4.1. Model Performance
 Overall, the model performed well and adequately simulated the observed values. The model satisfactorily captured seasonal variation in streamflow patterns, with slight overprediction during summer and fall, and slight underprediction during spring and winter. These over- and underpredictions are manifested in a slight underprediction of annual discharge. Therefore, the model captured the general annual hydrologic pattern over the period of 1964–2008 without any tendency in over- or underprediction.
 Although there was a slight overprediction of stream , the model captured the long-term trend of decreasing concentration. This long-term decline in stream is due to emission controls of SO2 associated with the 1970 and 1990 Amendments to the Clean Air Act [Driscoll et al., 2001]. In general, the model overpredicted concentrations. Because of the low selectivity coefficient for soil- exchange (Log K = –0.107) [Gbondo-Tugbawa et al., 2001], the exchangeable pool of is very small. The higher simulated concentration of and subsequent increase in soil pools triggered higher rates of nitrification and soil N mineralization, which contributed to the overprediction of in stream water. There has been an unexplained decline in measured stream concentrations at the HBEF and surrounding region despite a high chronic atmospheric deposition of N and the increasing age of the forest [Goodale et al., 2003, 2005] which is consistent with an overprediction of simulated stream concentrations. Modeling the N cycle in forest ecosystems is a challenge due to complexity, confounding factors, and limitations in knowledge about the N cycle in forest ecosystems, hampering the development of algorithms in the model that enable adequate depiction of stream water N losses. PnET-BGC incorporates current thinking of the nitrogen cycle of forest ecosystems to the extent that we understand it, but until a mechanism for the decrease in N loss can be identified and quantified it seems dishonest to modify an input or parameter of the model or invoke a poorly understood process to fit the measured data. Nevertheless, PnET-BGC is effective in simulating the response of the N to vegetation disturbance [Aber et al., 2002] and so likely captures the plant-soil perturbation associated with changing climate.
 The model calculates pH from a charge balance of all ions in stream water and mass law expressions of dissolved inorganic carbon, Al, and natural occurring organic acids [Gbondo-Tugbawa et al., 2001]. Accurate modeling of pH is a key component in most watershed models which simulate acid-base chemistry because many biological processes and effects are closely linked with pH [Gbondo-Tugbawa et al., 2001]. Simulation of pH is especially challenging in systems with ANC values close to 0 eq L−1, like the HBEF (pH 4.7–5.7) [Davis et al., 1987]. Since pH values are affected by all biogeochemical processes which influence the concentrations of ionic solutes, slight errors in the simulation of major elements can result in high variation and possible errors in pH predictions. Based on model performance criteria for pH, slight overprediction of and are compensated for, to some extent, by slight an overprediction of base cations. The underprediction of ANC values are mainly due to overprediction of and naturally occurring organic acids (i.e., DOC).
4.3. Modeling Results for Hydrology, Soil, and Stream Water Chemistry
 Under PnET-BGC model runs without CO2effects, warmer temperatures in the future caused a decrease in soil moisture and an increase in vapor pressure deficit, despite the increase in precipitation. These factors decrease evapotranspiration and cause midsummer drought stress, the extent of which is dependent on the climate-change scenario considered. Although wood net primary production (NPP) increased due to warmer temperatures and a longer growing season, repeated midsummer drought is projected to decrease maximum leaf area index, foliar NPP, and litterfall and fine root NPP [Aber and Federer, 1992; Campbell et al., 2009, 2011] (Table 5). Overall, these changes translate into less C sequestration in foliage and fine roots, and more in wood. Because of slower decomposition rates associated with woody litter, the model simulates a decrease in C transfer to humus. The increase in wood NPP does not offset the decline in the litter inputs (sum of leaf litterfall and fine roots) to the soil organic matter (SOM) pool.
 The assimilation of N, Ca, and other nutrients in plant tissues was similar to the pattern for C. The result of the shift in NPP was a decrease in litterfall elements, causing declines in the humus pool (Table 5). Because of water stress, the plant demand for N decreased and the available N pool for plants increased, resulting in a 6.6% decrease in the C:N ratio of the humus pool (Table 5). Although both model simulations and observed values show that the HBEF is currently a sink for atmospheric N deposition, future model simulations suggest that climate change may cause the site to shift to N source to downstream aquatic ecosystems. Note that previous experiments and measurements at the HBEF have demonstrated that the N cycle is very sensitive to ecosystem disturbance that affects forest vegetation [Likens et al., 1970; Houlton et al., 2003].
 The elevated export of from forest soils to surface waters is an environmental concern in the northeastern U. S. and elsewhere [Aber et al., 2003; Driscoll et al., 2003]. Elevated leaching losses of facilitate the depletion of cations from soil, and contribute to soil and surface water acidification [Driscoll et al., 2003]. High can lead to water quality impairments and can contribute to the eutrophication of coastal waters. It is challenging to model N losses from forest ecosystems, due to a poor understanding of processes that control N cycling, particularly those associated with immobilization and denitrification [e.g., Dail et al., 2001; Venterea et al., 2004]. Nitrogen retention is sensitive to a variety of factors, including the legacy effects of historical land use and disturbance which are often poorly characterized [Aber et al., 2002]. Despite these uncertainties, PnET-BGC is a useful tool for assessing the effects of climate change on the N cycle since it accounts for other disturbances including climate change, N deposition, and atmospheric CO2 simultaneously [Ollinger et al., 2009].
 Studies suggest that surface water DOC is increasing in Europe and the northeastern U. S. The alternative mechanisms explaining this phenomenon are declines in acidic deposition or climate change [Clark et al., 2010; Evans et al., 2006; Findlay, 2005; Freeman et al., 2001, 2004; Garnett et al., 2000; Monteith et al., 2007; Worrall et al., 2003]. PnET-BGC simulations suggest that DOC will decrease over the twenty-first century under all climate-change scenarios. This modeled decline in DOC is associated with a decline in litterfall and a decrease in soil C mineralization rates (Table 5). The trends in stream water DOC were modified under climate change in the presence of CO2 fertilization. The higher productivity of the forest (NPP and net ecosystem production) due to CO2 fertilization increased litterfall in comparison to values from model simulations without CO2 effects on vegetation (Table 5). An increase in the decomposition of the organic matter pool, triggered by higher temperatures, led to higher DOC concentrations in stream water. Note that when CO2 effects on vegetation were included in the simulations, large increases in stream DOC were not evident. Our model simulations would seem to be inconsistent with the hypothesis that climate change is driving increases in surface water DOC.
 While hydrochemical models such as PnET-BGC provide useful information about how ecosystems may respond to global change, they are somewhat limited by sources of uncertainty. First, there are only a few studies that have evaluated the effects of CO2 fertilization on NPP, especially in northern hardwood forest ecosystems [Ainsworth and Long, 2005; Curtis and Wang, 1998; Curtis et al., 1995; Ellsworth, 1999; Ellsworth et al., 1995; Lewis et al., 1996; Saxe et al., 1998]. Experimental manipulations show that increased atmospheric CO2enhances plant productivity, but the extent of this response over the long term in conjunction with other global-change drivers is not well established. Second, it is unclear how atmospheric deposition will change in the future, which could substantially influence the element responses. Moreover, we did not consider scenarios of future land disturbance, which could further affect hydrologic and biogeochemical dynamics. Third, changes in climate and other factors (e.g., pests, pathogens) may alter the composition of vegetation at the HBEF, which could also influence hydrologic (e.g., transpiration) and biogeochemical (e.g., uptake, litterfall, decomposition) fluxes. While changes in established tree species would occur slowly in response to climate change, the effects might be more pronounced at locations such as the HBEF, which are located in a transition forest zone (between northern hardwoods and red spruce-balsam fir forests). In this application, PnET-BGC model simulations assumed that the watershed consisted of a homogeneous distribution of northern hardwood forest. In the future it would be useful to evaluate the influence of shifts in species composition or to link PnET-BGC with a forest community model that projects changes species assemblages. The temperature conditions considered in some of the climate scenarios are beyond the conditions under which parameter values were developed for PnET-BGC. We are currently evaluating model performance for watersheds of lower latitude to assess this limitation. Finally, it is important to reduce the uncertainty of climate-change projections, particularly for precipitation, by continuing to improve climate models, downscaling techniques (e.g., station-based instead of gridded), and linkages with hydrochemical models.