	−0.64	1
TDS	0.85	0.47	0.90	0.96	0.19	0.97	0.98	0.96	−0.69	0.83	1
NO3 + NO2	0.91	0.64	0.91	0.94	0.31	0.96	0.97	0.97	−0.76	0.83	0.97	1

A log10‐transformation of the data for each water quality parameter was performed in an effort to approximate a normal distribution. Both the nontransformed and log10‐transformed data sets were explored simply to demonstrate whether the parameters chosen for each data set would differ in model development. Data transformation is commonly utilized to make data more symmetric, linear, and constant in variance [Helsel and Hirsch, 1992]. Geochemical data are most commonly transformed using log10 [Filzmoser and Reimann, 2000].

The models developed using multiple regression are expressed in the general form: Y = β 0 + β 1 X 1 + β 2 X 2 + β X … + ε, where Y denotes the dependent variable (combined radium concentration), β 0 is a constant, β 1 through β represent the regression coefficients for each independent variable X 1 through X , and ε designates the model error term. Models were developed with forward selection and backward elimination by utilizing various features of both SAS [SAS, 2003] and R software (Vienna, Austria Version 2.9.0) [R Development Core Team, 2007]. The models contained between 19 and 54 observations. Predictors are water chemistry and well location, which were represented by a dummy variable.

Collinearity diagnostics including condition index and proportion of variation values eliminated most models from consideration. Only Model A ((y 1 = −5.88 + 15.42(x 1) − 14.74(x 2) + 0.39(x 3), where x 1 = Site A (known radium site wells), x 2 = Site B (known calcium site wells), and x 3 = conductivity)) and Model B ((y 1 = −1.06 + 0.361(x 1) − 0.79(x 2) + 1.27(x 3), where x 1 = Site A (known radium site wells), x 2 = Site B (known calcium site wells), and x 3 = log10 conductivity)) are considered the best fitting and applicable (Table 5) for predicting combined radium concentrations at each site. Observed versus predicted radium concentrations for both models are illustrated in Figures 10 and 11. A cross validation was also performed in an effort to evaluate the reasonableness and predictive ability of these selected best fitting models by partitioning data into two subsets known as a training set (model building set) and a test set (predictor set) [Kutner et al., 2004]. The training set was based on 25 main study site observations. The test set was developed using three randomly chosen observations from within the area of main study not included in the model training set (wells 9, 11, and 17). The predicted values of both models are compared to the observed values in order to assess goodness of fit for each. A paired two‐sample t test was performed to assess whether the mean predicted values and mean observed values differed. In addition, because of such a small sample size (n = 3), the nonparametric equivalent Mann Whitney U test was utilized to assess the predicted and observed median values. Test results are considered statistically significant (P‐value < 0.05). As indicated by the outcomes of the paired t test and the Mann‐Whitney U test (Table 6), the null hypothesis is not rejected; there is no difference between the predicted and observed results. Therefore, statistically both Model A and Model B are believed to have good validity. Agreement between the predicted and observed concentrations for both models is quite good, suggesting that any normal assumption violation (non‐normal distribution, nonconstant variance, etc.) does not significantly affect the outcome of the tests.

Table 5

Results for Full MLR Models Developed From Nontransformed Data (Model A) and log10 Transformed Data (Model B) Including 54 Total Wells From the Main Study Area of Concern, Known Radium Wells (Site A), and Known Calcium Wells (Site B)

	β	SE	t‐value	Significance
Model A Adj. R 2 = 0.64; RMSE = 11.78
Intercept	−5.88	3.24	−1.81	0.07
Site A	15.42	4.45	3.46	<0.01
Site B	−14.74	3.73	−3.95	<0.01
Sp. Conductivity	0.39	0.04	9.34	<0.01
Model B Adj. R 2 = 0.56; RMSE = 0.413
Intercept	−1.06	0.32	−3.31	<0.01
Site A	0.361	0.1524	2.368	0.02
Site B	−0.79	0.13	−6.08	<0.01
log10 Sp. conductivity	1.27	0.19	6.63	<0.01

Figure 10

Model A observed versus predicted combined radium concentrations.

Figure 11

Model B observed versus predicted log10 transformed combined radium concentrations.

Table 6

MLR Model Validation: Mann‐Whitney U Test and Paired t Test Results for Significant Differences Between Predicted and Observed Combined Radium Concentrations

Mann‐Whitney U Test			Paired t Test
Model A (significance = 0.6625)	n	Median	Model A (P‐value = 0.228)	n	Median
Observed concentration (pCi/L)	3	10.68	Observed concentration (pCi/L)	3	8.61
Predicted concentration (pCi/L)	3	6.21	Predicted concentration (pCi/L)	3	7.41
Model B (significance = 1.00)			Model B (P‐value = 0.817)
Observed concentration (pCi/L)	3	1.029	Observed concentration (pCi/L)	3	0.721
Predicted concentration (pCi/L)	3	0.834	Predicted concentration (pCi/L)	3	0.766

Discussion

Existing file data for radium in public‐system water wells in South Carolina suggest that values >5 pCi/L (the MCL) are relatively common and occur primarily within the inner coastal plain Sandhills region. Thus, this area is the focus for concerns about the large number of untested private home wells used for drinking water.

The results here suggest that several commonly analyzed water “quality” (chemistry) constituents can be important indicators for excessive dissolved radium in groundwater located in the Sandhills region of the inner coastal plain of South Carolina and thus likely in similar shallow‐zone acidic‐sand geology of neighboring states and perhaps more widely. These “surrogate parameters” might come from file data on drinking‐water wells or from new screening surveys using the readily and inexpensively obtained chemical data. The surrogates might be very useful simply as threshold indicators obtained by (1) scatterplots of straight or log‐transformed data or (2) via regression analysis. Useful predictions may be a simple as “Parameter A measured above a concentration of X indicates a substantial risk of combined radium above the health‐base limit [MCL].” Because the health limit is low and data and associations in the region are “noisy,” a very useful prediction that will take little calibration involves finding wells with a substantial radium content, say several times the MCL, and thus of greater concern as a drinking‐water source. Just finding the worst wells would be a very useful beginning: wells at several times the MCL or higher are relatively common. Colleagues have found a home well in this area with 40 times the MCL (combined radium = 200 pCi/L).

The chemical data can also be used in a more sophisticated predictive manner. Modeling here sought the best possible linear regression model(s) for predicting radium concentrations within the wider study region (all three phases). The results of this modeling effort suggest conductivity to be the best predictor of combined radium in both Model A (R 2 = 0.64) and Model B (R 2 = 0.56), both of which include observations for all three sites. A comparison between predicted and observed combined radium concentrations for both models indicates general agreement, suggesting that any normal assumption violation (non‐normal distribution, nonconstant variance, etc.) did not significantly affect the outcome of the tests. These models are applicable to the greatest number wells and most useful especially since an instantaneous reading can be obtained at the well using a hand‐held conductivity meter. The results from this study indicate that basic predictive models can provide effective decision support for detecting radium in wells within the inner coastal plain. Models containing only one water quality variable in addition to a categorical variable for site were the best predictors of combined radium, thus would be the most useful as a screening tool. In most models developed containing more than one quantitative water quality variable, collinearity was significant.

Overall for the main study site wells, it appears that groundwater with a pH value >5.5 tended not to exceed the MCL for combined radium. Nitrate‐plus‐nitrite concentrations >5 mg/L (as N) occurred in six wells. All six wells exceeded the combined‐radium MCL, ranging between 13 and 51 pCi/L. Here nitrate contamination is undoubtedly due to surface influences, probably anthropogenic sources such as applied fertilizer and possibly by septic tanks [Aelion and Conte, 2004]. The bacterial nitrification process (NH4 to NO3 1−) releases acidity and likely contributes to the lowered pH of the groundwater [DePaul et al., 2005], especially in soils with minimal buffering capacity such as those found within the Sandhills region. At well #57, a combined radium concentration of 93.1 pCi/L is found with the highest measured nitrate‐plus‐nitrite concentration (15 mg/L as N) and the second lowest pH reading (4.28). This low‐pH situation is likely causing greater radium desorption from the aquifer material.

For the known calcium site wells, median conductivity was nearly twice as great as for the main study site wells and the known radium site wells. Additionally, the known calcium site wells generally did not exhibit the same positive association with combined radium as the other sites. Some other complicating factor(s) must be at work. Median pH values of the known calcium site wells were 6.28 compared to 4.78 for the known high radium wells and 4.81 for the main study site wells. Naturally occurring soluble mineral carbonates may occur at two boundaries of the coastal plain sands: (1) in slate belt metamorphic rocks beneath the sedimentary sands (into which some wells possibly penetrated and (2) sparse sedimentary carbonates within the sands, as trace outliers of the limestone found downdip and downflow).

The pH value and conductivity appear to offer the most practical predictors for identifying wells with elevated concentrations of combined radium because they are good indicators and are both easy and inexpensive to obtain (even by a hand‐held meter at the wellhead), especially when compared to laboratory radium analysis. Elevated nitrate can also be tested for in the field with ion‐specific electrode or indicator‐paper strips.

Conclusion

Presently, a large radium data set for private residential wells does not exist in this state, despite that radium is clearly a common problem in home water wells. In this study, 34 of the 57 wells tested for combined radium exceeded the MCL of 5 pCi/L, confirming earlier research work by others: radium is widespread and sporadic in the areas tested. Greater understanding of the problem is needed. As a beginning, surrogate or screening analyses were sought and seemingly obtained that could, with useful accuracy, estimate radium concentrations based upon inexpensive, easily obtained, chemical attributes of the groundwater.

The highly leached sands and low reactivity kaolinitic clays of the Sandhills host groundwater of very low TDS content and low buffering capacity. Thus, minor chemical alterations can appear sharply against these low natural backgrounds.

Combined radium concentration varies usefully with TDS, conductivity, nitrate‐plus‐nitrite, calcium, magnesium, hardness, and chloride, and pH (inverse association).

Conflict of Interest

The authors declare no conflicts of interest relevant to this study.

Supporting Information S1

Click here for additional data file.

Table S1

Click here for additional data file.

Table S2

Click here for additional data file.

Table S3

Click here for additional data file.

Table S4

Click here for additional data file.

Table S5

Click here for additional data file.

Table S6

Click here for additional data file.

Table S7

Click here for additional data file.

Table S8

Click here for additional data file.

Table S9

Click here for additional data file.

Table S10

Click here for additional data file.

Table S11

Click here for additional data file.

Table S12

Click here for additional data file.

Table S13

Click here for additional data file.

Table S14

Click here for additional data file.

Table S15

Click here for additional data file.

Table S16

Click here for additional data file.

Table S17

Click here for additional data file.