Predicting Acidification Recovery at the Hubbard Brook Experimental

Oct 28, 2010 - Predicting Acidification Recovery at the Hubbard Brook Experimental Forest, New Hampshire: Evaluation of Four Models. Koji Tominaga* ...
0 downloads 0 Views 2MB Size
Environ. Sci. Technol. 2010, 44, 9003–9009

Predicting Acidification Recovery at the Hubbard Brook Experimental Forest, New Hampshire: Evaluation of Four Models K O J I T O M I N A G A , * ,†,# J U L I A N A H E R N E , † SHAUN A. WATMOUGH,† MATTIAS ALVETEG,‡ BERNARD J. COSBY,§ CHARLES T. DRISCOLL,| MAXIMILIAN POSCH,⊥ AND AFSHIN POURMOKHTARIAN| Environmental and Life Sciences, Trent University, 1600 West Bank Drive, Peterborough, Ontario, K9J 7B8, Canada, Department of Chemical Engineering, Lund University, P.O. Box 124, Lund 221 00, Sweden, Environmental Sciences Department, University of Virginia, 291 McCormick Road, Charlottesville, Virginia, 22904-4123, United States, Department of Civil and Environmental Engineering, Syracuse University, Syracuse, New York, 13244-1190, United States, and Coordination Centre for Effects (CCE), National Institute for Public Health and the Environment (RIVM), P.O. Box 1, 3720 BA Bilthoven, The Netherlands

Received July 3, 2010. Revised manuscript received October 5, 2010. Accepted October 18, 2010.

The performance and prediction uncertainty (owing to parameter and structural uncertainties) of four dynamic watershed acidification models (MAGIC, PnET-BGC, SAFE, and VSD) were assessed by systematically applying them to data from the Hubbard Brook Experimental Forest (HBEF), New Hampshire, where long-term records of precipitation and stream chemistry were available. In order to facilitate systematic evaluation, Monte Carlo simulation was used to randomly generate common model input data sets (n ) 10 000) from parameter distributions; input data were subsequently translated among models to retain consistency. The model simulations were objectively calibrated against observed data (streamwater: 1963-2004, soil: 1983). The ensemble of calibrated models was used to assess future response of soil and stream chemistry to reduced sulfur deposition at the HBEF. Although both hindcast (1850-1962) and forecast (2005-2100) predictions were qualitatively similar across the four models, the temporal pattern of key indicators of acidification recovery (stream acid neutralizing capacity and soil base saturation) differed substantially. The range in predictions resulted from differences in model structure and their associated posterior parameter

* Corresponding author phone: +47 22854505; e-mail: [email protected]. † Environmental and Life Sciences, Trent University. ‡ Department of Chemical Engineering, Lund University. § Environmental Sciences Department, University of Virginia. | Department of Civil and Environmental Engineering, Syracuse University. ⊥ National Institute for Public Health and the Environment (RIVM). # Present address: Department of Biology, University of Oslo, P.O. Box 1066 Blindern, 0316 Oslo, Norway. 10.1021/es102243j

 2010 American Chemical Society

Published on Web 10/28/2010

distributions. These differences can be accommodated by employing multiple models (ensemble analysis) but have implications for individual model applications.

Introduction Dynamic watershed acidification models, such as MAGIC (model of acidification of groundwater in catchments) (1), PnET-BGC (photosynthesis and evapotranspiration-biogeochemistry) (2), SAFE (soil acidification in forest ecosystems) (3), and the VSD (very simple dynamic) model (4) are used to predict the rate of chemical response of soil and surface water to changes in acidic deposition (5, 6). Given the use of dynamic acidification models to support emission reduction policies, an understanding of uncertainty in model predictions (risk of inaccurate predictions) is important. One important source of uncertainty is model structure, which arises due to the formulation and simplification of the complex nature of ecosystems. Many dynamic acidification models serve the same function (i.e., predict changes in soil and surface water chemistry), but are nonetheless based on different structural representations. The influence of model structure on simulation can be examined by systematically providing comparative models with the same input information (7, 8). Moreover, simulations from multiple models can be combined into an ensemble of models, accommodating a range of alternative process formulations in prediction (9-13). Dynamic acidification models are typically calibrated by adjusting imprecisely known parameters so that model output (i.e., soil and surface water chemistry) correspond with observed data (14). Once calibrated, the models can be used for prediction under a prescribed deposition scenario. However, the hyperdimensional nature of model parameter space often allows many parameter sets to reproduce the same observed data, referred to as equifinality (15). This results in a range of simulations that are conditioned upon observations, but vary beyond the observation period. Bayesian techniques, such as Markov Chain Monte Carlo (16, 17) and the generalized likelihood uncertainty estimation (15), are increasingly used to accommodate this phenomenon. Alternatively, a combination of more simple Monte Carlo techniques (8, 18) and observation reproducibility criteria (19) may be used. Generally, longer and more detailed observed data provide a more challenging and rigorous test of model simulations. For this reason, long-term monitoring sites, such as the Hubbard Brook Experimental Forest (HBEF) (20), are valuable “instruments” to evaluate models. The objective of this study was to evaluate the prediction uncertainty and performance of four widely used dynamic watershed acidification models (MAGIC, PnET-BGC, SAFE, and VSD). To this end, all models were systematically applied to a small catchment at the HBEF and calibrated to observed soil and stream chemistry data. Variability in predictions was examined between and within models, in light of parameter and structural differences. The future response of soil and stream chemistry to decreases in sulfur deposition at the HBEF was assessed using the ensemble of models under two deposition scenarios: “current legislated emissions” as forecasted in 2008 (CLE) and “maximum technologically feasible reductions” (MFR).

Materials and Methods Study Site. The study was carried out using long-term data from Watershed 6 (W6), HBEF, New Hampshire (43°56′ N 71°45′ W). Since 1963, precipitation volume, bulk deposition VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9003

TABLE 1. List of Parameters Subjected to Monte Carlo Simulation, And Their Probability Distributions (N = Normal; U = Uniform) distribution parameters parameter 2-

type a

current deposition SO4 concentration current deposition Cl- concentrationa current deposition K+ concentrationa H+:SO42- charge ratio in deposition NH4+:SO42- charge ratio in deposition Ca2+:NH4+ charge ratio in deposition Mg2+:Ca2+ charge ratio in deposition NO3-:H+ charge ratio in deposition Na+:Cl- charge ratio in deposition dry deposition factor Ca2+, Mg2+, K+, SO42-b dry deposition factor Na+, Cl-b dry deposition factor NH4+, NO3-b organic soil depthc mineral soil depthc organic soil porosityc mineral soil porosityc organic soil-water contentc mineral soil-water contentc organic soil bulk densityc mineral soil bulk densityc organic soil cation exchange capacityc mineral soil cation exchange capacityc organic soil DOC concentrationc mineral soil DOC concentrationc DOC charge density for Oliver Organic soil solution pCO2c mineral soil solution pCO2c soil N immobilization fraction soil Ca2+ weathering rate soil Mg2+ weathering rate soil K+ weathering rate soil Na+ weathering rate initial Ca2+ exchangeable fraction initial Mg2+ exchangeable fraction initial K+ exchangeable fraction initial Na+ exchangeable fraction initial H+ exchangeable fraction organic soil gibbsite dissolution constantc mineral soil gibbsite dissolution constantc

N N N N N N N N N U U U U U U U U U U U U U N N N U U U U U U U U U U U U U U

mean (N) or minimum (U) 28.55 4.95 0.895 1.173 0.279 0.364 0.366 0.496 0.720 1.1 1.5 0.9 5.34 40 0.61 0.85 0.23 0.27 130 617 157.5 34 1305 279 13.27 9 19 0.75 5 1.5 10 15 0.15 0.05 0.07 0.025 0.225 6 8.1

standard deviation (N) or maximum (U) 0.99 0.47 0.123 0.014 0.007 0.010 0.011 0.009 0.013 1.3 1.6 1.1 7.42 60 0.71 0.95 0.44 0.37 158 755 192.5 77 72 15 1.087 11 21 0.99 25 10 25 30 0.25 0.075 0.12 0.05 0.375 7 9.1

unit -3

meq m meq m-3 meq m-3

cm cm

kg m-3 kg m-3 meq kg-1 meq kg-1 µmol L-1 µmol L-1 mmol g-1 C × atmospheric pCO2 × atmospheric pCO2 meq meq meq meq

m-2 m-2 m-2 m-2

yr-1 yr-1 yr-1 yr-1

log10(mol L-1)-2 log10(mol L-1)-2

a Mean of 2000-2004; historic and future deposition for each ion was scaled to current deposition. b Values were sampled individually for each ion. c Parameters for organic and mineral compartments were lumped into a single one-layer compartment except for PnET-BGC, which has an organic layer submodel.

chemistry, streamflow volume (runoff), and stream chemistry have been monitored at W6; the site and measurement schemes are well documented (20, 21). A variety of soil physicochemical measurements (e.g., soil depth, exchangeable cation fractions) from previous studies at HBEF were also used as input data (Table 1). A reconstructed hindcast deposition scenario (22), based on emission records for sulfur, nitrogen and PM10 (surrogate for base cation deposition), was employed for the period of 1850-1962 prior to bulk deposition collection (1963-2004). Overview. A principal consideration was to apply all models in an objective and systematic manner (Figure 1). Accordingly, common input parameter distributions (site physicochemical characteristics) were prepared for all models (Table 1). Input parameters were randomly sampled from these distributions (Monte Carlo simulation, n ) 10 000), and then systematically translated into model specific input formats (n ) 10 000 for each model) (8). This ensured that, for a given Monte Carlo iteration, all models were provided with consistent soil and deposition data. The models were run using these formatted inputs for the period of 1850-2100, producing uncalibrated simulation outputs (n ) 10 000 × 4 9004

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 23, 2010

models) that included soil (base saturation) and stream chemistry (pH, [Na+], [Bc] ) [Ca2+] + [Mg2+] + [K+], [SO42-], and acid neutralizing capacity [ANC] ) [Na+] + [Bc] - [SO42-] - [NO3-] - [Cl-]). The uncalibrated outputs were compared with observed soil and stream chemistry data; the first 30 simulations that reproduced observed data (“behavioral simulations”) for each model (n ) 30 × 4) were retained. These behavioral simulations were combined into an ensemble of multimodel simulations (n ) 120) to assess future changes in soil and surface water chemistry. Model Setup and Scope of the Simulations. All four models employed in the current study consider the following fluxes: deposition of major anions and cations, base cation weathering, cation exchange between soil solution and soil surface, and leaching of soil solution to surface water. Forest nutrient cycling (i.e., forest uptake and mineralization of organic matter) can have a significant impact on soil chemistry (23) but is modeled only by PnET-BGC. In the current study, forest nutrient fluxes simulated by PnET-BGC (24) were systematically passed to the other models as external inputs (Figure 1) to incorporate forest growth and

FIGURE 1. Modeling framework and data flow among the four dynamic acidification models.

TABLE 2. Soil and Surface Water Chemistry Criteria Used for Model Calibration variable

period

calibration criteria

a

soil base saturation (%) surface water pH surface water surface water surface water surface water a

1983 the simulated value is within 3.5% of the observed value 1963-2004 the mean of the simulation is within 1.85 times the standard deviation around the observation mean [Bc (Ca2+ + Mg2+ + K+)] (µeq L-1) 1963-2004 the mean of the simulation is within 0.5 times the standard deviation around the observation mean, and the correlation between the simulation and observation is greater than 0.74 + -1 [Na ] (µeq L ) 1963-2004 the mean of the simulation is within 3.5 times the standard deviation around the observation mean [SO42-] (µeq L-1) 1963-2004 the mean of the simulation is within 0.5 times the standard deviation around the observation mean, and the correlation between the simulation and observation is greater than 0.74 [ANC] (µeq L-1) 1963-2004 the mean of the simulation is within 1.0 times the standard deviation around the observation mean

Base saturation is the percentage of the cation exchange capacity occupied by base cations (Ca2+ + Mg2+ + K+).

disturbance history into all model simulations. Minimal model configurations (e.g., single lumped soil layer) and modification to model codes (e.g., weathering submodel was “turned-off” in SAFE, and a surface water compartment was added to VSD) were used in this study to facilitate standardization; model configurations and differences are presented in detail in the Supporting Information (SI). Model Calibration. Model calibration was conducted by filtering the Monte Carlo simulations according to arithmetic criteria based on observations (selected variables from volume-weighted stream chemistry (1963-2004) and soil base saturation (1983); Table 2) so that only behavioral simulations (i.e., simulations that do not violate the arithmetic criteria) were retained for further analyses. A parameter set that is behavioral for one model is not necessarily behavioral for another model given differences in model structure. The sample size of 30 calibrated simulations for each model ensured an adequate estimate of the median simulation. The behavioral criteria control the quality (fit) of the calibration; the stricter the criteria, the more constrained the calibration, resulting in fewer behavioral simulations. In this study, the criteria were defined to ensure at least 30 behavioral simulations for all models. The first 30 behavioral simulations for each model were retained, and the remaining, if any, were discarded to ensure an equal number of behavioral simulations for all models. Posterior Distributions. The behavioral simulations were used to examine the posterior parameter distributions for the calibrated models. Posterior distributions were obtained by tracing back input parameters for each behavioral simulation. Differences in the median of the posterior

parameter distributions among models were tested using Kruskal-Wallis one-way analysis of variance. Scenario Analysis by Ensemble Models. Future changes in soil and surface water at the HBEF were assessed using the ensemble of behavioral simulations under two future deposition scenarios: The CLE scenario was based on the Canadian Clear Air Regulatory Agenda and the U.S. Environmental Protection Agency’s Clean Air Interstate Rule, and represented a decrease of 13% for SO42- deposition by 2015 compared with 2000. The MFR scenario was based on full implementation of all presently available emission control technologies under forecasted anthropogenic activity (10, 25), and represented a reduction of 78% for SO42- deposition by 2030 compared with 2000 (Figure 2). The ecosystem response to the deposition scenarios was evaluated using stream [ANC] and soil base saturation, which are key indicators used in acidification studies. Surface water [ANC] is often used to indicate deleterious impacts to sensitive biota; simulations were compared with a widely used critical concentration level of 20 µeq L-1 (26). Soil base saturation indicates soil sustainability (27), and simulations were compared to a critical level of 10%.

Results and Discussion The success rate for behavioral simulations (percentage of simulations (n ) 10 000) that met all calibration criteria, Table 2) ranged from 0.3% (MAGIC) to 1.7% (SAFE). The success rate was limited by the initial parameter distributions, and therefore not indicative of each model’s inherent calibration efficiency. Calibration results (i.e., success rates for each VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9005

as VSD and SAFE do not model Na+ exchange (SI Table S1). The first major site disturbance was the 1919 tree harvest, which was predicted to have caused the acidification of soil solution and surface waters and increased leaching of base cations from the exchange complex. Acidification further developed as acidic deposition increased during the period 1950-1970; stream pH and [ANC] reached their lowest values during the 1970s.

FIGURE 2. Annual average historic (22) and future sulfate deposition under CLE (current legislated emissions; solid line) and MFR (maximum feasible reductions; dashed line) scenarios. criterion by each candidate model) are presented and discussed in the SI. All models predicted similar patterns of changes in stream and soil chemistry in the hindcast scenario (1850-2004; Figure 3); the bimodal pattern for simulated Na+ was expected

Although the pattern of biogeochemical response to acidification development was similar, there were substantial quantitative differences among model simulations. There was larger among-model uncertainty in base saturation (the 90th-percentile range: 25-38%) and stream pH (the 90thpercentile range: 6.2-6.8) during the 19th century (1850-1900) compared with the 20th century. Nonetheless, historic (1850-1900) base saturation (weighted-average of organic and mineral soils) was well above the critical level for forest soils (10%). There was less variability in simulated base saturation among models during the period 1920-2000; in contrast, stream pH variability decreased from 1.0 (1920-1940) to 10%) and stream [ANC] (>20 µeq L-1) were predicted to be met by the end of the 21st century, as at least 90% of all calibrated simulations satisfied these limits. In contrast, only 24% of the simulations (29 out of 120 behavioral simulations) met both critical chemical criteria under the CLE scenario. This difference derives solely from the difference in future sulfur deposition (Figure 2). The temporal recovery pattern is of particular importance from a policy perspective; in the current study, temporal recovery with a 90% probability, that is, when the 10thpercentile simulation meets the critical limit, was chosen as a recovery target. Under the CLE scenario, the 10th-percentile for [ANC] and soil base saturation increased slowly until 2050 and leveled off thereafter, remaining below the critical limit throughout the forecast period (2005-2100). In contrast, under the MFR scenario, the 10th-percentile values initially increased faster than under CLE, and continued to increase beyond 2050 for both variables. This trend led to a predicted eventual chemical recovery (meeting the critical criterion) by 2100 (soil base saturation by 2070, stream [ANC] by 2085). Significant (p < 0.05) differences in some of the calibrated input parameters were found between models, indicating that models have different behavioral posterior parameter distributions (Table 3). MAGIC and PnET-BGC’s posterior distributions for soil Bc weathering rate were lower than SAFE and VSD. Base cations in the stream are derived from deposition, weathering, and exchange from the soil complex; MAGIC and PnET-BGC predicted lower weathering to balance their higher exchange compared with SAFE and VSD. This pattern is consistent with the lower soil base saturation predicted by models with the Gaines-Thomas cation exchange submodel (MAGIC and PnET-BGC) compared with models with the Gapon cation exchange submodel (SAFE and VSD). Weathering rates at W6 have been previously estimated by mass balance (Na+: 23.5 meq m-2 yr-1), Na+/Ca2+ ratio (Ca2+: 17.2 meq m-2 yr-1), and modeling studies (using PnETBGC, Mg2+: 12.5 meq m-2 yr-1 and K+: 1.4 meq m-2 yr-1 (2)), resulting in an estimated base cation weathering (Ca2+ + Mg2+ + K+ + Na+) of 55 meq m-2 yr-1. The median base cation weathering rate for the behavioral simulations in this study ranged from approximately 60 meq m-2 yr-1 (MAGIC and PnET-BGC) to 70 meq m-2 yr-1 (SAFE and VSD, Table 3), slightly higher than the above estimate. Posterior distributions are influenced by model structure and provide insights into the conceptual ecosystem representation for each model. In this study, PnET-BGC had a lower exchangeable pool (compared with the other three models) and lower water content (compared with MAGIC and VSD, Table 3), implying that model performance improved with smaller elemental pools in soil and soil solution. In this study, the divergence in model simulations was caused by differences in model structure (e.g., ionic exchange between soil surface and soil solution) and behavioral input parameter ranges (i.e., posterior distributions). Uncertainties VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9007

TABLE 3. Median of Selected Posterior Parameter Distributions (n = 30 for Each Model)a parameters +

initial exchangeable Na fraction initial exchangeable base saturation (%)b Na+ weathering rate (meq m-2 yr-1) Bc (Ca2+ + Mg2+ + K+) weathering (meq m-2 yr-1) exchangeable pool (eq m-2) water content (cm) Gibbsite dissolution constant sulfate dry deposition factor

MAGIC

PnET-BGC

x

0.107 0.30xy 16.1x 43.3x 2.57xy 19.1x 9.25x 1.26x

SAFE

VSD xy

18.0xy 43.4x 1.59z 14.6y 8.97y 1.25x

0.097 0.28x 20.0y 50.9y 2.30x 16.6xy 9.16xy 1.23x

0.088y 0.31y 19.9y 51.3y 2.51y 17.8x 9.13x 1.22x

a Significant differences in median parameter values were tested using the Kruskal-Wallis test (p < 0.05). The significance letters (x, y, z) indicate the grouping of statistically non-significant medians in each parameter. Initial exchangeable fractions were supplied as input data to MAGIC, SAFE and VSD; they were calculated for PnET-BGC. b Base saturation is the percentage of the cation exchange capacity occupied by base cations (Ca2+ + Mg2+ + K+).

FIGURE 5. Predicted historic change (gray) and future recovery of (left) soil base saturation (%) and (right) stream ANC concentration (µeq L-1) using 10th- and 90th-percentile error range under the two scenarios (right of the dashed vertical line): CLE (current legislated emissions) and MFR (maximum feasible reductions). The red horizontal lines indicate the critical chemical limits for sensitive biota. surrounding the candidate models are not negligible, but can be accommodated by employing multiple models and parameter distributions (both initial and posterior distributions). Reducing uncertainties is paramount especially for sites with limited data (e.g., for regional applications). Under the CLE scenario, soil and stream acidification status at W6 would only experience limited improvement during the 21st century, and greater reduction in emissions (such as the MFR scenario in this study) will be necessary for a faster and greater recovery. The predicted limited recovery under the CLE scenario is specific to W6 (13.2 ha); sensitivity to acid deposition varies across the Hubbard Brook valley (∼4000 ha) (30), and on a regional scale. Successful calibration of the candidate models to a longterm data set supports wide usage of these models. Although there is variability among models and calibrations, the range in response is captured by the ensemble of models. This approach is useful for policy application as it provides an estimate of uncertainty associated with individual model predictions. Watershed acidification models remain an important research and management tool for assessing the potential impacts of future deposition scenarios.

Acknowledgments Funding for this study was provided in part by the Northeastern States Research Cooperative and a Collaborative Research and Development Grant from the Natural Sciences and Engineering Council (NSERC) and the Cumulative Environmental Management Association (CEMA). This research was also funded, in part, by the Canada Research Chairs Program and an NSERC Discovery Grant, and benefitted from additional support for K.T. by DAAD (German Academic Exchange Service) and The Research Council of Norway. Some data used in this publication were obtained 9008

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 23, 2010

by scientists of the Hubbard Brook Ecosystem Study; this publication has not been reviewed by those scientists. The HBEF is operated and maintained by the Northeastern Research Station, U.S. Department of Agriculture, Newtown Square, Pennsylvania and is a National Science Foundation (NSF) Long-Term Ecological Research (LTER) site. We are indebted to Gene E. Likens for the provision of precipitation and stream chemistry data with funding from the NSF (LTREB and LTER programs) and The Andrew W. Mellon Foundation. We also thank Scott Bailey and Tony Federer for providing soil samples for mineral species analysis and detailed hydrological properties for the Hubbard Brook Experimental Forest, and F. Dentener for the MFR scenario. This work was also made possible by the facilities of the Shared Hierarchical Academic Research Computing Network (SHARCNET), located in Ontario, Canada. We are grateful to Christian Huber for hosting the “Multi-Model Evaluation Workshop” at the Technische Universita¨t Mu ¨ nchen, Freising, Germany. In addition, the study benefited greatly from discussions with Christian Huber, Jana Kiekbusch, Thorjørn Larssen, Andreas Nicolai, Wendelin Weis, and Dick Wright.

Supporting Information Available Table S1 describing model features and configurations used in the current study, and Table S2 showing model calibration success rate against behavioral criteria. This material is available free of charge via the Internet at http://pubs.acs.org.

Literature Cited (1) Cosby, B. J.; Ferrier, R. C.; Jenkins, A.; Wright, R. F. Modelling the effects of acid deposition: refinements, adjustments and inclusion of nitrogen dynamics in the MAGIC model. Hydrol. Earth Syst. Sc. 2001, 5 (3), 499–517.

(2) Gbondo-Tugbawa, S. S.; Driscoll, C. T.; Aber, J. D.; Likens, G. E. Evaluation of an integrated biogeochemical model (PnET-BGC) at a northern hardwood forest ecosystem. Water Resour. Res. 2001, 37 (4), 1057–1070. (3) Alveteg, M. Dynamics of Forest Soil Chemistry, Ph.D. thesis, Department of Chemical Engineering II, Lund University, Lund, Sweden, 1998. (4) Posch, M.; Reinds, G. J. A very simple dynamic soil acidification model for scenario analyses and target load calculations. Environ. Model Software 2009, 24 (3), 329–340. (5) Hettelingh, J.-P.; Posch, M.; Slootweg, J.; Reinds, G. J.; Spranger, T.; Tarrason, L. Critical loads and dynamic modelling to assess European areas at risk of acidification and eutrophication. Water, Air, Soil Pollut.: Focus 2007, 7 (1), 379–384. (6) Wright, R. F.; Larssen, T.; Camarero, L.; Cosby, B. J.; Ferrier, R. C.; Helliwell, R. C.; Forsius, M.; Jenkins, A.; Kopacek, J.; Majer, V.; Moldan, F.; Posch, M.; Rogora, M.; Scho¨pp, W. Recovery of acidified European surface waters. Environ. Sci. Technol. 2005, 39 (3), 64A–72A. (7) Rose, K. A.; Cook, R. B.; Brenkert, A. L.; Gardner, R. H.; Hettelingh, J.-P. Systematic comparision of ILWAS, MAGIC, and ETD watershed acidification models. 1. Mapping among model inputs and deterministic results. Water Resour. Res. 1991, 27 (10), 2577–2589. (8) Tominaga, K.; Aherne, J.; Watmough, S. A.; Alveteg, M.; Cosby, B. J.; Driscoll, C. T.; Posch, M. Voyage without constellation: evaluating the performance of three uncalibrated processoriented models. Hydrol. Res. 2009, 40 (2-3), 261–272. (9) Ajami, N. K.; Duan, Q.; Sorooshian, S. An integrated hydrologic Bayesian multimodel combination framework: Confronting input, parameter, and model structural uncertainty in hydrologic prediction. Water Resour. Res. 2007, 43 (1), W01403. (10) Dentener, F. J.; Drevet, J.; Lamarque, J. F.; Bey, I.; Eickhout, B.; Fiore, A. M.; Hauglustaine, D. A.; Horowitz, L. W.; Krol, M.; Kulshrestha, U. C.; Lawrence, M.; Galy-Lacaux, C.; Rast, S.; Shindell, D.; Stevenson, D. S.; van Noije, T. P. C.; Atherton, C.; Bell, N.; Bergman, D.; Butler, T.; Cofala, J.; Collins, B.; Doherty, R.; Ellingsen, K.; Galloway, J. N.; Gauss, M.; Montanaro, V.; Mueller, J. F.; Pitari, G.; Rodriguez, J. M.; Sanderson, M.; Solmon, F.; Strahan, S. E.; Schultz, M.; Sudo, K.; Szopa, S.; Wild, O. Nitrogen and sulfur deposition on regional and global scales: A multimodel evaluation. Global Biogeochem. Cycles 2006, 20 (4), GB4003. (11) Devineni, N.; Sankarasubramanian, A.; Ghosh, S. Multimodel ensembles of streamflow forecasts: Role of predictor state in developing optimal combinations. Water Resour. Res. 2008, 44 (9), W09404. (12) Eyring, V.; Stevenson, D. S.; Lauer, A.; Dentener, F. J.; Butler, T.; Collins, W. J.; Ellingsen, K.; Gauss, M.; Hauglustaine, D. A.; Isaksen, I. S. A.; Lawrence, M. G.; Richter, A.; Rodriguez, J. M.; Sanderson, M.; Strahan, S. E.; Sudo, K.; Szopa, S.; van Noije, T. P. C.; Wild, O. Multi-model simulations of the impact of international shipping on atmospheric chemistry and climate in 2000 and 2030. Atmos. Chem. Phys. 2007, 7, 757–780. (13) Randall, D. A.; Wood, R. A.; Bony, S.; Colman, R.; Fichefet, T.; Fyfe, J.; Kattsov, V.; Pitman, A.; Shukla, J.; Srinivasan, J.; Stouffer, R. J.; Sumi, A.; K.E., T., Climate models and their evaluation. In Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Palnal on Climate Change.; Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tingnore, M., Miller, H. L., Eds.; Cambridge University Press: Cambridge, United Kingdom, 2007; pp 589-662.

(14) Janssen, P.; Heuberger, P. Calibration of process-oriented models. Ecol. Modell. 1995, 83 (1-2), 55–66. (15) Beven, K.; Freer, J. Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental systems using the GLUE methodology. J. Hydrol. 2001, 249 (14), 11–29. (16) Larssen, T.; Huseby, R. B.; Cosby, B. J.; Høst, G.; Høgåsen, T.; Aldrin, M. Forecasting acidification effects using a Bayesian calibration and uncertainty propagation approach. Environ. Sci. Technol. 2006, 40, 7841–7847. (17) MacDougall, G.; Aherne, J.; Watmough, S. A. Impacts of acid deposition at Plastic Lake: forecasting chemical recovery using a Bayesian calibration and uncertainty propagation approach. Hydrol. Res. 2009, 40 (2-3), 249. (18) Rose, K. A.; Brenkert, A. L.; Cook, R. B.; Gardner, R. H.; Hettelingh, J.-P. Systematic comparision of ILWAS, MAGIC, and ETD watershed acidification models. 2. Monte-Carlo analysis under regional variability. Water Resour. Res. 1991, 27 (10), 2591–2603. (19) Tarantola, A. Popper, Bayes and the inverse problem. Nat. Phys. 2006, 2 (8), 492–494. (20) Likens, G. E.; Bormann, F. H., Biogeochemistry of a Forested Ecosystem; Springer Science & Business: New York, NY, 1995. (21) Buso, D. C.; Likens, G. E.; Eaton, J. S. Chemistry of Precipitation, Streamwater, And Lakewater from the Hubbard Brook Ecosystem Study: A Record of Sampling Protocols and Analytical Procedures, General Technical Report NE-275; United States Department of Agriculture, Forest Service, Northeastern Research Station: Newtown Square, PA, 2000. (22) Gbondo-Tugbawa, S. S.; Driscoll, C. T. Factors controlling longterm changes in soil pools of exchangeable basic cations and stream acid neutralizing capacity in a northern hardwood forest ecosystem. Biogeochemistry 2003, 63 (2), 161–185. (23) Nodvin, S. C.; Driscoll, C. T.; Likens, G. E. Soil processes and sulfate loss at the Hubbard Brook Experimental Forest. Biogeochemistry 1988, 5, 185–199. (24) Aber, J. D.; Driscoll, C. T. Effects of land use, climate variation, and N deposition on N cycling and C storage in northern hardwood forests. Global Biogeochem. Cycles 1997, 11 (4), 639– 648. (25) Dentener, F. J.; Stevenson, D. S.; Cofala, J.; Mechler, R.; Amann, M.; Bergamaschi, P.; Raes, F.; Derwent, R. The impact of air pollutant and methane emission controls on tropospheric ozone and radiative forcing: CTM calculations for the period 19902030. Atmos. Chem. Phys. 2005, 5, 1731–1755. (26) Lien, L.; Raddum, G.; Fjellheim, A.; Henriksen, A. A critical limit for acid neutralizing capacity in Norwegian surface waters, based on new analyses of fish and invertebrate responses. Sci. Total Environ. 1996, 177, 173–193. (27) Holmberg, M.; Mulder, J.; Posch, M.; Starr, M.; Forsius, M.; Johansson, M.; Bak, J.; Ilvesniemi, H.; Sverdrup, H. Critical loads of acidity for forest soils: tentative modifications. Water, Air, Soil Pollut.: Focus 2001, 1 (1), 91–101. (28) Gapon, Y. N. On the theory of exchange adsorption in soils. J. Gen. Chem. USSR 1933, 3, 144–160. (29) Gaines, G. L.; Thomas, H. C. Adsorption studies on clay minerals. II. A formulation of the thermodynamics of exchange adsorption. J. Chem. Phys. 1953, 21 (4), 714–718. (30) Likens, G. E.; Buso, D. C. Variation in streamwater chemistry throughout the Hubbard Brook Valley. Biogeochemistry 2006, 78 (1), 1–30.

ES102243J

VOL. 44, NO. 23, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

9009