Data Retrieval and Harmonization

Discrete surface-water-quality data for the Delaware River Basin were retrieved primarily from the Water Quality Portal, a national data repository which serves data collected by hundreds of Federal, State, and local organizations (National Water Quality Monitoring Council, 2019; Murphy and Shoda, 2020). Data were then harmonized in a process that includes steps such as: (1) selecting and combining appropriate water-quality constituent names; (2) identifying comparable and problematic methods; (3) standardizing text fields, such as lab remarks; (4) converting results to common units; (5) identifying proper fractionation; (6) assessing missing, zero, or unrealistic result values; (7) synthesizing remark codes and data quality comments; and (8) preferentially removing duplicate records. The purpose of this process was to combine data from multiple sources and create a reliable and comparable dataset for use in analysis and modeling (Shoda and others, 2019a). Many of these steps were based on the processes described in Oelsner and others (2017). A more detailed description of the data harmonization process for the Delaware River Basin dataset used in this analysis, including the R code used to retrieve and harmonize data, is available in Shoda and others (2019a). In addition to the data retrieved from the Water Quality Portal, data for 27 sites were obtained from the New York City Department of Environmental Protection for a subset of constituents. These data were harmonized following the same steps outlined in Shoda and others (2019a).

Selection of Water-Quality Constituents and Additional Data Processing

Four groups of constituents were chosen for trend assessment, based on their importance to Delaware River Basin stakeholders and the availability of existing guidance on data harmonization for these constituents. Salinity constituents were defined as specific conductance (SC) and total dissolved solids (TDS). The term “salinity” commonly refers to the dissolved salt content in water. It can be measured holistically with SC, which is directly related to the ion content of water, or by measuring TDS, which represents the ions and other dissolved solids in water. More specifically, the major ions that contribute to salinity (calcium, chloride, magnesium, potassium, sodium, and sulfate) can be measured directly and represent their own constituent group. In this report, salinity and major ions constituents are often discussed together because of their interrelated nature. Nutrient constituents were ammonia, nitrate, orthophosphate (filtered and unfiltered), total nitrogen (TN), and total phosphorus (TP), and suspended solids constituents were suspended sediment and total suspended solids (TSS).

For some constituents assessed in this study, data from filtered and unfiltered fractions were combined based on analyses in Oelsner and others (2017), who found a low percentage difference between the filtered and unfiltered concentrations of certain constituents in thousands of samples collected nationally, including in the Delaware River Basin. Oelsner and others (2017) combined data for samples with filtered and unfiltered fractions of ammonia, nitrate, nitrite plus nitrate, chloride, and sulfate. Although not all major ions analyzed for this report were directly assessed by Oelsner and others (2017), fractions were also combined for calcium, magnesium, potassium, and sodium. In contrast, the filtered and unfiltered fractions of orthophosphate were not suitable for combination and are referred to as separate constituents in this report. Table 1.1 lists the constituents and fractions used in this study and indicates which were combined. After this step, a few additional data processing steps were performed, including the selection of one value per site, constituent, and day (“data thinning”) and adjustments to censoring based on published technical memorandums. These steps are documented in appendix 1.

Site Clustering

The harmonized dataset contains water-quality data for 5,014 monitoring locations across the Delaware River Basin (Shoda and others, 2019a), each having at least one water-quality sample. Data from all locations were either combined to form “cluster sites” or used independently as single “sites” in this analysis. In general, each monitoring location has a unique identifier that is specific to a collecting organization and sampling location on a river or stream. However, some collecting organizations assign a new monitoring location identifier to the same sampling location to distinguish sampling for a new project or a different network. When multiple monitoring location identifiers are associated with the same location or when monitoring locations are closely located on the same river or stream, their sampling is capturing virtually the same streamflow and water-quality conditions. When appropriate, combining data from multiple monitoring locations can increase the length of record and (or) increase the data density within a record, both of which increase the robustness of the dataset used for the estimation of trends. Nevertheless, precautions must be taken when combining data from different collecting organizations. Data from one organization may be biased compared with the data from another organization, which could lead to an artificial trend in the data due solely to differences in sample collection or analysis procedures between the collecting organizations.

In this study, the term “site” describes a particular location or stretch of stream that may have data from a single monitoring location (“single site”) or from multiple nearby monitoring locations (“cluster site”). Data from multiple monitoring locations were considered for grouping into a cluster site if they were: (1) co-located (in other words, a collecting organization assigned more than one identifier to the same location or the same location on a river was monitored by more than one collecting organization); (2) on the same flowline, defined as a unique stream segment in the medium resolution National Hydrography Dataset V2 (McKay and others, 2012) (otherwise known as a ComID), and the flowline was less than 1.5 kilometer long; or (3) on the same flowline and the distances between monitoring locations were no more than 10 percent of the length of the flowline (fig. 2). In the third scenario, when many sites were on the same flowline, hierarchal clustering and a distance matrix were used to determine the best grouping of monitoring locations. The greatest distance between any two sites in a cluster was 1.6 kilometers.

Data Screening

Data for all sites and constituents were screened following the steps described in this section (each site-constituent combination is hereafter referred to as a water-quality record). All water-quality records that passed the screening described in the following “Water-Quality Data Screening” section had trends estimated with the SKT. Those records that passed additional screening described in the “WRTDS-Specific Screening” section also had trends estimated with WRTDS.

Selection of Four Common Trend Periods

As described previously, each water-quality record was required to meet the water-quality screening criteria, and the length of the POR was allowed to vary for each water-quality record. A POR trend was determined for each water-quality record using the earliest and latest year of the record that met the screening criteria. There was not a clear or obvious policy implementation or other change point to target for assessment in the Delaware River Basin, and the approach used in this report allows for an understanding of how conditions changed over the longest possible period. For some assessments, however, there is a need to compare trends for the same periods. Therefore, trends were also determined for four additional trend periods, as available data allowed: 1978–2018, 1998–2018, 2003–18, and 2008–18. These four trend periods were determined using an approach that maximized the number of sites and the length of record for all constituents. In a precedent set by Oelsner and others (2017) with the intent to include as many sites and constituents as possible for each trend period, we allowed screened water-quality records ending 1 year earlier than the end of the trend period (2017) and (or) beginning one year later than the defined start of the periods (1979, 1999, 2004 and 2009) to be considered. As such, the WRTDS results for some sites and constituents have start or end years that vary by 1 year for these four trend periods. Because of this 1-year variation allowance for WRTDS results, all SKT trends were run using 1979, 1999, 2004 and 2009 as start years for the four trend periods, though for ease of interpretation, we refer to the trend periods as 1978–2018, 1998–2018, 2003–18, and 2008–18. The results section of this report focuses only on the concentration trends for the POR and the most recent trend period (2008–18), with results for the remaining trend periods available in Murphy and others (2020).

Trend Estimation

A parametric and nonparametric trend estimation technique were each used in this analysis, and each had related strengths and weaknesses. The WRTDS trend estimation methodology is a parametric test that provides a host of outputs to understand the evolving nature of water quality, including daily, seasonal, and annual estimates of water quality; diagnostic tools to characterize water-quality trends through time; uncertainty estimates for trends and annual estimates; and techniques to explore relations between concentration and streamflow. Furthermore, WRTDS can accommodate data with multiple censoring levels and variations in sampling frequency. The power of this technique is accompanied by a need for streamflow data and water-quality samples collected during high-flow conditions, which often limit the applicability of this method. The SKT is a nonparametric test and compared with WRTDS, the SKT and the Theil-Sen slope become less robust with increasing amounts of censored data, cannot accommodate multiple censoring levels, provide no information about the evolving patterns of change through time, and do not explore how concentrations relate to streamflow, which lead to reduced accuracy and power of the test (Hirsch and others, 2010). However, SKT trend estimates are widely applicable due to having less stringent data requirements and requiring less computing power and expertise to execute.

Evaluation of Trends

To better understand water-quality trends in the Delaware River Basin, results were evaluated in several ways. First, trends determined with SKT and WRTDS were assessed to understand their direction, magnitude, and rate of change. Second, the trends were compared to water-quality LOCs to contextualize the changes and understand their environmental significance. For trends that were increasing or decreasing towards a LOC, proximity to the LOC was assessed, and the rate of change was projected into the future to assess the potential timeframes required to cross above or below the LOC, respectively. Third, trends were summarized by land use and the change in land cover within each watershed, and general spatial patterns of the trends were evaluated. Lastly, when SKT and WRTDS could be applied to the same water-quality records, the two methodologies were directly compared.

In this report, the results of only one method (SKT or WRTDS) for each site, constituent, and trend period are presented. When WRTDS results were available, they were used preferentially to SKT results because WRTDS provides more detailed output and accounts for the effects of streamflow on changing water quality. In the analyses in which SKT and WRTDS results are presented side-by-side, the SKT trends are referred to as “unadjusted” to remind the reader that they differ from the FN trends generated by the WRTDS model. Most analyses in this report are for the POR trends; a few analyses summarize more current changes, and these use the most recent trend period of 2008–18. Additionally, there was only one site that passed the screening to calculate trends for suspended sediment concentration, so those results are not included in the summaries below but are available in Murphy and others (2020).

Trend Direction and Magnitude

Trend directions were defined as “increasing,” “decreasing,” and “nonsignificant,” using a p-value less than or equal to 0.1 to define the level of significance. To understand the magnitude of significant increases and decreases for the POR trends, the percentage change was calculated for each constituent and trend method for which there was a significant trend. The calculation of percentage change is the difference between the annual concentrations of the end year and start year divided by the start-year concentration and multiplied by 100. For the WRTDS results, percentage change was calculated using the estimated FN annual mean concentrations of the start and end years. For the SKT trends, percentage change was calculated by first estimating the start-year and end-year concentrations. The end-year concentration was calculated as the median concentration of the two most recent years of samples (for example, 2017 and 2018), and the start-year concentration was determined by projecting the Theil-Sen slope back from this end-year concentration to the earliest year in the POR. This approach for calculating start- and end-year concentrations by using the slope of the Theil-Sen line and the median concentration of the last 2 years of the record to determine the intercept of the line, better characterizes potential nonlinear changes compared to the more common approach (Helsel and others, 2020) of using the median concentration and median year of the trend period to determine the intercept (fig. 3). In the two cases in which this approach generated a negative start-year concentration, the lowest nonnegative estimated start concentration was used.

Description of Trend Estimations

Eight collecting organizations contributed water-quality data to the input dataset published in Murphy and others (2020), and each organization contributed data for between 1 and 16 constituents. The sizes of these contributions varied; organizations contributed between 1 and 2,128 data values per site and constituent. Overall, non-USGS organizations contributed about half of the water-quality data values to the input dataset. In addition to direct data contributions, other organizations often partner with the USGS to fund, or partially fund, water-quality and streamflow data collection that is executed by the USGS. In the final trend results dataset described extensively in this report (1 trend per site, constituent, and trend period), each organization contributed water-quality data to between 3 and 797 water-quality records that met data-screening criteria (table 2). Of the 1,160 POR trends, about a third (345) were generated with water-quality data from more than one organization.

There were 124 sites in the Delaware River Basin that passed the water-quality data screening for at least one constituent (fig. 4). Streamflow data from 57 USGS streamgages were used in WRTDS trend estimation, and 59 sites passed the additional WRTDS screening for at least one constituent. Many sites have trend results for multiple constituents with results from SKT, WRTDS, or both. It is common that at the same site, the length of water-quality records and the quality of those records differ between constituents. Thus, trend results are specific to each site, constituent, trend period, and method. As stated previously, only the results for one method were used per constituent, site, and trend period in this report, and WRTDS was used preferentially over SKT. The number of sites with POR trends varies depending on the method used and the constituent (table 3).

For most constituents, trend sites were generally well distributed across the basin, but several constituents have a dense cluster of sites in northern Delaware (figs. 1.5–1.19). The geographic distribution of trend sites, however, was directly related to the organizations that contributed data to this analysis and where their monitoring programs were, so some constituents are less well distributed. The New York City Department of Environmental Protection, for example, monitors the upper Delaware River Basin at sites upstream from some reservoirs that provide drinking water to New York City, and they are one of the few organizations that contributed filtered orthophosphate data, therefore sites with trends for filtered orthophosphate were exclusively in the northernmost part of the basin (fig. 1.10). Likewise, the USGS Pennsylvania Water Science Center was the only organization that contributed unfiltered orthophosphate data that passed the screening, therefore sites with these trends are only in Pennsylvania (fig. 1.11). Murphy and others (2020) contains details on which organizations contributed data for which constituents and sites. Sites with TDS trends are not in the upper Delaware River Basin (fig. 1.16), and sodium and potassium trends were generally not west of the main stem of the Delaware River (figs. 1.14 and 1.12). Sodium, in particular, had only three sites in the Lehigh and Schuylkill River subbasins (fig. 1.14).

Period of Record Trend Direction and Magnitude

Overall, the significant POR trends for nutrients (orthophosphate, TN, nitrate, TP, and ammonia) tended to be mostly decreasing, and the significant trends for the salinity constituents (SC and TDS) and some major ions (sodium, chloride, and calcium) were largely increasing (fig. 5). For both trend methods, the constituents with the largest percentage of sites with increasing trends over the POR were indicators of salinity: SC, TDS, sodium, chloride, and calcium (fig. 5). Likewise, for both methods, the largest percentage of decreasing trends were observed for sulfate, orthophosphate (filtered and unfiltered), and TN (fig. 5). TSS, ammonia, and TP shared the largest percentage of nonsignificant trends for both methods (fig. 5). Potassium and magnesium shared a fairly even distribution of sites with increasing, decreasing, and nonsignificant trends for both methods. Overall, the WRTDS and SKT trend results were largely in agreement, although for calcium, magnesium, nitrate, potassium, and unfiltered orthophosphate the trend direction with the largest number of sites was not consistent between methods, and the highest percentage of sites was nonsignificant for one method and significant for the other method.

Trend magnitudes followed a pattern similar to trend directions: the largest positive changes were for salinity constituents, chloride, and sodium and some of the largest negative changes occurred for nutrient constituents (fig. 6). For both methods, chloride, sodium, TDS, SC, calcium, and magnesium had the largest positive median percentage changes: their values ranged from 10 percent to 78 percent. Nutrient constituents, sulfate, and TSS had negative median percentage changes for both methods, ranging from −16 percent to −58 percent. The entire IQR of percentage changes for chloride, sodium, TDS, and SC was greater than zero for both methods, whereas the IQRs for ammonia, TSS, TP, sulfate, nitrate, and TN were less than zero for both methods. Additionally, sodium and chloride were in the top three constituents with the largest variation in percentage change (shown by IQRs) for both methods, indicating that percentage change can vary substantially between sites for these constituents and some magnitudes of change are much greater than the median (fig. 6).

To understand patterns in changing water quality, it is necessary to interpret the percentage of sites with a significant trend direction and the magnitude of those significant trends. Although nutrient trends were generally decreasing rather than increasing or being nonsignificant, the percentage of sites with significant decreases in nutrient constituents tended to be less than the percentage of sites with significant increases in salinity constituents (fig. 5), although nutrient trend magnitudes can be comparable to those for TDS and SC (fig. 6). Sulfate, with the largest percentage of sites with decreasing trends, had median percentage changes less than −34 percent for both methods. Although ammonia and TSS had some of the largest magnitude decreases, most trends for these constituents were nonsignificant (figs. 5 and 6). Potassium was the only constituent with a positive median percentage change for one method (WRTDS) and negative for the other (SKT), but that is to be expected because, as noted above, the trend direction with the largest number of sites differed between methods. For potassium, both of the median percentage changes were small and the IQR of percentage changes straddled zero.

The trends assessed in this section are for the POR, which varies for each water-quality record. Thus, the length of time represented by each increasing, decreasing, and nonsignificant trend may be different, and the water-quality records assessed might represent different periods (although all periods end in 2017 or 2018). The POR trends were used here because the scope of this analysis is intended to be comprehensive and to capture all potential long-term changes in the water quality of the Delaware River Basin. We sought to summarize not just the most recent changes (2008–18 trend period), but to understand change as far back in time as it was possible at all sites. The results of these same analyses for identical periods, the 2008–18 trend period, are in the appendix (figs. 1.1 and 1.2), and the results for the 2008–18 trend period and the POR mostly agree.

Evolution of Changes in Water Quality

Assessing the rates of change for constituents with significantly decreasing or increasing trends can illustrate whether those changes are happening more quickly or slowly over time. The 5-year rates of change (milligram per liter per 5-year subperiod) for sulfate were all negative, and there were no positive rates at any site throughout the subperiods investigated, showing a very consistent decreasing trend direction for this constituent. Additionally, negative rates of change for sulfate decreased between 1998 and 2013, indicating faster decreases during this period, and those decreases continued, but at a slower rate after 2013 (fig. 7). The median rates of change for TP were negative for the subperiods spanning from 1993 to 2008, but the negative changes slowed down over time, with rates getting closer to zero, until the 2013–18 subperiod, during which the median rate of change shifted from negative to positive. This shift indicated that, by the 2013–18 subperiod, a larger number of sites had positive rates of change (increases in TP annual mean FN concentration), compared to negative rates of change (decreases in TP annual mean FN concentration). The pattern in negative rates of change (decreases) for TN and nitrate was similar for the subperiods investigated, which was to be expected because nitrate is often the majority component of TN. These parameters had decreasing median rates that largely leveled off to roughly a constant rate of decrease between 2003 and 2018 (fig. 7).

The constituents with primarily increasing trends (SC, chloride, TDS, and sodium) tended to display increasing rates of change over the 25 years investigated, indicating increases have been occurring at faster rates over this time. The median rates of change for TDS and sodium, in particular, showed noticeable increases between 2003 and 2018, whereas SC increases were steady, but more subtle, and chloride median rates were more variable in this period. The IQRs of the rates of change for the constituents with increasing trends tended to be wide in the more recent subperiods, for example, 2013–18, indicating higher variation in the rate of increases for these years, and potential rates of change much higher than the median, perhaps possible to assess because there were a greater number of sites in these years (fig. 7). For each of the eight constituents, however, wide IQRs were present for many subperiods, demonstrating considerable variation in rates of change between sites, although some general tendencies across sites did exist.

Level of Concern Comparisons

Figure 8 shows the results of categorizing the POR trends relative to the LOC for both trend methods. POR start- and end-year concentrations for sulfate, nitrate, ammonia, and chloride were all below the LOC, with the exception of one site with an SKT trend for nitrate, for which start- and end-year concentrations were above the LOC (fig. 8). Sulfate, nitrate, and ammonia trends were largely decreasing away from the LOC or were nonsignificant. Most of the chloride trends were increasing towards the LOC. Both methods had sites with TDS trends (3 sites with WRTDS trends and 1 site with an SKT trend) that increased above the LOC over the POR, representing an environmentally relevant degradation in water quality (fig. 8). Three of these sites were on the same tributary to the Delaware River, and all were in Pennsylvania. The WRTDS results showed 77 percent of chloride trends and 38 percent of TDS trends increasing towards the LOC. Similar percentages were apparent for SKT results with 66 percent of chloride trends and 47 percent of TDS trends also increasing towards the LOC, ultimately representing a potential threat to water quality.

Start and end concentrations for TN and TP trends were above the LOC at most sites, but between 28 percent and 58 percent of sites were trending downward toward the LOC (specifically, 58 percent of trends determined with WRTDS and 40 percent determined with SKT for TN and 41 percent of WRTDS trends and 28 percent of SKT trends for TP), indicating the potential for continued improvements in water quality at those sites (fig. 8). Trends in TN and TP produced the highest percentage of sites that crossed the LOC threshold, both from above and below the LOC. For TN, five sites (13 percent) with WRTDS trends and four sites (7 percent) with SKT trends decreased from above to below the LOC during the POR, demonstrating improved water quality. For TP, only one site with an SKT trend decreased from above to below the LOC. Notable degradations in water quality were seen for one TP site with a WRTDS trend (Neversink River near Claryville, New York; fig. 1.18B) and one TN site with an SKT trend (Brodhead Creek near East Stroudsburg, Pennsylvania; fig. 1.17A), which increased from below to above the LOC (fig. 8), although these particular trends were nonsignificant, indicating that the water-quality records used to determine the trends might have been highly variable or sparse.

Identifying environmentally important changes that occurred over the longest period possible (POR trends) helps to place long-term changes in context (fig. 8), but to assess current water-quality threats and successes, understanding more recent change is also necessary. The 2008–18 trends that were approaching an LOC from either above (improving) or below (degrading) were investigated to understand how close the trend end concentrations (2018) were to the LOC. Of the 105 trends that were increasing toward the LOC from below, only five sites (5 percent) had end concentrations within 60 percent of the LOC, with trends in chloride, TN, TP, and TDS (fig. 9). For the 37 site constituents that were decreasing towards the LOC from above, only 3 sites (8 percent) had end concentrations within 160 percent of the LOC, with trends in TN and TP (fig. 9). Overall, 94 percent of trends from both methods were greater than 160 percent of the LOC or less than 60 percent of the LOC. Thus, even though some constituents had many trends approaching the LOC, many of the sites had concentrations that were still much higher or lower than the LOC, indicating that between 2008 and 2018, immediate degradation or improvement was not likely for many of the water-quality sites and constituents.

To understand how water quality might change in the future, the rate of change for the 2008–18 trends approaching the LOC in the “below LOC, increasing” and “above LOC, decreasing” categories was projected into the future to estimate the number of years to reach the LOC. For increasing trends in sulfate, nitrate, ammonia, chloride, and TDS that are approaching the LOC, the median number of years to reach the LOC was greater than 80 regardless of trend method, indicating that for most of these sites and constituents, the rates of change are small and there was no immediate threat of crossing the LOC (fig. 10). For chloride and TDS, however, notable outliers are observed, with the number of years to reach the LOC dropping as low as 12 for one site with a chloride SKT trend (Valley Creek at Wilson Road near Valley Forge, Pa.) (fig.10). Furthermore, the one TN trend (WRTDS result) and the one TP trend (SKT result) that were increasing towards the LOC had end concentrations near the LOC (fig. 9) and large enough rates of change to indicate only 2.5 and 11 years, respectively, to potentially reach the LOC (fig. 10).

Alternately, for TN trends decreasing towards the LOC, the median number of years to reach the LOC was less than 50, and for TP trends decreasing towards the LOC, the median number of years to reach the LOC was less than 8, indicating that decreases in TN and TP concentrations at these sites might cause the ecoregional nutrient criteria to be met relatively soon (fig. 10). In summary, most of the 2008–18 TN and TP trends that were decreasing towards the LOC had faster rates of change compared to the trends increasing towards the LOC, producing a comparatively short timeframe to achieve water-quality improvement (fig. 10).

Land-Use Assessment and Land-Surface Change

Of the 124 unique sites used in this analysis, 9 changed land-use categories at some point during the POR. In all cases, these nine sites became more urbanized. As of 2017, most sites were mixed use (63), about half that many were undeveloped (32), 22 sites were urban, and only 7 sites were agricultural (table 4). When subdivided by trend method, the relative frequency of sites in land-use categories is similar (table 4). Because of the small number of agricultural sites, water-quality trends at these sites will not be discussed.

Across all constituents, undeveloped sites tended to have water-quality trends that were lower in magnitude and had less variability between sites compared to mixed-use and urban sites (fig. 11). For SC, TDS, and most major ions, mixed-use and urban sites consistently had increasing trends, and urban sites had some of the largest increases, except for sulfate trends which decreased over the POR for sites in all land-use categories. For nutrients and TSS, mixed-use and urban sites tend to be decreasing but less consistently than the increases seen for the salinity constituents and most of the major ions (fig. 11). For percentage change, trends at undeveloped sites have relative magnitudes of change more similar to mixed-used sites (fig. 12). This is particularly evident for constituents such as TN, nitrate, TSS, SC, TDS, chloride, and sodium, indicating small magnitude changes in concentration observed at undeveloped sites are sizable compared to their starting concentrations (fig. 12).

Across all sites, cumulative land-surface change ranged from 0 to 49 percent, with a median of 10 percent, indicating that at half of the sites at least 10 percent of a given watershed had land-surface changes over the POR (fig. 13). This metric captures a range of land-surface changes, from permanent changes such as the expansion of suburban development, to transient changes such as the destruction of vegetation and infrastructure due to severe weather. The metric also captures multiple changes to the same spatial area over the POR, and it includes changes that may not cause a watershed to change land-use categorizations. Determining specifically what land-surface changes occurred in the watersheds included in this report would require an indepth look at additional land-surface imagery over time, a task not completed as a part of this analysis. Unsurprisingly, urban sites had the largest cumulative changes, followed by mixed-use and agricultural sites; undeveloped sites mostly had cumulative land-surface changes less than 5 percent over the POR (fig. 13). When plotted against the concurrent water-quality POR trend, cumulative land-surface change had a strong positive relation with all salinity-related constituents and major ions, except sulfate (fig. 14), indicating sites with more land-surface changes in the watershed also had larger increases in concentration for these constituents. For sulfate, decreases were consistent across all magnitudes of cumulative land-surface change, showing no relation between land-surface change and change in sulfate concentration. Similarly, TN and nitrate trends were unrelated to cumulative land-surface change. While TP and TSS showed no relation between land-surface change and the SKT trend, the WRTDS trend results showed negative relations with some scatter for these constituents (fig. 14). The plots of water-quality change versus cumulative percentage of land-surface change for the 2008–18 period are in the appendix (fig. 1.4).

Spatial Summary

Strong relations between geographic location and trend direction or magnitude were not observed in the POR trends for most constituents (figs. 1.5–1.19). Calcium, nitrate, orthophosphate (filtered and unfiltered), TDS, TN, and TP had no spatial patterns, or weak patterns. Chloride had large magnitude increases throughout the basin (fig. 1.7), and likewise, sodium and SC were increasing consistently basinwide (figs. 1.14 and 1.13). Significant SKT trends for ammonia were clustered in the southern part of the basin and were mostly decreases (fig. 1.5). Where there were significant trends for magnesium and potassium, the decreases tended to occur in the upper basin and increases in the lower basin, although the magnitude of these changes was small (figs. 1.8 and 1.12). Decreasing TSS trends estimated with SKT are evident in the southern part of the basin (fig. 1.19), whereas the largest decreases in sulfate tend to occur in the northern part of the Delaware River Basin (fig. 1.15). The POR trends do not represent the same period for each constituent and site, however, so this might make spatial patterns more difficult to detect.

Comparison of WRTDS and SKT Trends

As discussed above, for some sites, constituents, and trend periods, the criteria for both SKT and WRTDS were met, and trend results were generated with both methods. For the 993 results in which trend direction was available for both methods, 75 percent of the results agreed, generating significant trends in the same direction or nonsignificant trends with both methods (table 5). The other 25 percent of results produced significant trends with one method, but nonsignificant trends with the other (table 5). The methods produced significant trends in conflicting directions only once. The number of results in each trend direction comparison group, for each constituent, are in the appendix (table 1.2). In addition to the general agreement of trend direction between methods, when the two methods produced significant trends in the same direction, they also tended to result in a comparable percentage change. For the pairs of results in which the methods agreed on significance and trend direction, 70.4 percent of the pairs had a difference in percentage change that was less than 10, and an additional 18.6 percent of the pairs had differences in percentage change greater than 10 but less than 20 (table 6). The absolute differences in percentage change by constituent are in the appendix (tables 1.3). Note that due to data censoring, sometimes a tau value but not a sen slope was reported for SKT results (as discussed in the “Trend Estimation” section). Therefore, because SKT sometimes allows for the calculation of trend direction, but not percentage change, the number of results with significant trend direction agreement does not match the number of results that can be assessed for trend magnitude agreement.

Salinity and Major Ions

Chloride and SC were ranked among the top four constituents with the largest number of sites with water-quality data for both methods, indicating that they are among the best represented constituents in this analysis (table 3). Figure 7 showed generally increasing rates of change for these constituents, indicating that increasing trends are occurring at faster rates over time. Chloride, for example, had increasing trends with changes as high as 200 percent and rates of change increasing by more than 15 milligrams per liter per 5-year subperiod at some sites. Annual mean chloride concentrations, however, were still greatly below the 250 mg/L LOC (a secondary drinking water standard) at all sites, and the median number of years to reach the LOC was predicted to be greater than 100. In contrast, TDS showed the greatest cause for concern in the LOC analysis because concentrations at four sites increased above the LOC (also a secondary drinking water standard) during the POR, a notable percentage of sites (43 percent) were increasing towards the LOC, and two sites had trend-end concentrations (2018) relatively near the LOC (within 40 percent of the LOC). Calcium, magnesium, and potassium are represented in this analysis with 36 to 50 SKT trends and 19 to 35 WRTDS trends, depending on the constituent. These major ions have relatively even proportions of increasing, decreasing, and nonsignificant trends (fig. 5), and their changes are generally +/−25 percent and distributed around zero (fig. 6), indicating that water-quality changes in these constituents are small, variable between sites, and not of great concern basinwide. The greatest total changes per year and percentage changes per year for the salinity constituents, chloride, sodium, calcium, and magnesium occurred mostly at sites with urban land use, although sites with mixed and undeveloped land use were occasionally at comparable levels. Cumulative land-surface change had a strong positive relation with water-quality change for the salinity constituents, chloride, sodium, calcium, and potassium, indicating that changes on the land surface may be one of the factors contributing to widespread increases in salinity and these major ions. Increasing trends in SC, chloride, and sodium were basinwide, illustrating the widespread nature of this change.

Water availability for human health and aquatic life is threatened by increasing salinization that degrades the quality of freshwater. This degradation can threaten drinking water sources, which is of particular concern for the Delaware River Basin, which supplies drinking water to over 13 million people (Delaware River Basin Commission, 2020a). Assessing increases in chloride and the chloride-sulfate mass ratio are two ways that water-quality trends can be used to evaluate the potential for stream water to create corrosive conditions in drinking water infrastructure (Stets and others, 2018). Corrosion of water distribution systems can cause lead, copper, and other metals to leach into drinking water (Ng and Lin, 2016) and present a potential public health issue or the need for more costly drinking water treatment (Cañedo-Argüelles and others, 2016). The results presented in this report show widespread increases in surface-water chloride concentrations (fig. 5, fig. 1.7). Of the 21 sites with chloride and sulfate POR trends, 13 sites (62 percent) had increasing chloride trends and decreasing sulfate trends, indicating the potential for an increasing chloride-sulfate mass ratio.

Elevated salinity can also cause lethal and sublethal effects on aquatic communities (Hart and others, 1991). These effects are especially relevant in the Delaware River Basin, which supports a trout fishery and endangered fish and mussel species (Delaware River Basin Commission, 2020a). Water-quality degradation is a serious threat to these species, including the endangered Alasmidonta heterodon (dwarf wedgemussel) and other freshwater mussels, in the Delaware River Basin, which are sensitive to many contaminants, including chloride (U.S. Fish and Wildlife Service, 2011). Sodium chloride and calcium chloride have also been shown to have effects on the growth of trout species (Hintz and Relyea, 2017).

The salinity constituents (TDS and SC) had trend patterns similar to sodium and chloride, suggesting sodium and chloride were the source of overall salinity changes (that increases in TDS and SC may be caused by increases in chloride and sodium). A likely source of these major ions in the Delaware River Basin is sodium chloride deicing salt. Our finding the greatest change in SC, TDS, chloride, and sodium at sites with urban land use agrees with Stets and others (2018), who noted a high potential for increased corrosivity in urban areas. Additionally, the strong correlation between land-surface change and water-quality change for salinity and most major ions indicates that increasing urbanization and deicing salt application in urban areas might be a driver of these trends in the Delaware River Basin.

Previous studies in the Northeast United States, and the Delaware River Basin specifically, have noted similar increases in salinity-related constituents over time (Kaushal and others, 2005; Oelsner and others, 2017; Delaware River Basin Commission, 2019). This study found widespread increases in SC, TDS, chloride, and sodium, and found that increases in the salinity constituents and sodium are occurring more quickly in recent years. Continued salinization of rivers and streams in the Delaware River Basin at a quickening pace suggests the need for more urgent salinity management across the Delaware River Basin, including the need to definitively determine and mitigate the drivers of changes in salinity concentrations. Continued and additional monitoring of sodium in the western Delaware River Basin and TDS in the upper Delaware River Basin would provide information to fill the spatial gaps in this analysis. Calcium, magnesium, and potassium do not generally present water-quality concerns in the Delaware River Basin, although their continued study can advance the understanding of the drivers of salinity changes, ion exchange and mineral dissolution, and of groundwater/surface-water interactions, which can all affect water quality and have been noted as an area of interest in the Delaware River Basin (Fischer, 1999; USGS, 2020).

Alternately, sulfate trends have strong widespread decreases: more than 85 percent of sites for both trend methods show decreases (fig. 5) and exhibit some tendency to decrease at faster rates in recent years (fig. 7). The start and end concentrations of sulfate trends were below the LOC at all sites (fig. 8). At two sites, trends were increasing towards the LOC, but trend projections show greater than 2,000 years would be required to reach the LOC if the rate of change from 2008 to 2018 continued into the future (fig. 10). The decreases in sulfate concentrations seen in Delaware River Basin rivers and streams have been observed throughout the United States in response to decreases in the atmospheric deposition of sulfur related to the enactment of the Acid Rain Program, established under the Clean Air Act (EPA, 2020b). Instream changes in sulfate concentration and watershed land-surface changes were not related in our analysis (fig. 14), which suggests that atmospheric deposition was the primary driver of change. Despite general decreases of instream sulfate concentrations throughout the U.S., applications of sulfur to agricultural fields to assist in plant uptake of nitrogen have increased, in response to the reductions in atmospheric sulfate deposition and thus reductions in sulfate available for plant metabolism (Hinckley and others, 2020). There is record of sulfur application for these purposes in the Northeastern U.S., which has the potential to counteract decreasing trends, and might be the reason for the slowing rate of decrease for sulfate in the Delaware River Basin shown in figure 7 (Ketterings and others, 2012).

Nutrients

Several nutrient constituents (ammonia, nitrate, TN, and TP) are well represented in this analysis (42 to 63 SKT trends and 22 to 40 WRTDS trends), depending on the constituent. Orthophosphate and TN trends were among the constituents with the largest percentages of decreasing trends (fig. 5), but all nutrient constituents had negative median percentage changes (fig. 6), although these constituents can have a substantial percentage of sites with nonsignificant trends (fig. 5). Decreasing rates of change were observed for nitrate and TN, leveling out to a relatively constant rate of decrease in their instream concentrations by 2018 (fig. 7). A high percentage of trend-period start and end concentrations were above the LOC for TN and TP, although several sites had POR trends that decreased below the LOC, indicating meaningful water-quality improvement (fig. 8). If rates of change were to remain constant, every site with a TN trend decreasing towards the LOC would cross below the LOC threshold in less than 75 years, and this estimate would be 17 years for TP (fig. 10).

Nutrients threaten natural waters in many ways, including contributing to harmful algal bloom prevalence (which can cause serious human health consequences) and excessive algal growth (which upon senescence, can decrease dissolved oxygen concentrations critical for the health of fish and aquatic communities) (EPA, 2020a). In the Delaware River Basin, elevated concentrations of nitrogen have been identified as a potential cause of water-quality degradation in the Delaware Estuary (Delaware River Basin Commission, 2020b). In response to this threat, nutrient management initiatives addressing the main stem of the Delaware River and Estuary were established, and the recent status of nutrients in the Delaware River Basin has been reported as “very good” (Delaware River Basin Commission, 2019). Our analysis of long-term water-quality trends supports this assessment by showing generally decreasing concentrations, which indicates ongoing nutrient reduction strategies may be successful. Despite these general decreases, the LOC analysis showed that the majority of TN and TP concentrations in 2018 were greater than the EPA ecoregional nutrient criteria. These criteria represent a starting point for States to use in developing water-quality standards that protect freshwater systems from the effects of excess nutrient input (EPA, undated c). Exceedances of these criteria indicate that continued nutrient monitoring and management in the Delaware River Basin is warranted. In particular, orthophosphate monitoring is not well distributed spatially in the basin, and it was possible to estimate trends in orthophosphate at only a small number of sites, even with the less stringent SKT criteria.

Suspended Solids

As noted in the “Methods” section, only one site passed the criteria to generate a trend for suspended sediment concentration, so changes in sediment were represented in this analysis by TSS, for which 83 SKT and 32 WRTDS trends were calculated. TSS had the largest percentage of nonsignificant trends for all constituents with WRTDS trends (91 percent) and many nonsignificant SKT trends (59 percent) (fig. 5). When significant, TSS trends were decreasing with large negative percentage changes (fig. 6). Across both methods, only one TSS trend showed a significant increase over the POR. Nonsignificant trends occur for many reasons, such as generally stable water-quality concentrations over time, highly variable water-quality concentrations (noisy data), sparse data records given the variability of the data, and difficulty of the model to characterize the patterns of variability. Furthermore, sediment is notoriously difficult to model because most sediment is transported during high streamflows, which are typically less sampled than lower flows. Additionally, compared to suspended sediment concentration determinations, TSS determinations are less accurate, less precise, and are often biased low when there is a high fraction of suspended sand in the water column. TSS can poorly represent suspended sediment in natural waters (Gray and others, 2000): a potential explanation for the high proportion of nonsignificant trends for this constituent. However, an alternative explanation may be that TSS has been largely stable over the POR in the Delaware River Basin, and any small changes that may have occurred were not detected.

Understanding changing sediment in the Delaware River Basin is important for maintaining the health of surface waters in the basin, from the Delaware estuary to the upper most headwaters. The erosion of sediment from stream banks in the basin is a major source of new sedimentation in the estuary, and the Delaware Estuary Regional Sediment Management Plan notes the need to “fully characterize sediment processes and outputs” in the basin (Delaware Estuary Regional Sediment Management Plan Workgroup, 2013). Additionally, 90 percent of the water supply to New York City originates in the Catskill and Delaware watersheds, with the Cannonsville, Pepacton, and Neversink Reservoirs falling within the Delaware River Basin (Committee to Review the New York City Watershed Protection Program, 2020). These reservoirs are the largest unfiltered water supply source in the United States, and the maintenance of their water quality is critical in maintaining a potable water supply to the largest city in the United States. Therefore, understanding the trends and current conditions of turbidity is a top priority for the New York City Watershed Protection Program (Committee to Review the New York City Watershed Protection Program, 2020). Turbidity and TSS are strongly related, so the decreasing TSS trends in the Delaware River Basin are a positive result for watershed protection and potential sediment transport, indicating potential progress has been made in reducing sediment erosion in the Delaware River Basin. Furthermore, the large percentage of nonsignificant trends may also be a cautiously positive result, indicating that at many sites sediment concentrations may have changed very little and historical conditions have been maintained over time. Even in light of improving or maintained TSS conditions in many streams across the Delaware River Basin, continued data collection, especially at high flows, may allow for more robust trends to be generated in the future, and using suspended sediment concentration determinations in place of TSS may increase the accuracy and precision of the data and the ability to detect smaller changes over time.

Trend Method Strengths and Weaknesses

As described in the “Introduction” and “Methods” sections, the two trend methods used in this study have strengths and weaknesses to consider in the interpretation of their results. However, the analysis in the “Comparison of WRTDS and SKT trends” section indicates that, in terms of trend direction and magnitude, the methods produce mostly comparable results (tables 5 and 6). Across all combinations of site, constituent, and trend period used in the WRTDS-SKT comparison, 19 percent of the combinations had a significant trend detected using WRTDS but not using SKT, likely because controlling for streamflow allows WRTDS to identify a trend that may have been otherwise obscured due to variable weather and streamflow conditions. For the 6 percent of results in which SKT detected a significant trend in water quality, but WRTDS did not, it is likely these trends are due to several higher-than-average or lower-than-average flow years at the beginning or end of the trend period, causing an apparent trend in water quality. Because these apparent trends might be entirely due to variable streamflow (in other words, some years are wetter or drier than others), when flows return to more average conditions in coming years, they are likely to dissipate. The flow-normalization techniques used in WRTDS to characterize trends guard against these apparent, temporary trends by accounting for the interannual and often periodic variability of precipitation and streamflow, something SKT is incapable of capturing. Ultimately, conflicting trend directions occurred at only a single site, indicating the comparability of the results generated in this analysis. The magnitude of change is also important in assessing the status of water quality in a basin. Trend magnitudes estimated with WRTDS and SKT were largely consistent, with 89 percent of the compared results generating percentage changes that were within 20 percent of each other (table 6).

The WRTDS and SKT trend estimation techniques use distinctly different approaches for estimating trends, in particular, the inclusion of streamflow as a covariate with WRTDS. Despite these different approaches, the overall agreement of trend direction and magnitude in the Delaware River Basin indicates that SKT might be a suitable method to assess changing water quality when streamflow data is not available. Although WRTDS allows for a much more refined and certain understanding of changing water quality, analysts with limited data availability might be willing to use SKT trend estimation instead, considering the general agreement of results in the Delaware River Basin presented in this report. Tables 1.2 and 1.3 provide more information on the reliability of SKT results for specific constituents. The overall comparability of the trend results might indicate that streamflow does not substantially affect changes in water quality in the Delaware River Basin and that watershed management, such as changes in point and nonpoint sources, might be driving changing water quality.