Purpose and Scope

This report documents the following components of data compilation and evaluation: (1) a review of file information for 43 public water-supply system wells (hereafter referred to as wells) from the PADEP Northcentral and Southcentral regions to evaluate the SWIP and Microscopic Particulate Analysis (MPA) in GUDI determinations, (2) the addition of attributes to the PADWIS database including spatial anthropogenic (land cover and PADEP region) and naturogenic (geologic and physiographic, hydrologic, soil characterization, and topographic) data that could be potential indicators of GUDI designation or the presence of contaminants in MPA results, which was the impetus for the creation of three datasets (PADWIS database, PADWIS database subset, and MPA database subset) to be used for analysis (Gross, 2022), and (3) statistical summary and correlation analysis to evaluate existing GUDI sources in the databases with respect to hydrogeologic and source-construction characteristics that are currently utilized in the SWIP assessment methodology.

Data Availability

Files associated with 43 wells in PADEP’s Northcentral and Southcentral regions were examined to verify and compile missing data values for PADWIS attributes that are important for conducting the SWIP. Although the PADWIS database includes most of the criteria used for the SWIP, such as major aquifer type, static water level, or whether the aquifer is confined (Pennsylvania Department of Environmental Protection, 2001, 2002b), it does not provide sufficient information to show what data were utilized to classify a well as GUDI. For example, the SWIP STATUS attribute (table 1) in the PADWIS database indicates if an evaluation was completed, but this attribute does not provide results for SWIP monitoring or the total risk-factor scores from any collected MPA samples, so it is not known which, if any, of the SWIP monitoring criteria or guidelines (Pennsylvania Department of Environmental Protection, 2001, 2002b) were utilized during the evaluation. As a result, some PADWIS attributes could have verification performed using information in the well files to change incorrect values to correct values or to fill in missing values. In addition, files were examined to find additional attributes that were not included in the PADWIS database that might help to explain a well’s susceptibility to surface-water influence.

In the Southcentral region, the well depth (DEPTH [FT] attribute, table 1) was found in the files for 18 selected wells. Six (33 percent) of the 18 values differed from the data values stored in the PADWIS database. Casing depth (CASING DEPTH [FT] attribute, table 1) was available for 15 wells. Three (20 percent) of the 15 values differed from the data values stored in the PADWIS database. Depth to water in the well (STATIC WATER LEVEL [FT] attribute, table 1) was found in the files associated with 14 (74 percent) of the 19 wells. According to PADEP’s SWIP, static water level in the well is a necessary attribute for determining if a source is susceptible to surface-water influence (Pennsylvania Department of Environmental Protection, 2001).

Other variables of importance to determine if additional monitoring and evaluation are needed on a source include the spatial coordinates that define a well’s location (LATITUDE and LONGITUDE attributes, table 1). Spatial coordinates for 11 wells in the Southcentral region were found in files associated with the wells, yet the spatial coordinates listed in the files for 9 (82 percent) of these 11 wells differed by 130 to 14,800 feet (ft) from the coordinates provided in the PADWIS database. Some of these discrepancies could be caused by assuming an incorrect datum, but most errors exceeded the error that would arise with a shift from North American Datum (NAD) 1927 to NAD 1983.

Files associated with wells also were evaluated to obtain specific data that were collected by completing the SWIP but unavailable as PADWIS attributes. For example, the distance (threshold distance of 200 ft) from a well to the nearest surface-water feature, such as a flowline, water feature, or water body, is not an attribute available for assessment in the PADWIS database (Pennsylvania Department of Environmental Protection, 2016), but this is an attribute that may require a well to be identified for further evaluation. Data describing the distance to the nearest surface-water feature were found in files for 9 (47 percent) of 19 selected wells examined in the Southcentral region. For most of the GUDI wells in the Southcentral region, the distance to the nearest surface-water feature was not evaluated as part of the initial screening because the static water level in the wells did not exceed 50 ft below the land surface. This highlights an important point when considering the primary threat of surface-water influence to a well. Instead of evaluating the relation between groundwater and surface water (for example, streams and lakes), the focus for these nine wells was water-quality variations and the potential for rapid infiltration of precipitation if preferential-flow pathways are available at or near the wellhead. Evaluation of this primary threat is fundamentally different compared to determining a hydraulic connection between an aquifer and stream or lake.

Pearson-correlation coefficients, which are derived from data (water-quality parameters, precipitation, and local surface-water conditions) collected by the water system during the 6-month SWIP monitoring program, that are considered high (> 0.40) can be used to classify a well as GUDI or to elicit MPA testing, but these results are not included in the PADWIS database. Pearson-correlation results from the SWIP monitoring were available in files for 14 (74 percent) of 19 selected wells designated as GUDI in the Southcentral region, whereas 9 (82 percent) of the 11 non-GUDI wells had Pearson-correlation results from the SWIP monitoring.

Results associated with MPA sample collection are an important, and often final, part of the evaluation process since these results are reported as the definitive presence or absence of specific bioindicators. MPA results were available in files associated with 15 (79 percent) of 19 selected wells in the Southcentral region, and 4 (44 percent) of the 9 non-GUDI wells with the SWIP monitoring results had associated MPA results.

Application of the Surface Water Identification Protocol

The SWIP consists of up to three steps: (1) screening for sources susceptible to surface-water influence based on aquifer type and well characteristics, (2) monitoring for six months to evaluate the Pearson-correlation results among water level, precipitation, or stream stage with selected water-quality parameters, and (3) sampling for particulates and organisms with an MPA. Overall, the best information from the well files about how the SWIP was followed and ultimately why a well was classified as non-GUDI or GUDI is contained in the letter of determination from PADEP to the water supplier. The letter summarizes the specific evidence from the evaluation that supports the determination. A letter of determination was available for 20 (47 percent) of the wells included in file reviews from selected PADEP regions.

File review of selected wells showed that for some sources the GUDI determination in practice is much more complex and cannot be easily summarized in a database such as PADWIS. One of the GUDI wells in the Southcentral region that illustrates this complexity was in service in 2013 as a community well classified as non-GUDI when the operator noticed an increase in turbidity and nitrate concentrations following precipitation events (Thomas Yeager, Pennsylvania Department of Environmental Protection, written commun., October 24, 2017). After notifying PADEP, the well was shut down, owing to elevated bacteria levels and the presence of bioindicators in water samples, and reclassified as GUDI in March 2013. Further investigation by the well owner revealed a cracked well casing that allowed surface water to drain into the well. The well was reconstructed, after which 6 months of SWIP monitoring was conducted and several MPA samples were collected by PADEP and the operator. The monitoring and MPA results showed a low risk for surface-water influence, so the well was classified as “groundwater” (non-GUDI) in December 2015. This process, which started with an astute water-system operator, took nearly two years to complete.

Review of the well files shows that application of a uniform evaluation approach for GUDI is difficult because of highly variable site-specific factors that need to be considered in applying the SWIP steps, but its effect is difficult to quantify. This is especially notable in the evaluation of results from the 6 months of SWIP monitoring because the Pearson-correlation results among the monitored parameters are commonly moderate (close to but not exceeding 0.40). In such cases, the results of analyses of the bacteria samples collected during the monitoring are sometimes used to provide additional rationale for requiring an MPA sample. The most ambiguous cases seem to be those with: (1) moderate Pearson-correlation results during the monitoring, (2) detections of some bacteria, and (3) a low-risk MPA sample with some organisms detected. In some of these cases, the presence of bacteria and other organisms was used as an indication of surface-water influence and as evidence to classify the source as GUDI; in other cases, despite the presence of bacteria, the source was not classified as GUDI given the low-risk MPA results in addition to the implementation of disinfection treatment practices.

Implementation of the SWIP in PADEP’s Southcentral and Northcentral regions is further summarized in the following sections based on the examination of well files. Files were examined to determine: (1) why 19 wells in the Southcentral region were classified as GUDI, (2) why 11 similar wells in the Southcentral region were determined to be groundwater (non-GUDI), and (3) why 13 wells in the Northcentral region were classified as GUDI without undergoing SWIP monitoring.

Pennsylvania Drinking Water Information System

The original version (before the application of quality-control (QC) procedures by the USGS) of the PADWIS database provided by PADEP for the purposes of this study contained information for 12,445 public water-supply sources, of which 560 sources were identified as GUDI and 11,885 sources as non-GUDI. For the non-GUDI sources, the database was modified by PADEP before being provided to USGS so that it only included public water-supply system source types of wells. Available attributes for GUDI and non-GUDI sources are listed in table 1 along with the number of wells that have available data for each attribute.

Pennsylvania Drinking Water Information System Database

Quality-control practices were applied to the PADWIS database, which involved checking spatial coordinates; verifying collection type information; excluding any sources not designated as wells; and verifying or removing data values that were either obvious errors or populated as zero rather than as “no data.” More detailed information about database QC can be found in Gross (2022). The database QC process resulted in a dataset containing 12,147 wells, consisting of 335 (3 percent) GUDI wells and 11,812 (97 percent) non-GUDI wells, hereafter referred to as the PADWIS database.

Most (around 63 percent) of the wells in the PADWIS database are non-community wells, 20 percent of which are designated as non-transient and 80 percent as transient. Community wells account for about 37 percent of the wells in the PADWIS database, and less than 1 percent of the total wells are designated as other system types, such as retail water facilities or bottled, bulk, or vended water. Most (90 percent) of the 335 wells classified as GUDI are community wells, whereas 64 percent of the 11,812 wells designated as non-GUDI are non-community wells.

According to the PADWIS database, the PADEP Northeast region contains the most wells (36 percent), the Northwest region contains the least (8 percent), and the other four regions have an intermediate number of wells (12 to 16 percent; fig. 2). The Northcentral region contains the highest percentage of GUDI wells, with 186 (13 percent) of the 1,410 wells in the region designated as GUDI. Of the total 335 wells in the PADWIS database classified as GUDI, 56 percent are in the Northcentral region, and 76 wells (23 percent) are in the Southcentral region (fig. 2). The Southeast region has only one GUDI well.

The SWIP status attribute in the PADWIS database (Pennsylvania Department of Environmental Protection, 2016) indicates 98 (approximately 29 percent) of the 335 GUDI wells were designated GUDI where SWIP monitoring was not required. Most of these wells are in the Northcentral region (fig. 3). Of the 98 wells without SWIP monitoring requirements, 78 (80 percent) had a SWIP finalization date between 1991 and 2000, indicating the classification of wells as GUDI without required SWIP monitoring occurred prior to 2001. In addition, the PADEP’s Guidance for Surface Water Identification Protocol (Pennsylvania Department of Environmental Protection, 2001) indicates that MPA testing may be conducted for susceptible sources prior to the SWIP monitoring; thus, an initial MPA sample may have indicated a moderate or high risk of susceptibility to surface-water influence.

The SWIP STATUS attribute in the PADWIS database also indicates “Not Evaluated–Filtered” for 121 wells, of which 108 were classified as GUDI and 13 classified as non-GUDI. These 121 wells essentially bypassed the SWIP because adequate treatment techniques were either already in place or were being installed in compliance with the SWTR. Of the 121 wells meeting these criteria, most were in the Northcentral (39 wells; 32 percent) and Southcentral (34 wells; 28 percent) regions, likely because these two regions have the highest number of GUDI wells statewide.

Pennsylvania Drinking Water Information System Database Subset

A subset of the PADWIS database was extracted to include only community wells that were indicated in the database to have undergone the SWIP to determine a source classification of GUDI or non-GUDI; this was an effort to only include sources in the study analysis that had been through a comparable evaluation process to receive GUDI or non-GUDI classification (Pennsylvania Department of Environmental Protection, 2010). Therefore, sources indicated in the PADWIS database with the SWIP STATUS attributes of “Completed—Monitoring Not Required” or “Completed—Source Monitored,” were included in this database subset.

This subset of the PADWIS database contains information for 4,018 community wells with 176 wells identified as GUDI and 3,842 sources identified as non-GUDI, hereafter referred to as the PADWIS database subset, with GUDI wells clustered in the north-central and central parts of the State (fig. 4). In addition to the spatial distribution of the PADWIS database subset, figure 4 also shows the statewide spatial distribution of this subset in relation to major aquifer type and physiographic provinces, which will be discussed in greater detail. The regional distribution of the 176 GUDI wells in this data subset includes a majority of GUDI wells in the Northcentral (140 [80 percent] of 176) and Southcentral (21 [12 percent] of 176) regions and the remaining four regions containing between 1 and 9 GUDI wells.

Microscopic Particulate Analysis Database Subset

MPA sample results were obtained from the PADEP Bureau of Laboratories (BOL) files, and associated water-quality data were obtained from the PADEP BOL, Sample Information System. All groundwater samples that generated analytical laboratory results and included in this study were collected by PADEP staff and analyzed by BOL.

A digital database of 2,327 MPA results for samples collected from public water-supply systems between 1990 and 2014 was provided by PADEP. This database included analytical MPA results determined from methods in US Environmental Protection Agency (1992), including bioindicator counts and associated risk-factor scores for groundwater samples. All samples were collected and analyzed after the introduction of the SWIP in the early 1990s. Each MPA sample result was stored with a unique sample identification value and binary variables specifying absence or presence of primary bioindicators. The digital database of MPA results did not contain numeric values for the actual counts of individual bioindicators.

In addition, PADEP provided 1,979 individual MPA digital result sheets containing a sample number, date, and numeric information on the actual counts of primary bioindicators per 100 gallons. These sheets contained assigned risk-factor scores for each bioindicator based on those counts. A database of the digital result sheets was created to store the provided results, and this database was manually related to the original MPA database using unique sample identification values. In each database, records with the same sample number, collection date, and results were assumed to be duplicates, and only one record was retained. Conversely, records with the same sample number but different collection dates, collection times, and results were differentiated by “A” and “B” qualifiers added to the end of these sample numbers. This process yielded a total of 1,749 samples that were able to be matched between the two MPA databases to create a combined MPA database containing both the absence or presence indicator data and the numeric values for bioindicator counts.

Various amounts of records for public water-supply systems in the combined MPA database were stored with three of the source’s attributes—PWSID, SYSTEM NAME, and (or) SOURCE ID; see table 1 for attribute definitions—from the PADWIS database to indicate the source being tested. These three attributes were not always exact matches between the combined MPA and PADWIS databases and contained inconsistencies in naming associated with water system name changes, consolidation of sources following the completion of the MPA, or general transcription errors. Therefore, sources from the combined MPA database were manually matched with associated PADWIS database records based on whichever of these three attributes were available for each source. A total of 835 MPA results for 699 public water-supply systems from the combined MPA database were able to be joined with their respective source record in the PADWIS database. Of the 699 public water-supply systems, 631 had COLLECTION TYPE attributes of “Well,” while two sources had COLLECTION TYPE attributes of “Other,” and 66 were designated as “Spring.” Only the 631 sources with COLLECTION TYPE attributes coded as “Well” in the PADWIS database were included from the combined dataset of MPA results to be used for statistical analysis because the site-specific conditions causing a well to have a high MPA result are expected to be different than, and not necessarily representative of, site-specific conditions for sources like infiltration galleries, Ranney wells, cribs, and springs (Pennsylvania Department of Environmental Protection, 2002b). These 631 wells, which were associated with 749 MPA results, resulted in a dataset with 84 GUDI wells (a total of 117 results with the number of results per source ranging from 1 to 4) and 547 non-GUDI wells (a total of 632 results with number of results per source ranging from 1 to 4).

Water-quality samples are typically collected from wells concurrently with the MPA sample and analyzed by the BOL. Water-quality parameters or constituents tested for and associated with the MPAs typically include some combination of the following list of parameters:

PADEP was able to compile and provide results for all or some of the previously listed water-quality parameters for the 749 MPA results that were matched with the PADWIS database. Results of analyses for these water-quality parameters were compiled from the BOL Sample Information System database that houses laboratory results for samples collected by PADEP’s Safe Drinking Water Program staff, including results for groundwater samples collected concurrently with the MPA sample. For 95 wells that were sampled more than once, the MPA and water-quality results from the most recent sampling event were retained for statistical analysis for this study, which resulted in the removal of 118 samples. Therefore, the MPA data analyses included in this study account for the most recent MPA collected from the well, regardless of the season in which it was collected.

The MPA dataset used for analysis consisted of MPA and water-quality results for 631 wells (84 GUDI, 547 non-GUDI), hereafter referred to as the MPA database subset, with results for water-quality parameters (see previous list) populated for 49 to 417 wells. Of the 631 wells included in the MPA database subset, 419 (66 percent) are community wells, and 212 (34 percent) are considered non-community (transient and non-transient) and other use types (bottled water and retail water facility).

For the MPA database subset of 631 wells, 169 (27 percent) had an MPA total risk-factor score exceeding zero (fig. 4), and 53 (8 percent) had a moderate (a score between 10 and 19) or high (a score of 20 or greater) risk MPA result classification (fig. 5). MPAs associated with the MPA database subset were collected between 1996 and 2014. The years with the highest amounts of MPAs collected were between 2003 and 2007 (292 [46 percent] of 631 MPAs) and are highlighted in figure 5. Years 2003–06 had the highest amount of MPAs with total risk-factor scores exceeding zero (82 [49 percent] of 169 MPAs) and 2006 had the highest number of moderate or high-risk MPA result classifications (10 [19 percent] of 53 MPAs) (fig. 5). Most of the MPAs were completed in the month of April (100 [16 percent] of 631 MPAs) and were also generally collected from April through July (319 [51 percent] of 631 MPAs), which were also the months with the highest number of MPA total risk-factor scores exceeding zero (94 [56 percent] of 169 MPAs) (fig. 6).

The regional distribution of the 631 wells with associated MPAs is like that of the total amount of GUDI wells, with the Northcentral and Southcentral regions having the highest number of wells with MPAs (fig. 7). Unlike the geographic distribution of total GUDI wells, the Southeast region had just one GUDI well (fig. 3). The Southeast region had 89 (14 percent) of 631 wells with MPAs, with 26 (29 percent) of the 89 MPAs exceeding zero and only 2 (2 percent) of the 89 MPAs classified as moderate or high risk (fig. 7). The Northcentral region, which had the highest number of total GUDI wells (186 [56 percent] of 335), also had the highest number of wells with MPAs (276 [44 percent] of 631). In the Northcentral region, 58 (21 percent) of 276 MPAs exceeded 0, and 22 (8 percent) of 276 MPAs were classified as moderate or high risk (fig. 7).

Spatial Data

Spatial data consist of numeric and binary variables, representing anthropogenic (land cover and PADEP region) and naturogenic (geologic and physiographic, hydrologic, soil characterization, and topographic) data, which were compiled and extracted for the 12,147 wells in the PADWIS database. These data add attributes to the database that are potential indicators of the influence of surface water on water in the wells (table 2). Numeric (continuous) variables represent conditions at the spatial coordinates associated with the well (for example, percentage of developed land cover within a specific radius of the well, or percentage of clay content in the soil at the well), whereas binary variables indicate whether a well is within or outside of a mapped feature (for example, in a carbonate major aquifer type in the Piedmont physiographic province). Spatial data included only those mapped features that could be represented in a geographic information system (GIS); the mapped features ranged in scale from 1:24,000 to 1:250,000.

Availability of Data Associated with the Surface Water Identification Protocol

Data attributes associated with the SWIP have varying availability in relation to the 4,018 wells in the PADWIS database subset. More information about the SWIP and how PADEP identifies GUDI sources can be found in PADEP documentation (Pennsylvania Department of Environmental Protection, 2001, 2002a, 2002b, and 2008). The initial screening evaluation performed by water-system operators of community wells consists of a review assessing site-specific hydrogeologic attributes (Pennsylvania Department of Environmental Protection, 2001), including:

Aquifer condition and well criteria

Surface-water separation

Integrity criteria

Of nine site-specific hydrogeologic attributes potentially required for the initial screening process, six attributes (CONFINED, CARB, STATIC WATER LEVEL[FT], AQUIFER DEPTH[FT], NHD200, KARST200) were either already available in PADWIS or could be populated from spatial data. The other three attributes, which include presence of a documented surface-water recharge boundary, proximity to a man-made feature that exposes the aquifer, and well-construction deficiencies, would need to be populated on a case-by-case basis from file information or a site visit. Of the six available attributes, the three spatial attributes (CARB, NHD200, and KARST200; table 2) could be populated for all 4,018 wells in the PADWIS database subset. Of the three PADWIS attributes already available in PADWIS (table 1), the STATIC WATER LEVEL(FT) attribute was the most commonly populated for the PADWIS database subset, with 2,509 (62 percent) of 4,018 wells containing data. The CONFINED attribute was the least populated, with only 1,575 (39 percent) of 4,018 wells from the PADWIS database subset containing these data (table 5). The three PADWIS attributes were most well populated in the PADEP’s Northcentral region, with between 73 and 95 percent of the 545 wells in the region containing data. Conversely, data for these three attributes were most poorly populated in the Northeast region, which is also the region containing the most wells, with between 2 and 51 percent of the region’s 1,079 wells having data for the three PADWIS attributes.

In addition to the previously discussed attributes that are considered in the decision to initiate the SWIP monitoring, seven relevant variables (hereafter referred to as criteria variables) associated with how conditions at a well meet, or do not meet, the criteria for wells to be monitored are listed in table 6. Attributes describing these seven criteria variables were assessed to determine how many wells had enough data to determine whether conditions at the well met the criteria or not; 2,855 (71 percent) of 4,018 wells had sufficient data to describe at least one of the seven criteria variables listed in table 6. For 1,163 (29 percent ) of 4,018 wells, there were not enough available data to determine if any of the criteria were met; therefore, it is infeasible to use the PADWIS database alone to attempt to deduce why these 1,163 wells would have been subject to the SWIP monitoring according to criteria defined by PADEP and would require review of individual information files for each well; 34 of these wells for which criteria variable data were not available were designated as GUDI. Of the total wells for which no criteria data were available, most of them (519 [45 percent] of 1,163) were in the Northeast region and the smallest number of such wells (21 [2 percent] of 1,163) were in the Northcentral region. Conversely, for 1,014 (25 percent) of 4,018 wells, data were able to be compiled regarding whether or not a well met each of the seven criteria; most of these wells (389 [38 percent] of 1,014) were in the Southcentral region, while the least number of wells (14 [1 percent] of 1,014) with data for all seven criteria variables were in the Northeast region. Also, 1,841 (46 percent) of 4,018 wells had enough available data to determine if between one and five of the criteria were met or not.

Of the total 2,855 wells with sufficient data to describe conditions at the well that met at least one of the seven criteria (table 6), 1,797 (63 percent) of 2,855 wells did not meet any of the criteria, 973 (34 percent) of 2,855 met one of the criteria, and 85 (3 percent) of 2,855 met two of the criteria. Non-GUDI wells accounted for most (1,746 [97 percent] of 1,797) of the wells not meeting any of the seven criteria, which means that these wells must have passed the initial screening process and been designated as non-GUDI. Conversely, 902 non-GUDI wells met one of the criteria, whereas 65 non-GUDI wells met 2 of the 7 criteria, which means 967 total non-GUDI wells (34 percent of the 2,855 wells with sufficient data) that met at least one of the criteria required further evaluation and passed SWIP monitoring and (or) MPA testing so as to be designated as non-GUDI. In addition, 51 GUDI-designated wells did not meet any of the 7 criteria, and it was unclear from data provided in the PADWIS database why these wells were determined to be GUDI and would require review of individual information files for these wells. Data describing the criteria listed in table 6, including the number of criteria met by each well, were included in further statistical analyses to determine if these criteria are a statistically significant measure of a well having been designated as GUDI.

Geologic Variables and Well Characteristics

The major aquifer and physiographic province (table 7; fig. 4) in which a public water-supply system well was located were major factors considered in the analysis of the susceptibility of the well to microbial contamination that would result in it being classified as GUDI or resulting in an MPA total risk-factor score exceeding zero. The highest percentages of both GUDI wells and wells with MPA total risk-factor scores exceeding zero were completed in a carbonate aquifer (table 7). The lowest percentages of GUDI wells and wells with MPA total risk-factor scores exceeding zero were completed in a crystalline aquifer.

Carbonate aquifers, which consist primarily of limestone and dolomite, occur most commonly in areas of low relief in the Piedmont and the Ridge and Valley Provinces in central and southeastern Pennsylvania (fig. 4). Of wells completed in carbonate aquifers, 57 (12 percent) out of 461 wells in the PADWIS database subset were GUDI and 63 (30 percent) out of 209 wells in the MPA database subset had a total risk-factor score exceeding zero (table 7). The highly soluble minerals that make up the carbonate rocks of these aquifers facilitate the enlargement of fractures along bedding planes as dissolution occurs, thus forming interconnected openings that enhance permeability (Lindsey and others, 2006). These large fractures can in turn lead to the formation of karst features, such as sinkholes and caves, through which groundwater can flow at high rates. Karst features, along with the overlying permeable soils commonly associated with carbonate rocks, have limited filtering capacity and permit rapid infiltration of water and contaminants from the land-surface to groundwater (Lindsey and others, 2006). In the Piedmont Province, Fishel and Lietman (1986) noted higher concentrations of pesticides and nitrate in wells completed in carbonate aquifers than in those completed in non-carbonate aquifers, and in a study of carbonate aquifers across the United States, Lindsey and others (2009) documented the highest concentrations of nitrate and the most pesticide detections in carbonate aquifers of the Piedmont and Ridge and Valley Provinces. Kozar and Paybins (2016) identified carbonate aquifers containing karst features in West Virginia’s Ridge and Valley Province as the most susceptible to contamination in the State, owing to the rapid ease with which precipitation and surface water carrying land-surface contaminants can infiltrate and provide recharge to the aquifer systems. Results from these studies demonstrate the potentially greater vulnerability of carbonate aquifers in the Piedmont and the Ridge and Valley Provinces to contamination from land-surface runoff. A study by Bickford and others (1996) reported that water from wells in areas underlain by carbonate aquifers in south-central Pennsylvania had higher concentrations of bacteria than water from wells in areas with other types of bedrock. Although water samples collected for this study were not analyzed for protozoan pathogens, the presence of fecal bacteria in the samples indicates the potential for protozoans, such as Cryptosporidium and Giardia lamblia, and other pathogens of fecal origin to occur in groundwater. In addition, Embrey and Runkle (2006) noted that the waters that most commonly tested positive for coliform bacteria were those in carbonate aquifers, including those within the Ridge and Valley Province, where samples from more than 50 percent of wells included in the study tested positive for these bacteria. Also, several studies (Kozar and Paybins, 2016; Nnadi and Fulkerson, 2002; Sharek, 1998) outside Pennsylvania identified carbonate aquifers containing karst features as those most susceptible to surface-water contamination. Sharek’s (1998) study in Florida revealed that wells with high- and moderate-risk MPA results were associated with karst features, such as sinkholes, and higher groundwater recharge rates. Wells with high- and moderate- risk MPAs in Florida were also located in areas with surficial geology consisting of exposed limestone, dolomite, and clayey sand (Nnadi and Fulkerson, 2002; Sharek, 1998).

Crystalline aquifers, consisting mainly of igneous and metamorphic rocks, are found within the New England, Piedmont, and Ridge and Valley Provinces in southeastern Pennsylvania (fig. 4; table 4). Of the wells completed in crystalline aquifers, just 1 well in the PADWIS database subset was GUDI, and 15 (23 percent) of 65 wells in the MPA database subset had a total risk-factor score exceeding zero (table 7). The crystalline aquifers in the New England Province in Pennsylvania are within the Reading Prong, which represents the southernmost extent of the province. The crystalline aquifers in the Piedmont Province consist of diabase intrusions of the Gettysburg-Newark Lowlands and crystalline rocks of the Piedmont Upland. Crystalline aquifers of the Ridge and Valley Province constitute South Mountain, which is the northernmost part of the Blue Ridge mountains and contains the oldest rock formations in the State. Lindsey and Bickford (1999) noted that in some cases, crystalline bedrock is as susceptible to contamination as areas underlain by carbonate bedrock.

Siliciclastic aquifers, which are made up primarily of sandstone, siltstone, conglomerate, and shale, comprise most Pennsylvania aquifers and are present in the Appalachian Plateaus, Piedmont, and Ridge and Valley provinces (fig. 4; table 4). Of the wells completed in siliciclastic aquifers, 84 (3 percent) of 2,784 wells in the PADWIS database subset were GUDI and 66 (26 percent) of 252 wells in the MPA database subset had a total risk-factor score exceeding 0 (table 7). The Appalachian Plateaus Province is underlain by relatively flat-lying sedimentary rocks and some bituminous coal, whereas the sedimentary rocks in the Ridge and Valley Province are highly folded and fractured and contain anthracite coal-bearing units. A study by Johnson and others (2011) in the Ridge and Valley Province noted that bacteria detections in water samples from wells completed in siliciclastic-rock aquifers were less frequent than detections in wells completed in carbonate-rock aquifers, and in siliciclastic-rock aquifers, bacteria were detected only in samples from wells in agricultural areas, thus illustrating the role of agricultural practices and animals as sources of groundwater contamination. The Piedmont Province siliciclastic aquifers occur in the Gettysburg-Newark Lowlands and include sedimentary rocks interbedded with basalt flows or intruded by diabase dikes and sills (Trapp and Horn, 1997). Lindsey and Bickford (1999) noted that of the major aquifer types in Pennsylvania, siliciclastic aquifers generally have the least potential for leaching of contaminants into groundwater than aquifers consisting of other bedrock types. This is primarily due to the poor infiltration capacity of soils weathered from shale and the fact that siliciclastic aquifers generally have smaller fractures and more complex flow paths than other aquifer types found in Pennsylvania (Lindsey and Bickford, 1999).

Surficial aquifers consist of unconsolidated sand and gravel and are found primarily in the Appalachian Plateaus, Atlantic Coastal Plain, Central Lowlands, and Ridge and Valley provinces (fig. 4; table 4). In the Appalachian Plateaus, Central Lowlands, and Ridge and Valley provinces, these aquifers are mainly glacial outwash and alluvial deposits associated with prior glaciation of the northwestern and northeastern tiers of Pennsylvania and in the Susquehanna River valley (Braun, 2004). Also, river terrace deposits in the Appalachian Plateaus Province of western Pennsylvania form surficial aquifers that border the Allegheny-Monongahela River (Lindsey and Bickford, 1999). In the Central Lowland Province, which borders Lake Erie in northwestern Pennsylvania, beach deposits as well as glacial sand and gravel that overlie shale bedrock serve as surficial aquifers. The Atlantic Coastal Plain in southeastern Pennsylvania consists of unconsolidated sand, gravel, and clay material that overlies metamorphic rocks. Water quality in the unconfined aquifer of the Atlantic Coastal Plain is typically related directly to local land use and varies spatially depending on anthropogenic activities on the land-surface (Knobel and others, 1998). In an assessment of microbial quality of groundwater in the United States, Embrey and Runkle (2006) found that the lowest detection frequencies (less than 5 percent) for coliform bacteria were in samples from wells completed in aquifers consisting primarily of unconsolidated sand, gravel, and clay materials. They determined that hydrogeologic characteristics, proximity of contaminating sources, interactions with surface water, or well-construction features (including the age of the well) were factors most likely controlling the presence and transport of coliform bacteria in the groundwater. Of wells completed in surficial aquifers in Pennsylvania, 34 (9 percent) out of 361 wells in the PADWIS database subset were GUDI, and 25 (24 percent) out of 105 wells in the MPA database subset had a total risk-factor score exceeding zero (table 7).

Differences in well-construction characteristics between GUDI and non-GUDI wells were analyzed for the PADWIS dataset subset. Well-construction characteristic differences were analyzed for the MPA database subset for wells with MPA total risk-factor scores exceeding zero and wells with MPAs classified as moderate or high risk. Data were also analyzed according to major aquifer types for wells specified in the PADWIS and MPA database subsets. Specifically, well construction year (CONSTRUCTION), casing length (CASING DEPTH[FT]), source depth (DEPTH[FT]), and static water level (STATIC WATER LEVEL[FT]) attributes (table 1) were analyzed. The non-parametric Kruskal-Wallis test was used to evaluate the statistical significance (alpha level of 0.05) of differences in central tendency or, more specifically, differences in mean ranks. If the calculated probability (p-value) is less than a specified alpha value of 0.05, there is a 95-percent probability that groups are significantly different, which means there is only a 1 in 20 chance that the observation is due to random variability. A p-value of less than 0.05 indicates the median rank value from observations in one category was statistically different from the other (Helsel and Hirsch, 2002).

Kruskal-Wallis test results showed median values for construction year, well depth, and static water level differ significantly between GUDI and non-GUDI wells (p <0.05), whereas median values for casing depth did not differ significantly between the two designations. Overall, GUDI wells had significantly older median construction years, had shallower source depths, and had static water levels closer to the land surface than non-GUDI wells (fig. 8). Analyzing GUDI and non-GUDI well-characteristic differences among carbonate and siliciclastic aquifers yielded nearly the same results as when all major aquifers were grouped together. The exception was that well casing length differed significantly between GUDI and non-GUDI wells (p <0.05) in carbonate aquifers, with GUDI wells having shorter casing lengths than non-GUDI wells. Well-characteristic data in crystalline and surficial aquifers were insufficient for analysis. Likewise, an analysis of GUDI wells completed in carbonate and siliciclastic aquifers showed significant differences in source depths; depths were shallower in carbonate aquifers than in siliciclastic aquifers.

Kruskal-Wallis test results showed significant differences in median values for construction year and casing depth between wells with MPA total risk-factor scores exceeding 0 and wells with MPA total risk-factor scores of zero (p <0.05); wells with MPA total risk-factor scores exceeding 0 had older median construction years and shallower casing depths (fig. 9). Similarly, Sharek (1998) examined well age as a contributor to high MPA risk indices, and the study concluded that newer wells use more improved materials than those available for construction of older wells and generally are subject to more stringent, modern construction standards. Therefore, deterioration associated with increased age of a well is also a factor that would contribute to a greater vulnerability and higher risk of contamination. Likewise, Nnadi and Sharek (1999) found that MPA total risk-factor scores generally increased with well age. Also, Sharek (1998) collected and analyzed samples and investigated characteristics of 62 wells in 7 Florida counties to determine contributors to high- and moderate-risk MPA results and found that wells with deeper casings had lower MPA risk indices, whereas wells with larger diameters had higher MPA risk indices due to the higher capacity of water that can be pumped from the well. Kozar and Paybins (2016) found that wells with casings deeper than 45 feet below the land surface had less bacterial contamination than wells with shorter casings.

Well construction characteristics did not differ significantly between wells that had moderate- or high-risk MPAs and wells that had low-risk MPAs. Also, data for well construction characteristics were insufficient to analyze wells in relation to categorical MPA total risk-factor scores according to major aquifer type, since only 53 wells had moderate- or high-risk MPAs, with that number decreasing further upon being divided according to four major aquifers. Differences in well-construction characteristics between wells with MPA total risk-factor scores exceeding zero and those with scores of zero were also analyzed according to carbonate and siliciclastic aquifers and yielded the same results discussed for all aquifer types. Differences in well characteristics were not analyzed for crystalline or surficial aquifers because each of these settings had less than 30 wells with MPA total risk-factor scores exceeding zero. No significant differences were found for wells with MPA total risk-factor scores exceeding zero whether they were completed in carbonate or siliciclastic aquifers.

Pennsylvania Drinking Water Information System Database Subset

Data from the PADWIS database subset, consisting of information for 4,018 wells, were analyzed using Spearman’s rho to evaluate the statistical significance of correlations between: (1) the designation of a community well as GUDI, and (2) well-characteristic (table 1), spatial (table 2), and criteria (table 6) variables. Correlations between variables that were statistically significant (p <0.0001) and with Spearman’s rho values greater than or equal to 0.1 or less than or equal to −0.1 are indicated in table 8. Of the 10 well-characteristic variables, a total of 3 had a statistically significant (p <0.0001) correlation with a well being designated as GUDI and had Spearman’s rho values greater than or equal to 0.1 or less than or equal to −0.1 (table 8). From the well-characteristic variables, aquifer confinement and average well yield had positive correlations for a well designated as GUDI, whereas static water level was negatively correlated with a well designated as GUDI. These findings indicate that wells designated as GUDI may be present in unconfined aquifers and have high average yield and shallow static water levels relative to the non-GUDI wells.

Six of the 37 spatial variables were statistically significant (p <0.0001) with Spearman’s rho values greater than or equal to 0.1 or less than or equal to −0.1 (table 8). The NE, SE, and SIL spatial variables had a negative relation with a well being designated as GUDI, which illustrates that wells assumed to be drilled in siliciclastic aquifers in the Northeast or Southeast PADEP regions have decreased potential of being designated as GUDI. This finding coincides with conclusions drawn by Lindsey and Bickford (1999), who noted that of the major aquifer types in Pennsylvania, siliciclastic aquifers generally have a lesser potential for leaching of contaminants into groundwater than do aquifers underlain by other bedrock types. In addition, the CARB, RVCARB, and NC spatial variables had positive correlations with a well being designated as GUDI, indicating that wells assumed to be drilled in carbonate aquifers are especially vulnerable to contamination from land-surface run-off (Bickford and others, 1996; Embrey and Runkle, 2006; Fishel and Lietman, 1986; Lindsey and others, 2006; Lindsey and others, 2009).

Also, four of the eight criteria variables were statistically significant (p <0.0001) with Spearman’s rho values of at least 0.10 (table 8). Wells completed in unconfined aquifers with static water levels less than or equal to 50 feet below the land surface (UNC50; table 6) and wells completed in carbonate aquifers with static water levels less or equal to 100 feet below the land surface (CARBLE100; table 6) had positive correlations with a well being designated as GUDI, and wells in confined aquifers with an aquifer depth less than or equal to 50 feet below the land-surface and located within 200 feet of a surface-water body (CONFNHD; table 6) had negative correlations with a well being designated as GUDI. Interestingly, the aquifer depth PADWIS attribute (AQUIFER DEPTH [FT]; table 1) used to create the statistically significant CONFNHD criteria variable (table 6) did not have a statistically significant correlation with a well being designated as GUDI. Also, the total number of the seven criteria (NumCriteriaMet; table 6) that are met by wells had a positive correlation with a well being designated as GUDI. Overall, these findings for the criteria variables are similar to those illustrated by the PADWIS attributes, which is expected because the criteria variables were created primarily on the basis of PADWIS attributes.

Microscopic Particulate Analysis Database Subset

The MPA database subset, consisting of 631 community and non-community wells, was examined to determine the significance of variables to the MPA total risk-factor score for each well. Ten well-characteristic variables (table 1), 37 spatial variables (table 2), 13 water-quality parameters associated with MPA sample collection, and 23 bioindicator variables associated with MPA results were analyzed using Spearman’s rho to assess correlation significance between these variables and the MPA total risk-factor scores and assigned hazard-level score (1-low, 2-moderate, 3-high).

Spearman’s rho computation was performed on rank-transformed water-quality variables rather than on actual values because the computation of this nonparametric statistic can include concentrations reported as less than the minimum reporting level (MRL). This enabled the use of all the data in nonparametric statistical analyses without making assumptions about the distribution of values less than the MRL, below the lower limit, or above the upper limit for the explanatory variables further described here (Helsel and Hirsch, 2002). Analytical results for nitrate, sulfate, and turbidity had MRLs of 0.04 mg/L, 15 and 20 mg/L, and 0.3 and 0.5 Nephelometric Turbidity Units, respectively. Reported nitrate concentrations below the MRL, reported sulfate concentrations below the highest MRL, and turbidity values below the highest MRL were censored to the highest MRL associated with each constituent or property. This enabled the use of all the data in nonparametric statistical analyses without making assumptions about the distribution of the data below the MRL for nitrate, sulfate, or turbidity (Helsel and Hirsch, 2002). Results of analyses for E. coli and total coliform were quantified with both less-than and greater-than values, bounding the data at upper and lower limits, whereas results of analyses for fecal coliform were quantified with greater-than values, bounding the data at an upper limit. E. coli results were quantified with less-than values of 0 and 1 CFU/100 mL and a greater-than value of 200 CFU/100 mL; total coliform results were quantified with less-than values of 0 and 1 CFU/100 mL and 6 greater-than values ranging from 80 to 2,400 CFU/100 mL; fecal coliform results were quantified with greater-than values of 60 and 200 CFU/100 mL. Analytical results for these three bacteria species were censored to the highest lower limit (1 CFU/100 mL for E. coli and total coliform) and the lowest upper limit (200, 80, and 60 CFU/100 mL for E. coli, total coliform, and fecal coliform, respectively) associated with each constituent. For MPA result bioindicator counts, counts of diatoms, coccidia, other algae, rotifers, and plant debris were quantified with greater-than values of 150, 300, 150, and 150 and 200, respectively, bounding the data at an upper count limit. Counts of bioindicators greater than these upper count limits were censored to the lowest associated value for each upper count limit for each parameter. Those variables that were statistically significant (p <0.0001) and had Spearman’s rho values greater than or equal to 0.1 or less than or equal to −0.1 are indicated in table 9.

Of the 10 well-characteristic variables, only one had a correlation with MPA total risk-factor score that was statistically significant (p <0.0001) with a Spearman’s rho value less than -0.10 (table 9). The aquifer depth variable, which describes the depth to the major water-bearing zone, had a negative correlation with MPA total risk-factor scores, suggesting that wells completed in aquifers with depths closer to the land-surface (shallower and relatively younger groundwater) had higher MPA total risk-factor scores than wells completed in aquifers with depths further from the land-surface (deeper and relatively older groundwater). Of the 37 spatial variables, two geologic variables (CARB and RVCARB; table 2) had statistically significant (p <0.0001) Spearman’s rho correlations. The statistical significance of wells completed in carbonate aquifers, especially in the Ridge and Valley Province, indicated that these aquifers are especially vulnerable to contamination and MPA samples with high total risk-factor and hazard-level scores.

Of the 13 water-quality parameters analyzed, fecal coliform and coliform had statistically significant positive correlations with both MPA total risk-factor scores and assigned hazard-level scores (table 9). Also, out of the 23 bioindicator variables analyzed, 17 had statistically significant positive correlations with MPA total risk-factor scores, while 20 had statistically significant positive correlations with assigned hazard-level scores. Of the bioindicator variables, the risk-factor score, sample count, and presence or absence variables for algae, rotifers, and diatoms had the highest positive correlations with both MPA total risk-factor scores and assigned hazard-level scores, suggesting that these bioindicators are more commonly present in MPA samples with high risk-factor and hazard-level scores than bioindicators such as insects/crustacea, plant debris, giardia, or coccidia. These results highlight those bioindicators that are more commonly found in Pennsylvania groundwater as part of the SWIP and those bioindicator variables that best correspond with MPA samples with high total risk-factor and hazard-level scores and thus could influence the designation of a well as GUDI.