Constituent concentrations were compared to Federal and State drinking water-quality benchmarks to provide context (table 1). Benchmarks were selected in the following order of priority: United States Environmental Protection Agency (USEPA) or SWRCB-DDW maximum contaminant level (MCL) or action level (AL), which ever had the lowest concentration (24 constituents), SWRCB-DDW secondary maximum contaminant levels (SMCL; the upper SMCL was used for constituents with lower and upper recommended values; 5 constituents), HAL (2 constituent), then SWRCB-DDW notification level – response level, NL-RL (1 constituent) [USEPA, 2018; SWRCB-DDW, 2018a,b]. Sample concentrations (C) are defined as “high”, “moderate”, and “low” relative to the benchmark concentration (B): High C > B; Moderate B/2 < C ≤ B; Low C ≤ B/2.

Six constituents did not have a benchmark, but these constituents may contribute to or explain trends of other constituents with benchmarks (table 1). For example, calcium does not have a benchmark but contributes to total dissolved solids (TDS) concentrations. Therefore, a calcium trend may help explain TDS trends.

Table 1. List of water-quality constituents analyzed for trends with the number of wells with at least one sample, the constituent screening level, and water-quality benchmark.

The Mann-Kendall (MK) rank correlation [Kendall, 1938], which is a non-parametric, rank-based statistical test, and Sen’s slope estimator [Sen, 1968] were used to assess trends in water quality data. Trends were accepted as statistically significant when MK rank correlation p-values were below a significance level (α) of 0.05 and the Sen’s slope estimator was not zero. Positive Sen’s slopes indicate increasing concentrations while negative slopes indicate decreasing concentrations. Tests were computed using the Python scripting language [PSF, 2016] for constituents at wells with 4 or more unique laboratory analyses (number of analyses minus number of equal values or ties).

Before a statistical test was applied, water-quality data were processed to reduce biases in trend detection caused by serial correlation, changing reporting levels, and from seasonal patterns. In general, the most common detection level reported with the SWRCB-DDW data was used as a truncation level such that non-detections and concentrations below the truncation level were recoded to the most common detection level for each constituent listed in table 1. Non-detect values above the truncation level were removed from the dataset. To reduce the effects of serial correlation and to test for trends in data that display significant water-quality differences among two pumping seasons, water-quality data was classified as a Summer sample if the sample date was between May 1^st and October 31 ^st or a Winter sample if the sample date was outside the Summer date range. For each season, the median concentration and date was computed for all summer and winter samples when more than one result was measured in a season. This method produces at most two data points for each year.

Tests for trends were applied to different time periods to identify long-term trends (LTT), recent trends (RT), reversals in trends (TRV) and trends that have seasonal concentration differences (figure 1). The entire period of recorded data was used to identify LTTs while RTs were evaluated with water quality data collected since the year 2000. LTTs and RTs were computed for datasets with 4 or more unique laboratory analyses.

Reversals of trends (TRV) show a change in trend direction either from decreasing to increasing or from increasing to decreasing. TRVs were computed for datasets with at least 8 unique laboratory analyses spanning at least 8 years. TRVs were determined by looking for opposite trends in two continuous segments; one segment from the oldest data and one segment from the newest data. To determine if a change in slope occurred, the MK test was computed multiple times by incrementally varying the size of the oldest (S_old=i) and newest (S_old=N-i) segments, where i goes from 4 to the number of data points (N). Because this analysis can produce multiple sets of segments with TRVs around the inflexion point, the set of newest data with the largest change in trend slope was reported (figure 1). This procedure identifies trends that have reversed direction once over the entire period of record rather than trends with frequent reversals caused by variability over shorter durations (<8 years).

Seasonal trends can result from cyclical periods of pumping and non-pumping that cause changes in the water sampled by a well differences [Bexfield and Jurgens, 2014]. Trends can be masked, or the rate of change can be over/under-estimated by seasonal differences in water-quality data [Hirsch et al., 1982; Helsel and Hirsch, 1995]. Seasonality was identified using the Mann-Whitney test for differences between seasonal populations of water-quality data when there were at least four unique analyses in each season. If differences in concentrations between seasons were significant, MK rank correlation and Sen’s slope estimator were computed for each set of seasonal data. A seasonal trend was statistically significant if at least one MK test p-value was below the significance level and the Sen’s Slope estimator was not zero. This approach to seasonal trends is different than the computation by the Seasonal MK trend test, which is a sum of the individual Kendall’s S statistic among seasons and generally requires trends to be in the same direction for most seasons to be significant.

Figure 1. Examples if (A) long-term, (B) recent, (C) reversing, and (D) seasonal trends.

Each well was scored for the concentration (C) relative to its benchmark (B) and scored for the magnitude and direction of the Sen Slope (SS) trend relative to half the benchmark. The time to half the benchmark (T_hb), concentration score (S_C), and trend score (S_T) is calculated as:

Spatial weighting was used to determine the areal proportion of the groundwater resource with different classes of degradation or improvement in a study area or physiographic province. Spatial weighting counteracts biases caused by differences in the spatial density of wells, so that areas with higher densities of wells or more frequent sampling will receive the same weight as other grid cells with lower densities of wells [Belitz et al., 2010].

Physiographic provinces, study areas, and grid cells were the same as those used by Belitz et al. [2015]. Study areas correspond to California Department of Water Resources groundwater basins [Bulletin 118] and some areas outside of groundwater basins. Each study area was divided into a network of equal-area grid cells. The study areas investigated by Belitz et al. [2015] included 95% of the area statewide where public-supply wells (PSWs) are located and 99% of the population supplied by PSWs. Therefore, the gridded area is essentially the entire area of the groundwater resource used for public supply in California.

A cell score, S_cell, was calculated for each grid cell by aggregating the concentration and trends scores for the wells in the cell (Nw):

Cell scores that are negative indicate that water-quality trends are predominately improving while positive cell scores indicate that water-quality trends are predominately degrading. When positive and negative trends scores for wells in a cell are equal, the cell score is zero or indeterminate. Cell scores can be computed using any of the trend tests determined above; however, only recent trends results were used to compute cell scores because they provide the most recent picture of groundwater quality trends statewide.

Cell scores were grouped into one of nine classifications: 1) not tested; 2) no trend; 3) improving (decreasing trends) with high concentrations; 4) improving with moderate concentrations; 5) improving with low concentrations; 6) degrading (increasing trends) with low concentrations; 7) degrading with moderate concentrations; 8) degrading with high concentrations; 9) indeterminate. Negative scores indicate concentration trends are predominantly decreasing (improving) in an area while positive scores indicate increasing concentrations (degrading).

Belitz, K.B., Jurgens, B., Landon, M.K., Fram, M.S., Johnson, T., 2010, Estimation of aquifer scale proportion using equal area grids: Assessment of regional scale groundwater quality. Water Resources Research vol. 46, W11550, DOI: 10.1029/2010WR009321

Belitz, K.B., Fram, M.S., Johnson, T.D., 2015, Metrics for assessing the quality of groundwater used for public supply, CA, USA: Equivalent-population and Area. Environmental Science and Technology, 49, 14, 8330-8338. DOI: 10.1021/acs.est.5b00265

Bexfield, L.M. and B.C. Jurgens. 2014. Effects of seasonal operation on the quality of water produced by public-supply wells, Groundwater p 15. DOI: 10.1111/gwat.12174.

California Department of Water Resources. California’s Groundwater—Bulletin 118, Update 2003; California Department of Water Resources: Sacramento, CA, 2003; http://www.water.ca.gov/groundwater/bulletin118/index.cfm.

California State Water Resources Control Board – Division of Drinking Water (SWRCB-DDW), 2016; EDT Library and Water Quality Analyses Data and Download Page; http://www.waterboards.ca.gov/drinking_water/certlic/drinkingwater/EDTlibrary.shtml.

California State Water Resources Control Board, 2018. Division of Drinking Water, Chemicals and Contaminants in Drinking Water Website, accessed February 21, 2018 at: https://www.waterboards.ca.gov/drinking_water/certlic/drinkingwater/Lawbook.shtml.

California State Water Resources Control Board, 2018. GAMA – Groundwater Ambient Monitoring and Assessment Program Website, accessed February 21, 2018 at: https://www.waterboards.ca.gov/gama/geotracker_gama.shtml.

Helsel, D.R., Hirsch, R.M., 1995. Statistical Methods in water resources. Elsevier Science, p. 529.

Hirsch, R. M., Slack, J. R., Smith, R.A., 1982, Techniques of trend analysis for monthly water quality data. Water Resources Research, 18 (1), 107-121. DOI: 10.1029/WR018i001p00107.

Hirsch, R. M., & Slack, J. R., 1984. A non-parametric trend test for seasonal data with serial dependence. Water Resources Research, 20, 727–732. doi: 10.1029/WR020i006p00727.

Jurgens, B., Jasper, M., Nguyen, D.H., and Bennett, G.L., 2018, USGS CA GAMA-PBP Groundwater-Quality Results--Assessment and Trends: U.S. Geological Survey website, at https://ca.water.usgs.gov/projects/gama/water-quality-results/.

Kendall, M.G. 1975. Rank correlation methods. Griffin. London.

Koterba, M.T., F.D. Wilde, and W.W. Lapham. 1995. Ground-water data-collection protocols and procedures for the National Water-Quality Assessment Program—Collection and documentation of water-quality samples and related data. U.S. Geological Survey Open-File Report 95-399, 113. Reston, Virginia: USGS.

PSF, 2016. Python Software Foundation. Python Language Reference, version 3.5. Available at http://www.python.org

Sen, P.K. 1968. Estimates of the regression coefficient based on Kendall’s tau. Journal of American Statistical Association, 63, 1379-1389.

U.S. Geological Survey, variously dated, National field manual for the collection of water-quality data: U.S. Geological Survey Techniques of Water-Resources Investigations, book 9, chaps. A1-A10, available online at http://pubs.water.usgs.gov/twri9A/.

U.S. Geological Survey, 2018, National Water Information System data available on the World Wide Web (USGS Water Data for the Nation) at URL [U.S. Environmental Protection Agency. 2018a. Safe Drinking Water Act Amendments of 1996; Office of Ground Water and Drinking Water; Washington, DC, 1996. https://www.epa.gov/sdwa (accessed March 2018).

U.S. Environmental Protection Agency. 2018b. Drinking water standards and regulations: National Primary Drinking Water Regulations. Website, accessed February 21, 2018 at: https://www.epa.gov/dwstandardsregulations.U.S.

Geological Survey, 2018a, National Water Information System: Web Interface Website, accessed February 21, 2018 at: https://www.waterboards.ca.gov/gama/geotracker_gama.shtml.

AL: action level
GAMA-PBP:USGS Groundwater Ambient Monitoring and Assessment – Priority Basin Project (GAMA-PBP)
LTT: long-term trends
MCL: maximum contaminant level
MK: Mann-Kendall
N: Nitrogen
NAWQA: USGS National Water Quality Assessment
NL-RL: notification level – response level
PSWs: public-supply wells
RT: recent trends
SWRCB: State Water Regional Control Board
SWRCB-DDW: California State Water Resources Control Board Division of Drinking Water
TDS: total dissolved solids
TRV: reversals in trends
USEPA: United States Environmental Protection Agency