Disentangling climatic and anthropogenic controls on global terrestrial evapotranspiration trends

Intensified global hydrological cycle has been observed and modeled during the past few years (Huntington 2006, Gerten et al 2008, Wang et al 2010, Durack et al 2012, Douville et al 2013, Sterling et al 2013, Wu et al 2013, Gedney et al 2014). Terrestrial evapotranspiration (ET) is arguably the central component of this changing hydrologic cycle, and functions as a vital link between energy, water and carbon cycles, thereby having important implications for the availability and usage of fresh water resources by humans and terrestrial ecosystems (Seneviratne et al 2006, Trenberth et al 2009, Fisher et al 2011, Wang and Dickinson 2012).

Natural environmental factors (e.g. precipitation, temperature, incident solar radiation, soil moisture, wind and atmospheric teleconnections) regulate ET and its variability across different terrestrial ecosystems (Teuling et al 2009, Jung et al 2010, Wang et al 2010, Vinukollu et al 2011, Zhang et al 2012, Miralles et al 2014). These natural controls and limitations/co-limitations of ET are scale-dependent. Their mechanistic understanding is very important to predict the tendency and variability of ET (Wang and Dickinson 2012). Human-induced land use/land cover change, ground water withdrawals, and irrigation can directly alter the amount and timing of ET by modifying the local water and energy balances (Piao et al 2007, Gerten 2013, Leng et al 2013, 2014a, 2014b, Lo and Famiglietti 2013, Sterling et al 2013, Lei et al 2014c). Human activities that contribute to greenhouse gas emissions, atmospheric nitrogen deposition (NDE), and ozone pollution can also alter ET indirectly through changes in physiological, structural and compositional responses of plants (Gedney et al 2006, Betts et al 2007, Sitch et al 2007, Cao et al 2009, Leakey et al 2009). Discriminating these anthropogenic perturbations from natural factors is expected to increase in importance as anthropogenic transformation of the Earth System becomes more pervasive (Seneviratne et al 2010, Gerten 2013).

Based on mechanistic and empirical algorithms that are driven by remotely sensed observations, a variety of globally gridded diagnostic ET products have been compiled and evaluated in recent years (Willmott et al 1985, Fisher et al 2008, Jiménez et al 2009, Jung et al 2009, Sheffield et al 2010, Zhang et al 2010b, Miralles et al 2011, Mueller et al 2011, Vinukollu et al 2011, Zeng et al 2012, Schwalm et al 2013). These gridded ET estimates offer crucial sources and benchmarks for quantitative investigations of historical ET dynamics over the land surface. However, the accuracy of these observation-based ET products has yet to be reconciled due to limitations in underlying hypotheses and errors in input datasets (Mueller et al 2011, 2013, Polhamus et al 2012). Moreover, due to their reliance on the satellite observations, these datasets offer a limited historical temporal record that encompasses only a few decades (Badgley et al 2015).

To predict future changes in ET patterns, process-based simulation and understanding of the magnitudes, mechanisms and interactions that control historical ET dynamics will be required and should be within uncertainty of both historical and present-day observations. Mechanistic land surface models (LSMs), driven by measurement-based environmental properties, are useful tools for the detection and attribution of natural and anthropogenic effects on ET dynamics. For the past decade, global factorial LSM experiments have been conducted and analyzed by different modeling groups to investigate the separate effects of environmental stresses on land surface and subsurface runoff, river flow, ET and water use efficiency (Gedney et al 2006, 2014, Piao et al 2007, Shi et al 2011, 2013, Tian et al 2011, Liu et al 2012, Tao et al 2014). The role of climate impacts on these hydrologic variables has been characterized predominantly across different regions of the globe. The relative role of natural environmental change versus anthropogenic activities, however, was modeled to be heterogeneous and geographically dependent. Nevertheless, due to large differences in initial model conditions, driver data, and complex parameterizations that govern models, the simulated ET was demonstrated to vary in magnitudes and responses across models at both temporal and spatial scales (Wang et al 2010).

To disentangle these differences in simulated ET patterns and the relative role of model sensitivity and structure, the experimental setup and boundary/initial data must be similar among different participating models. We leveraged the controlled factorial experiments and model simulation protocol from the Multi-Scale Synthesis and Terrestrial Model Intercomparison Project (MsTMIP) (Huntzinger et al 2013). Further, we synthesized a global ET time series (1982–2010) based on a diverse set of diagnostic ET products (table 1), and the methodology reported recently in Mueller et al (2013). The partitioning of ET (e.g canopy transpiration (Tr) and evaporation from wet canopy and bare soil (ET–Tr)) and the variation of those ET components are poorly understood and less constrained by observations (Lawrence et al 2007, Jasechko et al 2013, Swenson and Lawrence 2014, Wang et al 2014). The MsTMIP modeling framework can advance our understanding of trends in ET by providing predictions of the individual ET components. In this study, we thus further investigated the contribution of individual influencing factors to the spatial and temporal characteristics of these ET constituents.

We created a merged diagnostic ET data (DIA) from 11 long-term diagnostic datasets, all based on different assumptions and constrained with extensive in situ observations or satellite retrievals or both (table 1). We remapped the monthly raw datasets from their original spatial resolutions to the half-degree resolution of the model output from 1982 through 2010 based on data availability. Following Mueller et al (2013), we applied both physical and statistical constraints for quality control and bias corrections. For the physical constraint, we developed a dataset of seasonal net radiation maxima using the surface radiation budget (SRB3.0) datasets (Gupta 1983). We then excluded grid points with values exceeding net radiation maxima by more than 25%. The outliers were identified as values that exceed ±3 standard deviations (Weedon 2011). Then the median values of these quality-controlled multiple ET estimates were treated as the merged product, and were comprehensively compared with the LSM results in this study. As shown in figure S1, the annual anomalies of the previously synthesized ET in Mueller et al (2013) are well within the spread of this newly-merged diagnostic data product. This updated product however, provides longer-term dynamics and is more amenable for studies at multi-decadal timescales.

To isolate the contributions of environmental drivers to multi-year ET variations, we utilized the factorial ET simulations from the MsTMIP data archive. Driven by the same environmental forcing (climate variability and trends, rising atmospheric CO₂ concentrations causing fertilization and reducing stomatal opening, nitrogen deposition, land use/land cover change, and soil texture and vegetation types), these state-of-the-art LSMs were employed to identify the principal drivers of interannual variability and multi-decadal changes of ET. Because the evaporation component for canopy and soil, and the snow sublimation, were not separately archived in the standard model outputs in the MsTMIP I protocol (Huntzinger et al 2013, 2015), we included all relevant available outputs, namely the ET, Tr and the total evaporation (ET–Tr). Four model experiments: (1) SG1 (time varying climate), (2) SG2 (time-varying climate and land use change history), (3) SG3 (time-varying climate, land use, and atmospheric CO₂), and (4) BG1 (time-varying climate, land-use, atmospheric CO₂ and nitrogen deposition), were analyzed to quantify the effects of each environmental forcing factor on the study variables for the years 1982 through 2010. The transient simulations began in 1901, turning on one time-varying driver at a time. Simulations BG1 or SG3 were used to address the combined impacts from various historical forcing agents for models with (BG1) or without (SG3) an explicit nitrogen cycle. Simulation or simulation differencing was used to quantify the contribution to ET and ET component changes from climate change (CLI) (derived from SG1), land use/land cover change (LUC) (derived from SG2-SG1), rising atmospheric CO₂ (CO₂) (derived from SG3-SG2), NDE (derived from BG1-SG3), or all forcing (ALL) (derived from BG1 or SG3) (table 1). To account for the overall effects from human activity (OTH), we derived the human-induced ET to be the difference between the BG1 and SG1 or SG3 and SG1 simulations.

Annual cropland area and total tree coverage information for the 1982–2010 period were derived from the merged product of the SYNergetic land cover MAP (Jung et al 2006) and the annual time series of the land use harmonization data (Hurtt et al 2011). Additional details on the aforementioned driver data and experimental design can be found in Wei et al (2014a, 2014b) and Huntzinger et al (2013, 2015).

Growing season ET generally dominates the annual sum over the vegetated area of land (Wang et al 2007). We focused our analysis on growing season ET for all observational and modeled data. The dynamic annual growing season information, used to mask the monthly ET between 1982 and 2010, was first determined from the global inventory modeling and mapping studies normalized difference vegetation index (NDVI3g) dataset (Pinzon and Tucker 2014) using a Savitzky–Golay filter (Chen et al 2004, Jonsson and Eklundh 2004). It was then refined by excluding the freeze period identified by the Freeze/Thaw Earth System Data Record (Kim et al 2011, 2012). In particular, the growing season of tropical rainforests was set to 12 months and it started in January.

Across the globe, statistically significant increasing trends of ET were recorded from 1982 to 2010 in the observation-based ET estimates (DIA) (1.18 mm yr⁻²) and modeled ET from the ALL simulation (0.93 ± 0.31 mm yr⁻²) (figures 1 and S2, and table S1). Significantly positive annual correlations between the simulated ALL ET and the observed ET were obtained, particularly in the Northern hemisphere (NH) (Land: R² = 0.58, p < 0.01, NH: R² = 0.72, p < 0.01, and the Southern hemisphere (SH): R² = 0.46, p < 0.01). The simulated multiyear increasing trend and interannual variability of the ALL ET were mainly explained by the CLI ET. In contrast, the overall human-induced OTH ET was predicted to decrease somewhat, and to exhibit relatively small interannual variations.

Zoom In Zoom Out Reset image size

Spatial analysis of linear trends of ET for the merged observation product revealed remarkably consistent increasing tendency over most continents (figure 2(a)). Local hotspots of reduced ET were diagnosed to occur in the arid regions of Western North America, central Africa, Northern China and Southeastern Asia. By contrast, the modeled changes of ALL ET underestimated the magnitude of ET changes in Eastern North America and Western Europe, and missed the ET decreases in central Africa. But the placement of increasing or decreasing trends in ALL ET largely agreed favorably with those of the observed ET trends, indicating the suitability of examining multi-year ET trends using the all-factor simulations.

Zoom In Zoom Out Reset image size

Spatial patterns of ET changes that are consistent between the ALL and CLM estimates confirm the dominance of climate forcing in explaining annual ET trends (figures 2(b), (c) and 3(a)). This dominant climatic response of ET trends was chiefly associated with concurrent annual precipitation changes (spatial R² = 0.34 for ALL ET and precipitation trend, and spatial R² = 0.30 for CLI ET and precipitation trends, respectively, P < 0.01), and tended to show large spatial heterogeneities of sign and magnitude (figure 4(a)). The spatial dominance patterns based on the partial correlations among the total growing season ET, precipitation, temperature and incident solar radiation affirmed that for the MsTMIP models, annual precipitation drove not only the interannual variability of ET, but primarily accounted for the multiyear ET trends over most land areas (figures 4(e) and (f)). Combined anthropogenic effects tended to decrease ET, most notably in Northeastern North America, Western Amazon, Northwestern Europe and tropical Asia (figure 2(d)). These effects were subject primarily to the net physiological and structural impacts of CO₂ concentration on the growth of plants in ecosystems (figures 2(e), 3(d), S2 and S3(a)).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Increasing nitrogen deposition led to increasing leaf area index (LAI) (figures 4(b) and S3(b)), and consequently to enhanced terrestrial ET, particularly over South America, Africa and Southeastern China (figures 2(f) and S2). The areas undergoing strong increase in forest fraction and decrease in cropland fraction, such as in central Eastern North America and central Europe, clearly showed increasing annual ET (figures 2(g), 4(c) and (d)). In contrast, regions with evident loss of trees, such as Eastern China and Southeastern South America, show a downtrend of annual ET. Compared to the CO₂ and nitrogen deposition effects, however, the effect of LUC on land ET was important locally. Relatively large uncertainties from the LUC were also found between individual models (figures S2 and S6).

Trends for the Tr and total ET–Tr were dominated by the climatic changes across various continents. For Tr, 85.4% of the study area was impacted by the climatic changes, and 88.7% for ET–Tr (figures 3(b), (c), S4(a)–(f)). Congruent with the response of ET changes to rising CO₂ (48.4 ppm during the period 1982–2010), most areas, especially these regions covered by tropical broadleaf evergreen trees and high latitude shrubs, showed decreasing Tr. This is due to the CO₂-induced reduction in stomatal conductance overwhelming the LAI-induced increase of canopy evaporation and transpiration under elevated CO₂ concentration (figures 3(e) and S4(j)). On the other hand, CO₂ fertilization would enhance canopy LAI through increasing photosynthate allocation to leaves, and caused more canopy transpiration and evaporation than the reduced transpiration by CO₂ physiological effects, especially over dry areas with sparse vegetation (e.g. the Western North America, central Eurasia, and Australia) (figures S3(a) and S4(j)). Reversed ET–Tr trends in these arid regions imply that decreasing soil evaporation was the dominant factor in changing ET–Tr (figures S4(j)–(l)). For most areas that showed decreasing Tr but increasing ET–Tr under CO₂ enrichment, the augmented evaporation of intercepted rainfall and increasing soil evaporation may have been coincidental.

Increasing ET caused by nitrogen deposition was due to enhanced Tr (figures 2(f), S4(m) and S5). A decrease of ET–Tr caused by the nitrogen deposition effect, as seen in central North America and in Western Europe, was due to reduced soil evaporation (figures S4(n) and S5). The latter is a consequence of the increasing LAI providing more shade and so reducing solar energy for soil evaporation. In addition, the increasing Tr further depleted soil water, which reduced soil evaporation. In the evergreen broadleaf forests of the Western Amazon and Congo basin, nitrogen deposition and higher LAI resulted in increasing canopy evaporation. The increase in canopy evaporation more than offset the decrease in soil evaporation and hence dominated the increasing ET–Tr and even the nitrogen-induced increase in total ET (figures S4(m)–(o)).

LUC led to a decreasing trend in Tr across densely inhabited regions that had experienced substantial land use perturbations (e.g. clearing trees for crops) during the study period. These occurred mainly in Southeastern South America and the Eastern China (figures 4(c), (d), S4(p) and S5). Tr trends showed a general negative sign over central Eastern North America and Western Europe, where croplands had been replaced mainly by forests and woodlands. This reduction of Tr with reforestation implies that the tree species that replaced the crops had lower stomatal conductance than the crop species, the younger and smaller trees of the returning forests had lower LAI than the croplands they replaced, or the available soil water for plants decreased because of the removal of irrigation. These aspects deserve further study.

Between 1982 and 2010, the observation-based and simulated ALL ET consistently showed a significantly increasing trend across the globe. These findings are consistent with previous studies, which reported an intensified global hydrological cycle in response to global warming following the Clausius–Clapeyron law (the relationship between equilibrium water vapor pressure and temperature, about 7% per °C of warming) (Held and Soden 2006), as well as increasing importance of the radiative component of ET (Johnson and Sharma 2010). Climatic factors accounted for much of the spatial and temporal variations in terrestrial ET, Tr and ET–Tr. This supports previous studies regarding the prevalent climatic mechanisms controlling the long-term ET trends such as temperature, precipitation, soil moisture, energy and internal climate variability (Teuling et al 2009, Jung et al 2010, Wang et al 2010, Vinukollu et al 2011, Zhang et al 2012, Ukkola and Prentice 2013, Miralles et al 2014).

In our study, the rising atmospheric CO₂ concentration, as tested by model factorial experiments, induced an overall suppression of Tr and hence a general decreasing ET. Our results further suggest that the sign of change and regional pattern of these CO₂ physiological effects on ET were moderated by changes in LAI. The overall response of ET was eventually determined by the balance among the changes of Tr, canopy evaporation and soil evaporation. These results are consistent with modeled and observed plant physiological responses to the increase of CO₂ concentration in the atmosphere (Betts et al 2007, Leakey et al 2009). They also reiterate previous findings that show the concurrent physiological and structural responses of vegetation to rising CO₂, and associated hydrological effects (Gedney et al 2006, Leipprand and Gerten 2006, Ainsworth and Rogers 2007, Betts et al 2007, Kurc and Small 2007, Piao et al 2007, Cao et al 2009, Leakey et al 2009, Lei et al 2014b).

Simulation experiments that consider NDE showed enhanced global LAI as a result of increasing nutrient availability (figures 4(b) and S3(b)). The nitrogen-induced enhancement of canopy Tr and canopy evaporation, however, was regionally offset by decreasing soil evaporation, and led to lower ET for the nitrogen fertilization effect. Nonetheless, mineralized nitrogen in the rooting system was governed by not only the amount of deposited N, but also by leaching and denitrification, which are affected by environmental conditions (Hovenden et al 2014). This highlights the necessity of better understanding the interactions among these environmental drivers, and the underlying mechanisms responsible for biogeochemical and hydrologic cycles.

Previous modeling studies (Boisier et al 2012, 2014, Shi et al 2013, Sterling et al 2013, Tao et al 2014) agree with our results that anthropogenic activities modified ET and its components locally, and human-induced LUC effects tended to counteract each other at a global scale. We found large uncertainties associated with LUC impacts among the MsTMIP LSMs, particularly over the NH and areas having marked land cover conversions. Though based on the same merged LUC dataset, different LSM groups prescribed the dynamic evolution of plant functional types with model-specific classifications (Wei et al 2014a, 2014b). The sensitivity of biophysical and biogeochemical processes to the reconstructed historical scenario of LUC, moreover, varied considerably from model to model (Huntzinger et al 2013). For example, for the SIB3-JPL models, abnormally higher LUC ET was simulated over the NH and global land compared to that of other models (figure S6). In SIB3-JPL, ET is a function of stomatal conductance and is sensitive to changes in photosynthetically active radiation (PAR). In LUC simulations, plant functional type changes over time, but the PAR is prescribed from present day NDVI climatology and is thus fixed to modern vegetation. This can lead to a bias in gross primary production in cases where grasslands are converted to forests, since the NDVI and resulting fraction of incident PAR absorbed by green leaving in the canopy (fPAR) are calculated from a modern day forest ecosystem but used to estimate stomatal conductance and ET for the historical grassland it replaced. The sensitivity to land-use change and cultivated ecosystems (e.g., irrigated croplands) reinforce the need for better LUC characterization, improved parameterization of ET in croplands, and the development of forcing datasets (e.g., PAR) that are not artificially dependent upon land cover. Improvements in these areas may help reduce the large inter-model spreads in the responses of ET to LUC.

Quantitative estimation of ET partitioning has been refined recently, but information on long-term variations and the precise drivers of each ET component are lacking (Jasechko et al 2013, Wang et al 2014). By using a multi-model ensemble, we assessed the annual trends of the Tr and ET–Tr over nearly three decades, and further estimated their spatial-temporal responses to various environmental stresses. These modeled results, however, remain rather uncertain without observational constrains that are sufficiently long and representative. Comprehensive synthesis of long-term observation-constrained ET components is needed to improve our understanding of the controlling mechanisms, and to better characterize the partitioning schemes.

The relative contribution of climate and anthropogenic activities to the spatio-temporal changes in ET was quantitatively characterized with the newly-merged ET and multifactor ensemble simulations from MsTMIP. In the LSMs, climate, CO₂, nitrogen deposition, and land use impacts were separated experimentally to determine the ET variations between 1982 and 2010. Climate, and in particular, changes in precipitation, was the dominant control of multi-year ET trends and variability. The overall CO₂ physiological and structural effect induced decreasing plants transpiration and the total ET, especially in areas where vegetation was dense. Compared to climate change and the elevated CO₂ effects, the impacts of nitrogen deposition and land use change on ET were less important and acted locally. Other detailed explorations are needed, such as the implementation of more compelling statistical techniques and fully-coupled modeling systems (Douville et al 2013, Wu et al 2013, Gedney et al 2014) to detect and attribute the natural and anthropogenic effects on ET with more certainty. ET-related feedback studies are also required to account for land-atmosphere interactions and anthropogenic impacts in the integrated earth system models (Seneviratne et al 2010, Bond-Lamberty et al 2014, Collins et al 2015) and to understand future trajectories of drought (Sheffield et al 2012, Zarch et al 2015). Given that human activities continue to grow and intensify in the Anthropocene Epoch, we emphasize utilizing multi-stream datasets and multi-modeling frameworks to better diagnose and project anthropogenic influences on terrestrial ET, hydrologic cycle and overall climate change.

This research was supported partially by the Terrestrial Ecosystem Science Scientific Focus Area (SFA), which is sponsored by the Terrestrial Ecosystem Science (TES) Program in the Climate and Environmental Sciences Division (CESD) of the Biological and Environmental Research Program in the US Department of Energy Office of Science. This research was supported partially by the Biogeochemistry–Climate Feedbacks SFA and project under contract of DE-SC0012534, which are both sponsored by the Regional and Global Climate Modeling (RGCM) Program in the Climate and Environmental Sciences Division (CESD) of the Biological and Environmental Research Program in the US Department of Energy Office of Science. CRS was supported by National Aeronautics and Space Administration (NASA) Grants #NNX12AP74G, #NNX10AG01A, and #NNX11AO08A. JBF carried out this research at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with NASA. Funding for the Multi-scale synthesis and Terrestrial Model Intercomparison Project (MsTMIP, http://nacp.ornl.gov/MsTMIP.shtml) activity was provided through NASA ROSES Grant #NNX10AG01A. Data management support for preparing, documenting, and distributing model driver and output data was performed by the Modeling and Synthesis Thematic Data Center at Oak Ridge National Laboratory (ORNL, http://nacp.ornl.gov), with funding through NASA ROSES Grant #NNH10AN681. Finalized MsTMIP data products are archived at the ORNL DAAC (http://daac.ornl.gov). This is MsTMIP contribution #5.

Acknowledgments for specific MsTMIP participating models:

Biome-BGC: Biome-BGC code was provided by the Numerical Terradynamic Simulation Group at University of Montana. The computational facilities provided by NASA Earth Exchange at NASA Ames Research Center.

CLASS-CTEM-N+: This research was funded by the Natural Sciences and Engineering

Research Council (NSERC) of Canada Discovery and Strategic grants. CLASS and CTEM models were originally developed by the Climate Research Branch and Canadian Centre for Climate Modelling and Analysis (CCCMa) of Environment Canada, respectively.

CLM: This research is supported in part by the US Department of Energy (DOE), Office of Science, Biological and Environmental Research. Oak Ridge National Laboratory is managed by UT-Battelle, LLC for DOE under contract DE-AC05-00OR22725.

CLM4VIC: CLM4VIC simulations were supported in part by the US Department of Energy (DOE), Office of Science, Biological and Environmental Research (BER) through the Earth System Modeling program, and performed using the Environmental Molecular Sciences Laboratory (EMSL), a national scientific user facility sponsored by the U.S.DOE-BER and located at Pacific Northwest National Laboratory (PNNL). Participation of M Huang in the MsTMIP synthesis is supported by the U.S.DOE-BER through the Subsurface Biogeochemical Research Program (SBR) as part of the SBR Scientific Focus Area (SFA) at the Pacific Northwest National Laboratory (PNNL). PNNL is operated for the US DOE by BATTELLE Memorial Institute under contract DE-AC05-76RLO1830.

DLEM: The Dynamic Land Ecosystem Model (DLEM) developed in the International Center for Climate and Global Change Research at Auburn University has been supported by NASA Interdisciplinary Science Program (IDS), NASA Land Cover/Land Use Change Program (LCLUC), NASA Terrestrial Ecology Program, NASA Atmospheric Composition Modeling and Analysis Program (ACMAP); NSF Dynamics of Coupled Natural-Human System Program (CNH), Decadal and Regional Climate Prediction using Earth System Models (EaSM); DOE National Institute for Climate Change Research; USDA AFRI Program and EPA STAR Program.

Integrated Science Assessment Model (ISAM) simulations were supported by the US National Science Foundation (NSF-AGS-12-43071 and NSF-EFRI-083598), the USDA National Institute of Food and Agriculture (NIFA) (2011-68002-30220), the US Department of Energy (DOE) Office of Science (DOE-DE-SC0006706) and the NASA Land cover and Land Use Change Program (NNX14AD94G). ISAM simulations were carried out at the National Energy Research Scientific Computing Center (NERSC), which is supported by the Office of Science of the US Department of Energy under contract DE-AC02-05CH11231, and at the Blue Waters sustained-petascale computing, University of Illinois at Urbana- Champaign, which is supported by the National Science Foundation (awards OCI-0725070 and ACI-1238993) and the state of Illinois.

LPJ-wsl: This work was conducted at LSCE, France, using a modified version of the LPJ version 3.1 model, originally made available by the Potsdam Institute for Climate Impact Research.

ORCHIDEE-LSCE: ORCHIDEE is a global land surface model developed at the IPSL institute in France. The simulations were performed with the support of the GhG Europe FP7 grant with computing facilities provided by LSCE (Laboratoire des Sciences du Climat et de l'Environnement) or TGCC (Tre`s Grand Centre de Calcul).

VISIT: VISIT was developed at the National Institute for Environmental Studies, Japan. This work was mostly conducted during a visiting stay at Oak Ridge National Laboratory.

This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-00OR22725 with the US Department of Energy. The United States Government retains and the publisher, by accepting the article for publication, acknowledges that the United States Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or reproduce the published form of this manuscript, or allow others to do so, for United States Government purposes. The Department of Energy will provide public access to these results of federally sponsored research in accordance with the DOE Public Access Plan (http://energy.gov/downloads/doe-public-access-plan).