Hydroclimatic variability and change in the Chesapeake Bay Watershed

Endeavors to ecologically restore and protect Chesapeake Bay (CB), the largest estuary system on the east coast of the United States, are challenging under a changing climate (Pyke & Najjar 2008). Since the streamflow carries large quantities of nutrients from uplands into the estuary, quantifying streamflows in the CB Watershed becomes critical (Neff et al. 2000; Kimmel et al. 2006; Miller & Harding 2007; Najjar et al. 2009). The future variability in temperature and precipitation will directly affect the water budget components of the CB Watershed including groundwater flow, evapotranspiration (ET), and surface runoff. Knowledge about altered hydrologic characteristics caused by climate change is also necessary to properly manage the water resources of this region.

The CB is the largest and the most productive estuary in the United States and provides 5% of US fisheries (Najjar et al. 2010), and its watershed (165,000 km²) is home to more than 17 million residents and includes many metropolitan areas such as Washington, DC and Baltimore. The CB region has chronic water quality issues including low dissolved oxygen level, excessive chlorophyll a and chemical contaminant concentrations and poor water clarity (Kemp et al. 2005). Excessive nutrient loading from multiple sub-watersheds in the CB Watershed has been recognized as one of the major reasons (Hagy et al. 2004). Efforts to reduce pollutant sources in the watershed have continued for close to three decades (Shenk et al. 2012).

Recent studies have extensively investigated this watershed to better understand the hydrologic and water quality issues to achieve sustainability (USEPA 2010; Shenk et al. 2012). However, there is only limited knowledge about the changing hydrology of the CB Watershed and this hampers the management of this basin. Downscaled climate variables from improved general circulation models (GCMs) coupled with a physically based hydrological modeling approach is essential for any realistic evaluation. The watershed has undergone enormous land use and population changes which also increases the complexity. The Susquehanna River basin (71,225 km²), which is the largest sub-basin showed −4% decrease in annual flow under doubling CO₂ (Neff et al. 2000) to 24% increase for the decade of 2090–2099 with a 1% increase of CO₂ per year (Najjar 1999). Other hydrologic modeling studies showed a wide range of changes in the annual streamflow projections: −32 to 6% under the CO₂ doubling scenario (Moore et al. 1997), −25 to 33% with 1% increase in CO₂ per year between 2090 and 2099 (Wolock & McCabe 1999), and 9 to 18% increase for the period of 2070–2099 under the SRES A1FI and B1 scenarios (Hayhoe et al. 2007) all of which used the mid to late 20th century to obtain base periods for comparisons. The Phase 5.3 CB Watershed assessment included an initial estimation of flows, associated sediments, and nutrient loads using the Climate Assessment Tool (USEPA 2010). This approach followed simple adjustments to historical precipitation and air temperature records using standard arithmetic operators applied to both monthly and seasonal data (USEPA 2010).

As climate models and downscaling techniques are constantly refined, they are useful to estimate regional climate characteristics (Maurer et al. 2010). Implementing the latest climate data, i.e. the fifth phase of the Coupled Model Intercomparison Project (CMIP5) evaluated by the Fifth Intergovernmental Panel on Climate Change Assessment Report (AR5) (Taylor et al. 2012), and the CB Watershed Model (CBWM), which was extensively developed for last two decades (Shenk et al. 2012), would be a vigorous combination to better understand the watershed behavior in a changing climate. The objectives of this study are to (1) estimate changes in precipitation and temperature using the CMIP5 GCMs downscaled data, (2) simulate streamflows using the CBWM Phase 5.3, and (3) investigate the water budget components to quantify the expected direction and magnitude of changes under climate change conditions.

Two Representative Concentration Pathway (RCP) scenarios in the Intergovernmental Panel on Climate Change AR5, RCP4.5, and RCP8.5 were implemented in this study. The RCP4.5 scenario implies that greenhouse gas emissions will be stabilized by the middle of the 21st century with a radiative forcing of 4.5 Wm⁻² in the year 2100 (Thomson et al. 2011). On the other hand, the RCP8.5 scenario assumes little effort to reduce greenhouse gas emissions with an anthropogenic radiative forcing of 8.5 Wm⁻² in 2100 (Riahi et al. 2011). In total, 80 climate projection scenarios (40 ensemble members in each RCP scenario) from a combination of the RCP4.5 and RCP8.5 were selected based on the availability of Daily Bias Correction Constructed Analogues (BCCA) CMIP5 climate data (Maurer et al. 2007; Hidalgo et al. 2008; Reclamation 2013) (Appendix B1, available with the online version of this paper). The statistically downscaled BCCA CMIP5 climate data included daily precipitation and daily maximum/minimum temperature data at 1/8th degree spatial resolution over the contiguous United States for the period 1950–2099 as well as the gridded observation data derived from the weather stations for the period of 1950–1999.

The CBWM built from the Hydrological Simulation Program-Fortran (HSPF) is widely used to simulate streamflows and pollutant loads that are transported from the uplands to the CB (Shenk et al. 2012). For this study, the Phase 5.3 CBWM (USEPA 2010) was implemented. The land segment, the basic representative area for the runoff simulation, generally corresponds with the county boundary that contains data on human and livestock population, pollutant sources, etc. The CBWM inputs include multiple years of land segment data including land use, manure, fertilizers, streamflow withdrawal, point sources, and atmospheric deposition of nutrients for multiple years during the 1984–2005 simulation period (USEPA 2010). Land use consists of 26 categories, with 16 of them categorized as agricultural lands, two as forests, four as urban areas, and four as other land uses. Streamflow withdrawal inputs include irrigation, industrial, and municipal uses. For many large industrial uses, such as thermoelectric generation, only a small amount of water consumption or loss, i.e. evaporation, is assumed and most of it is directly returned to the streams through permitted discharge from each facility. Withdrawal for irrigation and municipal purposes are considered as 100% consumptive use in the CBWM. However, collected and treated water from municipal uses are considered as point source flow within the model. Streamflow withdrawal data are provided on a monthly basis over 1984–2005 (USEPA 2010).

As part of the simulation effort, first a total of 238 land segments within the watershed were simulated and the results were subsequently transferred to 1,174 river segments for river routing (Preston & Brakebill 1999). Since the CBWM was calibrated and validated for the period 1985–2005 using an automated method developed for the CBWM (USEPA 2010), the refined model parameter values were simply adopted for our implementation.

The climate data were processed to generate weather inputs for the CBWM. Projected climate data covered the CB Watershed with 1,244 grid cells at 1/8th degree resolution; however, our study did not simulate the hydrology at this resolution. The climate grid cells that fell under each land segment were averaged since each land segment simulation in the CBWM needed only a single weather dataset. If a land segment was too small to include a grid cell, the closest grid cell to the land segment was assigned. The integration process combined 1,244 climate grids into 238 weather data points for the land segments of the CBWM. For the preparation of weather input files for the CBWM, precipitation and temperature data were disaggregated from daily to hourly time steps using the WDMUtil (Hummel et al. 2001). Then, hourly potential evapotranspiration (PET) and evaporation from open water bodies were calculated based on the Hamon (Hamon 1961) and Penman (Penman 1948) methods, respectively. Our analysis included the weather data for all 40 ensemble members under the two chosen RCP scenarios.

Assessing the historic simulations of climate models is the first necessary step to examine the trends and variability in precipitation and temperature as these models exhibit appreciable differences. It is important to note that the differences between the GCM simulations and observation were not unrealistic since the GCMs are known to contain some uncertainties in characterizing natural climate variability (Hayhoe et al. 2007). In addition, since the BCCA downscaling method corrected the GCM outputs based on the probabilistic distributions for the historic period 1950–1999 (Maurer et al. 2010; Reclamation 2013), the downscaled climate data might not exactly represent the observed climate events.

Historic temperature showed an upward trend of 0.023 °C/decade in the observed annual mean values over the period of 1950–1999 in the CB Watershed. The average temperatures of the CMIP5 models (±standard deviation) were also showing increasing temperatures of 0.139 ± 0.097 °C/decade for the same region and period (Table 2). For the period 1970–1999, the temperature increase from the observed data was about 0.150 °C/decade while the increase of the CMIP5 models (±standard deviation) was on the order of 0.251 ± 0.158 °C/decade. The ranges in CMIP5 trends were from −0.033 to 0.364 °C/decade and from −0.081 to 0.563 °C/decade for the periods of 1950–1999 and 1970–1999, respectively, although most of the ensemble members showed increasing trends.

Temperature increases were higher in the northern part of the watershed (0.013 °C/decade and 0.145 ± 0.100 °C/decade for the observed and ensemble averages, respectively) than in the southern region (0.010 °C/decade and 0.131 ± 0.095 °C/decade for the observed and ensemble averages, respectively). This implied that the northern part of the watershed experienced more temperature changes than the southern part and this could offer insights into differing hydrologic impacts between these two regions. But a weak cooling in the southern portion (Allard & Keim 2007) was not identified in this study except for few ensemble members.

The ensemble average of temperature changes in the future was depicted seasonally and spatially (Appendix A1, A2, and B2, available with the online version of this paper). The historic annual average temperature in the CB watershed ranged from 8.5 to 14.1 °C with the lowest and highest annual averages in the Susquehanna River and York River basins, respectively. Rising temperature conditions were shown from north to south and from inland to shores. Under RCP4.5, the annual average temperature of the ensemble was projected to increase by about 1.5–1.8 °C, 2.3–2.7 °C, and 2.8–3.2 °C for the periods of 2020–2039, 2040–2069, and 2070–2099 relative to the baseline period (1970–1999), respectively. The difference in the annual averages between the northern (e.g., Susquehanna River basin) and the southern (e.g., James River basin) parts were about 0.3–0.5 °C. Seasonal temperature projections generally coincided with annual changes for the future time period and were expected to have lesser spatial variance in spring and fall as opposed to summer and winter. The winter projections in temperature are in the range of 1.3–1.8, 2.0–2.7, and 2.4–3.2 °C for the three future periods, respectively.

Temporal and spatial temperature patterns were analyzed from the RCP8.5 scenario that showed increased temperature changes when compared with the RCP4.5 scenarios, e.g., 1.8–5.2 °C increase under RCP8.5 but 1.7–3.0 °C under RCP4.5 for the whole future period (2020–2099) across the CB Watershed. Approximately 60–65 days per year in the 2090s were also identified which were above the 95th percentile temperature value of the 1990s under the RCP4.5 scenarios, while 20–40 days/year (Hayhoe et al. 2007) and 30–40 days/year (Diffenbaugh et al. 2005) for the same period were estimated in the Northeastern US region.

Both spatial and temporal variability complicated the assessment of trends in precipitation over a region, unlike temperature. The observed precipitation during the period 1950–1999 was increasing by 18 mm/decade with a p-value of 0.24, but the observed precipitation between 1970 and 1999 was calculated as decreasing by −9.6 mm/decade (p = 0.74) (Table 2). Floods in 1972 and 1975 brought the trends down for the period of 1970–1999. The positive tendency was evident when the two flood years were excluded from the analysis. The ensemble average precipitation showed increasing tendency (p = 0.48 and 0.49) with the mean ± 1σ of 6.2 ± 10.4 and 11.4 ± 24.4 mm/decade for the two historic periods, respectively.

On the other hand, the mean annual precipitation values of the ensemble members matched well with observed precipitation records with at least 99.8% agreement for the period 1950–1999. Similarly, good agreements for the monthly averages between the observed and simulated precipitation for 1950–1999 were noted. However, for the period 1970–1999, the differences in mean annual precipitation between observations and projections ranged from −0.2 to −5.0%, suggesting that the GCMs slightly underestimated over this region.

The GCMs projected wetter conditions, i.e. more precipitation in the future when compared with the baseline period (1970–1999) for both RCP4.5 and RCP8.5 scenarios (Figure 2). About 11 and 13 out of 40 ensemble members projected less precipitation in the near future (2020–2039) under the RCP4.5 and RCP8.5 scenarios, respectively, whereas only two ensemble members suggested drier conditions at the end of the century (2070–2099). The GCMs showed −5 to 20% of changes in mean annual precipitation, and the changes in each future period were projected to increase over time. The RCP8.5 scenario resulted in a wider range of precipitation change than the RCP4.5, presenting more variability among the GCMs under the RCP8.5 scenarios.

The historical precipitation was 1,079 mm for the baseline period and showed a great spatial variability ranging from 1,037 to 1,143 mm (Appendix A3, available with the online version of this paper) with more precipitation over the regions close to the estuary. The ensemble average precipitation was projected to increase by about 1.6%, 3.9%, and 5.1% for the future periods 2020–2039, 2040–2069, and 2070–2099, respectively under the RCP4.5. However, the increasing trends ranged were about 2.3%, 5.1%, and 8.4% under the RCP8.5 scenarios (Appendix A3, A4, and B2, available online). Least decreasing trends in precipitation were projected in the fall season with ranges from −3.1 to −0.6% and −3.6 to −0.7% under RCP 4.5 and RCP 8.5 scenarios, respectively, but for the other seasons, the trends appeared increasing. Higher changes were projected in winter ranging between 5.1 and 10.2% for the RCP4.5 and 5.6 and 15.5% for the RCP8.5 scenarios. Wetter conditions in the northern region of the CB Watershed were projected in the spring and winter while the southern region was to have more precipitation during the summer and fall in the future. This pattern was reported by other studies that predicted a consistent increase in winter precipitation and no change or decreases in summer rainfall in the Northeastern USA, which included the northern part of the watershed (Hayhoe et al. 2007; Solomon et al. 2007; Kunkel et al. 2013).

The simulated streamflow, driven by each downscaled GCM output for the historic period, was compared with observed flows from 222 gaging stations within the CB watershed. The simulated flow could capture the seasonal characteristics of observed flows including the magnitude and timing of peak/low flow. It should be noted that since the simulated climate data did not fully reflect the specific timing of events, the resulting streamflow projections also did not follow the observed flow exactly; however, the flow characteristics generally provided the expected changes in hydrology.

Three non-parametric statistical tests were conducted to evaluate if the derived flow replicated the observations in terms of annual flow volumes: the Mann–Whitney U test which was for testing central tendency (Wilks 2006), i.e. mean flow; the Siegel–Tukey test which tested variance between two groups (McCuen 2003), i.e. variability of flow; and the Kolmogorov–Smirnov (KS) test which assessed whether two samples were drawn from the same distribution, including central tendency, variance, and overall shape of distribution (Wilks 2006). All the tests were carried out at a 95% confidence interval (p = 0.05). Comparison of central tendencies, variability, and distribution between the observed and the simulated annual flows was tested for the 40 ensemble members in each gaging station (Appendix B3, available with the online version of this paper). Simulated mean annual flows for 76–99% of gaging stations did not exhibit significant difference with observed flows in the watershed. The statistical comparison of variability between the simulated and observed annual flows showed similarity in 66.6–91.8% of the gages, with most differences in the Susquehanna River basin. This suggested that the simulated flows obtained from GCM-derived climate data for the historic period (1984–2005) shared a similarity with observed annual flows in terms of mean and variability. The CanESM2.3 and CanESM2.1 showed the best performance in both the Mann–Whitney U test and the KS test while the CanESM2.2 was high ranked in the KS test and the Siegel–Tukey test, presenting a slight discrepancy between the best-performing GCMs for resembling central tendency and variability of the observed annual flow. The statistical analysis indicated that simulated flow forced by the climate model output was limited to reproduce the characteristics of the observed flow well with both the mean annual and interannual variability.

Monthly analysis of projected flows for the historic period showed that they were generally capturing the observed magnitudes in terms of low and high flows (Appendix A5a–A5f), although the ensemble could not capture several high flow events at SL and PM. For example, a flood month in 1993 at SL and two months in 1993–1994 at PM exceeded the highest value, and the high flow during spring and summer in 1996 at PM was not reflected in the simulated flows. On the other hand, observed low flows in the drought years of 1999–2000 at SL, PM, RU, and YP, and 1999–2003 at JL were generally captured. Seasonal comparison between the simulated and measured flow for the historical period indicated that the flow regime driven by the GCMs could capture the measured flow regimes (Appendix A5g–A5l). In SL, PM, YP, and JA, the average monthly flow in February to August was underestimated, while the monthly flow at RU and JL from October to February was overestimated. Simulated average flows in SL showed peak flow in March while the observed flow was recorded in April as a high flow month. (Appendix A5 is available with the online version of this paper.)

Historic simulations showed biases in streamflows which could potentially undermine the analysis of future projections. Despite certain correction techniques applied to the climate model data, biases in simulated flow could come from uncertainties in the input data, hydrologic model structure, and observed data itself. In this study, ratios of GCM-driven flows to the observations for the historic period were calculated using the average monthly flow values (January to December), and then were multiplied to projected flows to adjust the future projections.

It has been reported that the projected streamflow results in the Mid-Atlantic varied between −40 and 30% using two to four GCMS and hydrologic models (Najjar et al. 2009, 2010). For this study, 19 GCMs (40 ensemble members) were employed to understand the characteristics of climate variability and uncertainty and to investigate the hydrologic response over a wide spectrum. Because the projected flow driven by the GCMs varied dramatically, the flow projections were divided into two groups to look into the characteristics of increasing and decreasing flows: one that was above the observed flow for baseline period (1985–2005) and the other below the observational records.

The annual streamflow projections varied among the ensemble members, which ranged from −30.3 to 15.2% and −41.1 to 8.7% for the period 2070–2099 with the RCP4.5 and RCP8.5 scenarios, respectively. The numbers of ensemble members that showed increasing annual flows were eight and three under the RCP4.5 and RCP8.5 for the period 2070–2099 in the CB Watershed (Table 3 and Appendix B5, available with the online version of this paper). It resulted in mean ± σ of 6.3 ± 5.4% and 4.7 ± 3.3% respectively, which were lower than reported in Hayhoe et al. (2007). On the contrary, the projected scenarios resulted in decreased streamflows with mean ± σ of −12.4 ± 7.5% and −21.8 ± 9.2% by 2070–2099 relative to the baseline period under the RCP4.5 and RCP8.5 scenarios, respectively. Hayhoe et al. (2007) projected increases in runoff ranging from 9% to 18% by 2070–2099 relative to a historic period in the PA and New Jersey (NY) region, which roughly covers the Susquehanna River basin, under the A1FI and B1 scenarios. These differences in streamflow projections can be explained by the variability of the projected precipitation magnitudes among the climate models, which suggested 7% to 14% increase by 2070–2099 relative to 1961–1990 in the PA-NY region (Hayhoe et al. 2007) and 5.2% to 8.3% increase in this study. The projected warming shown in these studies were similar to increases over the baseline periods of 2.9–5.3 °C (Hayhoe et al. 2007) and 3.2–5.5 °C (this study) by 2070–2099. However, the methods used in estimating ET, Hamon (this study) versus Penman (Hayhoe et al. 2007) methods, could have caused differences in the future flow projections. The streamflow could decrease up to 15–40% due to warming up to 4 °C in the Mid-Atlantic region (Najjar et al. 2009).

Although the annual flow projections showed decreasing trends, seasonal flow changes varied among the major basins in the CB Watershed. Fourteen and six ensemble members under the two RCP scenarios projected winter flows to increase from 8.1 ± 5.4% to 8.1 ± 5.9% for 2070–2099 relative to the baseline period in the CB Watershed, which was the greatest number of ensembles expecting increasing flow compared to any other seasons. Decreases in winter flows showed the fewest changes relative to the other seasons (Table 3 and Appendix B5). On the other hand, the greatest number of ensembles predicted flow reduction in fall with the greatest variability (i.e. standard deviation). Those projections corresponded with precipitation projections thus presenting a wetter winter and drier fall.

Two hydrologic indices were calculated to assess changes in projected high and low flows: a 7-day duration low flow with a 10-year return period (7LQ10) and a 3-day duration high flow with a 10-year return period (3HQ10). Those were computed based on the minimum and maximum moving averages of daily streamflow with a 10-year return period assuming the log-Pearson-III type distribution. While the projected low and high flows showed a great variability among GCMs, the average of the flow ensemble decreased both in the 7LQ10 and 3HQ10 values over the future periods with the exception of the high flow in the York River basin. Decreases in low flows over the entire watershed were estimated to be about −28.9 ± 35.4% and −58.4 ± 20.9% in 2070–2099 under the RCP4.5 and RCP8.5 scenarios, respectively, while reductions in high flows were projected to be between −8.0 ± 22.2% and −19.8 ± 23.0% (Appendix B6, available with the online version of this paper). This indicated that the projected climate impacted the low flows more than the high flows. Projected decreases in low flow could have important implications for water resource use and management. A weak decrease in the high flow was detected because variation (standard deviation) between the 3HQ10s exceeded the average of flow ensemble in most major basins. Only the Susquehanna River basin showed decreases in 3HQ10, and this could be explained by the highest temperature increase in the northern region. On the contrary, the York River basin had increased high flows under the RCP4.5 scenario. The highest increase in summer precipitation (12.1 ± 7.8% to 15.6 ± 10.8% under RCP4.5) possibly resulted in increased high flows.

Evaluation of the water budget components due to the anticipated changes in precipitation and temperature is expected to provide insights into the management of water resources in the watershed. Projected increases in temperature accelerated ET while increasing precipitation trends promoted runoff. Estimated hydrologic variables (ET and runoff consisting of surface flow, interflow, and groundwater flow) were used to assess the long-term implications of the simulations results.

As shown in Table 5, for the baseline period (1985–2005) the annual total precipitation of 1,063 mm resulted in 660 mm (62%) and 400 mm (38%) of ET and runoff, respectively. The runoff ratio varied among the basins ranging from 31 to 44% with the highest value in the Susquehanna River basin.

The impacts of climate change across the CB watershed was investigated using the climate model projections of temperature and precipitation. The hydrologic simulation model, Phase 5.3 CBWM, developed for the CB watershed management was implemented to evaluate streamflow, ET, and runoff.

Increases in temperature from 3.0 to 5.2 °C were projected between 2070 and 2099 under the RCP4.5 and RCP8.5 scenarios. Increases in precipitation were projected between 5 and 8% by 2070–2099 relative to the baseline under the RCP scenarios. However, one-third of climate models predicted less precipitation in the near future (2020–2039), whereas only two GCMs expected drier conditions towards the latter part of the century (2070–2099). Warmer and wetter winters over the CB watershed were projected, and the northern part of the CB watershed was wetter and warmer than the southern region.

The simulated annual flow with the climate model inputs from two climate models, CanESM2.3 and NorESM1-M, appeared realistic. Streamflow projections, assuming fixed land use and no water withdrawals, varied among the ensemble members over time, although most ensemble members projected decreasing streamflow at the end of the 21st century relative to the baseline period. The average changes in flow were projected as −12.4 to 6.3% and −21.8 to 4.7% for 2070–2099 under the RCP4.5 and RCP8.5 scenarios, respectively. Projected high flows (3HQ10) as well as low flows (7LQ10) also decreased. A transition in peak flow time was expected at SL from April to March, which corresponded with warmer and wetter winter projections in the northern region.

Total simulated ET increased by about 17–28% in 2070–2099 relative to the baseline period under the two RCP scenarios. Although the wetter conditions were expected in the future by the end of the century, the land surface runoff was mainly projected to decrease due to increased ET therefore evaluation of ET using different approaches requires attention in partitioning the water budget. Changing streamflows and the cascading impacts on water quality can be a concern because it closely relates to the management of the CB watershed and its stream network. However, this research is not directly linked to water quality. Studies linking nitrogen and phosphorous loading into the CB attributed to both non-point source and point source pollution impacting the water quality. Any future water quality management research should therefore emphasize a thorough hydrologic analysis with appropriate physically based hydrologic models.

We acknowledge the World Climate Research Programme's Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Appendix B1 of this paper) for producing and making available their model output. For CMIP, the US Department of Energy's Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This project was funded, in part, by the Virginia Agricultural Experiment Station (Blacksburg) and the Hatch Program of the National Institute of Food and Agriculture, US Department of Agriculture (Washington, DC).