Despite decades of effort toward reducing nitrogen and phosphorus flux to Chesapeake Bay, water‐quality and ecological responses in surface waters have been mixed. Recent research, however, provides useful insight into multiple factors complicating the understanding of nutrient trends in bay tributaries, which we review in this paper, as we approach a 2025 total maximum daily load (TMDL) management deadline. Improvements in water quality in many streams are attributable to management actions that reduced point sources and atmospheric nitrogen deposition and to changes in climate. Nutrient reductions expected from management actions, however, have not been fully realized in watershed streams. Nitrogen from urban nonpoint sources has declined, although water‐quality responses to urbanization in individual streams vary depending on predevelopment land use. Evolving agriculture, the largest watershed source of nutrients, has likely contributed to local nutrient trends but has not affected substantial changes in flux to the bay. Changing average nitrogen yields from farmland underlain by carbonate rocks, however, may suggest future trends in other areas under similar management, climatic, or other influences, although drivers of these changes remain unclear. Regardless of upstream trends, phosphorus flux to the bay from its largest tributary has increased due to sediment infill in the Conowingo Reservoir. In general, recent research emphasizes the utility of input reductions over attempts to manage nutrient fate and transport at limiting nutrients in surface waters. Ongoing research opportunities include evaluating effects of climate change and conservation practices over time and space and developing tools to disentangle and evaluate multiple influences on regional water quality.

In the Chesapeake Bay watershed, estimated fluxes of nutrients and sediment from the bay’s nontidal tributaries into the estuary are the foundation of decision making to meet reductions prescribed by the Chesapeake Bay Total Maximum Daily Load (TMDL) and are often the basis for refining scientific understanding of the watershed-scale processes that influence the delivery of these constituents to the bay. Two regression-based flux and trend estimation models, ESTIMATOR and Weighted Regressions on Time, Discharge, and Season (WRTDS), were compared using data from 80 watersheds in the Chesapeake Bay Nontidal Water-Quality Monitoring Network (CBNTN). The watersheds range in size from 62 to 70,189 square kilometers and record lengths range from 6 to 28 years. ESTIMATOR is a constant-parameter model that estimates trends only in concentration; WRTDS uses variable parameters estimated with weighted regression, and estimates trends in both concentration and flux. WRTDS had greater explanatory power than ESTIMATOR, with the greatest degree of improvement evident for records longer than 25 years (30 stations; improvement in median model R2= 0.06 for total nitrogen, 0.08 for total phosphorus, and 0.05 for sediment) and the least degree of improvement for records of less than 10 years, for which the two models performed nearly equally. Flux bias statistics were comparable or lower (more favorable) for WRTDS for any record length; for 30 stations with records longer than 25 years, the greatest degree of improvement was evident for sediment (decrease of 0.17 in median statistic) and total phosphorus (decrease of 0.05). The overall between-station pattern in concentration trend direction and magnitude for all constituents was roughly similar for both models. A detailed case study revealed that trends in concentration estimated by WRTDS can operationally be viewed as a less-constrained equivalent to trends in concentration estimated by ESTIMATOR. Estimates of annual mean flow-adjusted (ESTIMATOR) and flow-normalized (WRTDS) concentration for years initially constituting the end of a water-quality record showed a similar degree of variability as data for additional years were incrementally added and the initial estimates “aged.” On the basis of the results of this broad comparison of the two models, the U.S. Geological Survey is adopting WRTDS as the primary model for estimating constituent fluxes and trends throughout the CBNTN. Nutrient and sediment flux and trend estimates, based on WRTDS, are summarized narratively and tabulated in appendixes for all stations for which fluxes or trends were reported through water year 2012. WRTDS also was used to explore the sensitivity of flux and trend estimates to three data-quality issues common in many large-scale monitoring networks and evident in some of the CBNTN records. The potential effects of inconsistency in annual sampling effort and inconsistency in storm sampling effort were explored by way of a subsampling experiment using eight of the most densely sampled long-term (1985–2012) stations in the CBNTN as baseline datasets. From each dataset, a set of 10 “design guideline” subsamples was selected, consisting of 12 monthly samples and 8 targeted storm samples per year. The selection was conducted in a manner that preserved the overall intensity of storm sampling in the baseline data. These 10 subsamples were further manipulated to create “heterogeneous” subsamples by removing storm samples prior to 2003. The maximum relative difference between flow-normalized flux estimated in a single year from any of the 10 design guideline subsamples and values estimated in the corresponding year from baseline data was smallest for dissolved inorganic nitrogen (median of 8 stations = 6 percent of baseline estimate), but more appreciable for total phosphorus and sediment (medians of 22 and 32 percent, respectively). The maximum relative difference between flow-normalized flux estimated from from the 10 heterogeneous subsamples and values estimated in the corresponding year from baseline data was more pronounced, with medians for 8 stations of 15, 30, and 53 percent of the corresponding baseline estimates for dissolved inorganic nitrogen, total phosphorus, and sediment, respectively. The worst-case maximum relative differences between flow-normalize flux estimated in a single year from the 10 heterogeneous subsamples and values estimated in the corresponding year from baseline data were 25 percent for dissolved inorganic nitrogen, 37 percent for total phosphorus, and 250 percent for sediment. The results for the heterogeneous subsamples indicate that changes in storm sampling frequency can result in appreciable distortion of estimated trends in flow-normalized flux, especially for total phosphorus and sediment. Trend lines estimated from heterogeneous subsamples tended to converge with the trend lines estimated from baseline data after 2003. In contrast, 2003–12 trends based on subsamples truncated by discarding all data prior to the induced heterogeneity in 2003 showed appreciable biases and differences in slope, relative to the corresponding 2003–12 segment of the trend computed from the design guideline subsamples. Overall, the results indicate that for particulate constituents, load and trend estimates computed using long-term records recently converted to CBNTN design guideline sampling protocols will be most reliable if the trend is computed using the entire record, but reported only for the period that design guideline sampling protocols were followed. Inconsistencies related to changing laboratory methods were also examined via two manipulative experiments. In the first experiment, increasing and decreasing “stair-step” patterns of changes in censoring level, overall representing a factor-of-five change in the laboratory reporting limit, were artificially imposed on a 27-year record...

An understanding of riverine water‐quality dynamics in regional mixed‐land use watersheds is the foundation for advances in landscape biogeochemistry and informed land management. A differential implementation of the statistical/process‐based model SPAtially Referenced Regressions on Watershed attributes (SPARROW; Smith et al., https://doi.org/10.1029/97wr02171) is proposed to empirically relate a regional pattern of changes in flow‐normalized constituent flux, over a multiyear period, to contemporaneous changes in spatially referenced explanatory variables. In a pilot application, the differential model, called Spatiotemporal Watershed Accumulation of Net effects (SWAN), is used to explore factors influencing changes in flow‐normalized flux of total nitrogen over the period 1990–2010 at 43 sites in the nontidal Chesapeake Bay watershed. A seven‐parameter model explains 80% of the transformed variability in independently estimated flux changes, indicating that storage effects having characteristic time scales greater than 20 years had a small influence, relative to changes in inputs, on regional water‐quality response. Results suggest that 1990–2010 changes in total‐nitrogen flux are largely the outcome of increased nonpoint‐source pollution associated with urban and suburban development, modulated to the point of negation by terrestrial losses stemming from widespread increases in air temperature and precipitation. The loss mechanism is qualitatively consistent with denitrification; however, increases in aboveground biomass, agricultural nitrogen exports, or hydrologic flushing are also plausible contributors. Although qualified by a small sample size and constraints on explanatory data availability, the pilot suggests that SWAN is a promising approach for broadening scientific understanding of factors driving regional water‐quality change and for supporting evidence‐based land‐management decisions.

Many agricultural watersheds rely on the voluntary use of management practices (MPs) to reduce nonpoint source nutrient and sediment loads; however, the water‐quality effects of MPs are uncertain. We interpreted water‐quality responses from as early as 1985 through 2020 in three agricultural Chesapeake Bay watersheds that were prioritized for MP implementation, namely, the Smith Creek (Virginia), Upper Chester River (Maryland), and Conewago Creek (Pennsylvania) watersheds. We synthesized patterns in MPs, climate, land use, and nutrient inputs to better understand factors affecting nutrient and sediment loads. Relations between MPs and expected water‐quality improvements were not consistently identifiable. The number of MPs increased in all watersheds since the early 2010s, but most monitored nutrient and sediment loads did not decrease. Nutrient and sediment loads increased or remained stable in Smith Creek and the Upper Chester River. Sediment loads and some nutrient loads decreased in Conewago Creek. In Smith Creek, a 36‐year time‐series model suggests that changes in manure affected flow‐normalized total nitrogen loads. We hypothesize that increases in nutrient applications may overshadow some expected MP effects. MPs might have stemmed further water‐quality degradation, but improvements in nutrient loads may rely on reducing manure and fertilizer applications. Our results highlight the importance of assessing MP performance with long‐term monitoring‐based studies.

In 2007, the U.S. Geological Survey partnered with Fairfax County, Virginia, to establish a long-term water-resources monitoring program to evaluate the hydrology, water quality, and ecology of Fairfax County streams and the watershed-scale effects of management practices. Fairfax County uses a variety of management practices, policies, and programs to protect and restore its water resources, but the effects of such strategies are not well understood. This report used streamflow, water-quality, and ecological monitoring data collected from 20 Fairfax County watersheds from 2007 through 2018 to assess the effects of management practices, landscape factors, and climatic conditions on observed nutrient, sediment, salinity, and benthic-macroinvertebrate community responses.Urbanization, climatic variability, and an increase in management practices occurred within Fairfax County during the study period. Impervious cover, housing units, wastewater infrastructure, and (or) stormwater infrastructure increased in most study watersheds. Climatic conditions varied among study years; countywide estimates of average-annual air temperature differed by about 3 degrees Celsius, and total precipitation ranged from about 34 to 63 inches per year. The effects of the management practices, implemented to reduce nitrogen, phosphorus, and (or) sediment loads, are considered in this study. These management practices primarily consist of stormwater retrofits and stream restorations; however, stream restorations account for most of the financial investment and expected load reductions. Management practices were implemented in half of the study watersheds, and most practices were installed and reductions credited late in the study period.Changes in hydrologic response during storm events were evaluated over the study period because many management practices that were implemented were designed to achieve nutrient and sediment reductions by slowing or intercepting runoff. The average number and length of storm events was mostly unchanged throughout the monitoring network. Four watersheds with 10 years of streamflow data showed a mixture of trends in stormflow peak, volume, and rate-of-change. Event-mean nutrient and sediment concentrations from these watersheds were evaluated during storm events and generally showed increases in total phosphorus (TP) and suspended sediment and reductions or no changes in total nitrogen (TN).Landscape inputs of nitrogen and phosphorus and the percentage of inputs delivered to streams were estimated for the study watersheds. Estimated phosphorus from fertilizer and nitrogen from atmospheric deposition represented large nutrient inputs in most watersheds; amounts of other nonpoint sources varied based on land use. Estimated nitrogen inputs declined throughout Fairfax County and in most study watersheds from 2008 through 2018; in comparison, phosphorus input changes were relatively small. Most nonpoint-nutrient inputs were retained on the landscape and did not reach streams, with slightly more nitrogen retention than phosphorus, on average. Retention rates were lower for years with more precipitation and streamflow. After adjusting for streamflow, TN and TP loads were generally higher for years with more nutrient inputs. Calculated as a function of flow-adjusted loads, TP retention declined at most stations from 2009 through 2018, in comparison, TN retention was relatively unchanged.Landscape and climatic conditions affected spatial differences and changes in Fairfax County stream conditions from 2009 through 2018. TN concentrations were higher and increases over time were larger in watersheds with elevated septic-system density. TP concentrations were higher in watersheds with more turfgrass; concentrations were lower, but had larger increases over time, in watersheds with deeper soils. Suspended-sediment concentrations were higher in watersheds with greater stream densities. Specific conductance was higher in watersheds with more developed land use and shallower soils. Benthic-macroinvertebrate index of biotic integrity (IBI) scores were lower in watersheds with high road density and had larger increases over time in bigger, more developed watersheds. Annual variability in TN and TP concentrations and benthic-macroinvertebrate IBI scores was affected by precipitation; annual variability in suspended sediment concentrations and specific conductance was affected by air temperature.After accounting for influences from landscape and climatic conditions, expected management-practice effects were not consistently observed in monitored stream responses. These effects were assessed by comparing expected management-practice load reductions with the timing, direction, and magnitude of changes in storm-event hydrology, nutrient and sediment loads, median-annual water-quality conditions, and benthic-macroinvertebrate IBI scores. An important consideration for future investigations of management-practice effects is how to control for water-quality and ecological variability caused by geologic properties, the urban environment, precipitation, and (or) air temperature. The interpretation of management-practice effects in this report was likely influenced by a combination of factors, including (1) the amount, timing, and location of management-practice implementation; (2) unmeasured landscape and climatic factors; (3) uncertain management-practice expectations; (4) hydrologic variability; and (5) analytical assumptions. Through continued data-collection efforts, particularly after management practices have been completed, many of these factors may become less influential in the future.