Drivers of Future Northern Latitude Runoff Change

Abstract

Identifying the drivers of changing continental runoff is key to understanding current and predicting future hydrological responses to climate change. Potential drivers of runoff change include changes in precipitation and evaporation caused by climate warming, physiological responses of vegetation to elevated atmospheric CO₂ concentrations, increases in lower-atmosphere aerosols and anthropogenic land-cover change. In this study, we present a series of simulations using an intermediate-complexity climate and carbon cycle model to assess the contribution of each of these drivers to historical and future continental runoff changes. We present results for global runoff, in addition to northern latitude runoff that discharges into the Arctic and North Atlantic oceans, to identify any potential contribution of increased continental freshwater discharge to changes in North Atlantic deep-water formation. Between 1800 and 2100, the model simulated a 26% increase in global runoff and a 32% runoff increase in the northern latitude region. This increase was driven by a combination of increased precipitation from climate warming and decreased evapotranspiration caused by the physiological response of vegetation to elevated CO₂. When isolated, climate warming (and associated changes in precipitation) increased runoff by 16% globally and by 27% at northern latitudes. Vegetation responses to elevated CO₂ led to a 13% increase in global runoff and a 12% increase in the northern latitude region. These changes in runoff, however, did not affect the strength of the Atlantic Meridional Overturning Circulation, which was affected by surface ocean warming rather than by runoff-induced salinity changes. This study indicates that physiological responses of vegetation to elevated CO₂ may contribute to changes in continental runoff at a level similar to that of the direct effect of climate warming.

[Traduit par la rédaction] Il est primordial de déterminer les causes premières des changements dans le ruissellement continental afin de comprendre les réactions hydrologiques par rapport aux changements climatiques à l'heure actuelle et de les prédire. Voici des facteurs éventuellement responsables : les changements dans les précipitations et l’évaporation causés par le réchauffement climatique, la réaction physiologique des plantes à l'augmentation des concentrations de CO₂ dans l'atmosphère, l'augmentation de la concentration d'aérosols dans la basse atmosphère et les changements dans la couverture terrestre, qui découlent des activités humaines. Dans notre étude, nous présentons une série de simulations fondées sur un modèle du climat et du cycle du carbone de complexité intermédiaire afin d’évaluer le rôle de chacun de ces facteurs dans les changements du ruissellement continental dans le passé et à l'avenir. Nous présentons des résultats pour le ruissellement dans le monde, en plus du ruissellement sous les latitudes nordiques qui se déversent dans les océans Arctique et Atlantique Nord, afin d’établir la contribution éventuelle de l'augmentation de la décharge d'eau douce en provenance du continent dans les changements dans la formation d'eaux profondes dans l'océan Atlantique Nord. Entre 1800 et 2100, le modèle a simulé une augmentation de 26% du ruissellement mondial et de 32% du ruissellement sous les latitudes nordiques. Cette progression s'expliquait par la hausse des précipitations conjuguée au réchauffement climatique et à la diminution de l’évapotranspiration attribuable à la réaction physiologique des plantes à l'augmentation des concentrations de CO₂. Quand nous avons isolé ces facteurs, nous avons constaté que le réchauffement climatique (et les changements survenus dans les précipitations, induits par ce réchauffement) a augmenté de 16% le ruissellement à l’échelle mondiale et de 27% le ruissellement sous les latitudes nordiques. La réaction physiologique des plantes à l'augmentation des concentrations de CO₂ a entraîné une progression de l'ordre de 13% du ruissellement mondial et de 12% du ruissellement sous les latitudes nordiques. Cependant, ces changements ne se sont pas répercutés sur la circulation de renversement méridienne de l'Atlantique, qui était affectée par le réchauffement de la surface des océans plutôt que par des changements dans la salinité induits par le ruissellement. Notre étude démontre que la réaction physiologique des plantes à l'augmentation des concentrations de CO₂ peut entraîner des changements dans le ruissellement continental comparables à l'effet direct du réchauffement climatique.

Keywords:

• climate change
• continental runoff
• physiological forcing
• vegetation feedbacks
• ocean circulation

1 Introduction

Predicting runoff change is challenging given the complexity and uncertainty associated with the climate's response to anthropogenic forcings and, particularly, the hydrological changes associated with anthropogenic climate warming. However, quantifying a range of possible responses is critical given the important ramifications of a modified hydrological cycle (Allen and Ingram, 2002; Milly et al., 2005). Analysis of historical stream discharge has identified a trend towards increasing continental runoff at high northern latitudes (Labat et al., 2004; Dai et al., 2009). There is some disagreement over historical records at low to mid-latitudes, with the result that estimates of global runoff changes range from an increasing trend (Labat et al., 2004) to no statistically detectable trend (Dai et al., 2009). Climate model predictions of future runoff change are also highly variable; although there is general agreement that continental runoff is expected to change on the order of 5–15% per degree of global warming. These projected changes include an anticipated decrease in runoff in many temperate and subtropical regions and a corresponding increase at high latitudes as a result of an overall intensification of the global hydrological cycle (Solomon et al., 2010).

Although these model projections account for changes in precipitation and evaporation associated with anthropogenic climate warming, they generally do not include other potentially important drivers of continental runoff change. In particular, several recent studies have highlighted the potential role of terrestrial vegetation responses to elevated atmospheric CO₂ concentrations as a contributor to global runoff changes. This so-called “physiological forcing” results primarily from decreased stomatal conductance because plants are able to use available soil moisture more efficiently in higher ambient CO₂ concentrations. The effect of vegetation physiological forcing has been shown in some regions to amplify the response of runoff to climate warming (Betts et al., 1997, 2007; Levis et al., 2000; Medlyn et al., 2001; Gedney et al., 2006; Bernacchi et al., 2007; Cao et al., 2009, 2010).

In general, vegetation physiological forcing tends to decrease transpiration, leading to increased soil moisture and a corresponding increase in continental runoff (Levis et al., 2000; Piao et al., 2007). This effect can be damped somewhat by plant structural responses to CO₂ changes, whereby enhanced leaf growth can lead to an increase in available surface area for water transpiration and a corresponding increase in moisture recirculation to the atmosphere. In so doing, plant structural responses are thought to at least partially offset the effect of the physiological forcing, though the outcome is also highly dependent on regional landscape characteristics (Betts et al., 1997; Levis et al., 2000; Piao et al., 2007).

Anthropogenic land-cover change has also been identified as a potential contributor to historical runoff change through its influence on vegetation and land surface characteristics associated with cropland establishment and abandonment (Piao et al., 2007). For example, deforestation and/or agricultural expansion might be expected to affect runoff as a result of changes in evapotranspiration and soil water retention. In addition, the differing albedos of forests and crops could modify regional radiative forcing and could therefore affect precipitation patterns via altered surface temperatures. At a global scale, however, studies have found relatively little impact of land-cover change on historical runoff changes (Gedney et al., 2006).

Finally, the growing body of knowledge on the effect of anthropogenic aerosols on climate processes has revealed several potential means by which increased atmospheric aerosols may affect continental runoff. Aerosols in the atmosphere absorb and scatter shortwave radiation resulting in both a cooling of the atmosphere and a decrease in the energy received at the Earth's surface. This has the potential to decrease precipitation (as a result of atmospheric cooling) or to decrease surface evaporation (because of the lower availability of surface energy), either of which could affect overall continental runoff. In addition, anthropogenic aerosols can act as cloud condensation nuclei, leading to changes in their physical and chemical properties, potentially further altering precipitation patterns (Boucher et al., 2004; Solomon et al., 2007). Some studies have found a reduction in precipitation and evaporation as a result of lowered surface solar radiation from aerosol forcing (e.g., Liepert et al., 2004). However, the results of Gedney et al.’s (2006) analysis of twentieth century runoff observations showed little evidence of aerosol impacts over northern latitudes.

In this study, we present transient model simulations of historical and future climate change using an intermediate complexity global climate and carbon cycle model. In a series of sensitivity experiments, we assess the relative contributions of physical climate changes caused by greenhouse gas increases, vegetation physiological forcing and structural changes, land-use and land-cover changes, and anthropogenic aerosols as potential drivers of observed and projected future continental runoff changes. We focus, in particular, on high northern latitudes, which have seen increased historical runoff, because of the possibility that changes in continental runoff could have implications for ocean circulation and deep water formation in the North Atlantic. Our use of transient simulations with a coupled climate and carbon cycle model enables an examination of these potential links between land surface processes and ocean circulation and, in addition, advances previous research focused on equilibrium-only climate simulations.

2 Methodology

We use here the University of Victoria's Earth System Climate Model (UVic ESCM) version 2.9 (Weaver et al., 2001; Eby et al., 2009), a coupled climate-carbon model of intermediate complexity. Version 2.7 of the UVic ESCM was a contributing model to the Coupled Carbon Cycle Climate Model Intercomparison Project (C⁴MIP; Friedlingstein et al., 2006) and to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (Meehl et al., 2007). Version 2.9 has an equilibrium climate sensitivity of 3.5°C and a transient climate response of about 2°C, which falls approximately within the mid-range of more comprehensive climate models (Meehl et al., 2007).

The atmospheric component of the UVic ESCM is a reduced complexity (1 vertical level) energy-moisture balance model, with a spherical grid resolution of 3.6° longitude by 1.8° latitude. The atmospheric model controls heat and moisture transport through Fickian diffusion, as well as moisture transport via advection. Precipitation in the model occurs when relative humidity exceeds 85% (Weaver et al., 2001). This simple atmosphere is coupled to a three-dimensional ocean general circulation model, the Modular Ocean Model, version 2.2 (MOM2.2; Pacanowski, 1996) which has 19 vertical levels (Weaver et al., 2001). MOM2.2 is driven by surface boundary conditions from the atmosphere and land surface; in particular, the freshwater flux (as a function of precipitation, evaporation and runoff) is added to the surface ocean in the form of a salinity flux. The maximum Meridional Overturning Circulation (MOC), whose value corresponds to the strength of deep-water formation in the North Atlantic (Weaver et al., 2001), is also calculated in the ocean model.

The land surface scheme used is the Met Office Surface Exchange Scheme (MOSES) (Cox et al., 1999; Essery et al., 2003), coupled to a biochemical vegetation model, Top-down Representation of Interactive Foliage and Flora Including Dynamics (TRIFFID), which simulates biophysical processes for five plant functional types found in the model (Cox, 2001). TRIFFID is a dynamic vegetation model, which simulates the structure and spatial distribution of vegetation in response to changing climate conditions. MOSES simulates land-atmosphere carbon fluxes and calculates various components of the hydrological cycle, including evapotranspiration, soil moisture and runoff. Evapotranspiration in the model is calculated using the Penman-Monteith formulation and varies as a function of both local climate conditions and atmospheric CO₂ concentration (Cox et al., 1999; Essery et al., 2003). The version of MOSES implemented in the UVic ESCM is a simplified version of MOSES2 as described in Essery et al. (2003); in particular, soil moisture here is stored in a one-metre soil layer, as opposed to the four soil layers in the original version. Excess moisture is diverted to the ocean as runoff via 32 realistic river basins, according to a river-routing scheme described in Weaver et al. (2001). We note that version 2.9 of the UVic ESCM does not represent permafrost, wetlands or frozen soils; this is an area of active model development (Avis et al., 2011) but represents a limitation of the current model in representing Arctic continental runoff responses to climate warming.

In our study, we performed five transient model simulations following a 10,000 year spin-up period under pre-industrial forcing. For the reference simulation (ALL) we included all potential drivers of runoff change: the Special Report on Emissions Scenarios (SRES) A2 business-as-usual CO₂ concentration scenario, including additional prescribed forcing from non-CO₂ greenhouse gases (Nakicenovic et al., 2000); physiological and structural responses of vegetation to CO₂ changes; historical land-cover change from Ramankutty and Foley (1999); and the direct effect of anthropogenic sulphate aerosols, represented as a prescribed aerosol optical depth following historical and future A2 aerosol emissions (see Matthews et al. (2004) and University of Victoria (2010) for more information about aerosol forcing in the UVic model). Subsequently, we performed additional simulations wherein each of the four runoff drivers was turned off. For aerosol effects and land-use change, the effect was removed from each respective simulation by holding the forcing constant at pre-industrial conditions. We isolated the effect of vegetation-CO₂ (V-CO₂) physiological forcing by holding CO₂ concentrations constant at pre-industrial levels with respect to the vegetation component of the model while allowing CO₂ radiative forcing to affect the physical climate model. Conversely, the climate effect of greenhouse gases was isolated by masking the radiative forcing from increased greenhouse gases in the climate model, while allowing CO₂ increases alone to affect vegetation growth.

Model simulations are summarized in Table 1. In the results presented here, we isolated the effect of individual drivers by calculating the difference between each of the four simulations and the reference simulation ALL. We extracted and analyzed runoff changes both globally and for land areas that discharge into the Arctic and North Atlantic oceans north of 45°N latitude, and between 105°W and 90°E longitude (see grey area marked on Fig. 3). Within this region, we calculated runoff, land temperature, precipitation and evaporation over the river basins whose discharge areas were within the specified latitude and longitude ranges. Ocean fields (temperature and salinity) were calculated for all ocean surface areas within the region.

Table 1. Description of model simulations.

Simulation	Description
ALL	Reference simulation with all runoff drivers included: CO2 + non-CO2 radiative forcing; vegetation-CO2 physiological forcing; direct effect of sulphate aerosols and historical land cover change.
GG	Effect of CO2 and non-CO2 greenhouse gas forcing alone*
V-CO2	Effect of vegetation-CO2 physiological forcing alone*
AER	Effect of direct sulphate aerosol forcing alone*
LCC	Effect of historical land cover change alone*
*Effect of single driver calculated as the difference between a simulation without this process and the reference simulation ALL.

3 Results

Under present-day conditions, globally averaged runoff in the model was 1.1 Sv (where 1 Sv = 10⁶ kg s⁻¹). In our selected northern latitude region, which includes river drainage into the North Atlantic Ocean and part of the Arctic Ocean, total runoff was 0.13 Sv at present day. These values compare well with recent estimates by Dai and Trenberth (2009), who reported a total continental discharge of about 1.15 Sv for the year 2004 and an Arctic Ocean discharge of 0.14 Sv.

Simulated trends in globally averaged runoff, global land surface air temperature, land precipitation and land evaporation are shown in Fig. 1 (global) and Fig. 2 (northern latitude region). Over the twentieth century, global runoff in ALL increased by 7.5 mm yr⁻¹, or 3.4% relative to pre-industrial levels, accompanying a 0.6°C increase in global temperature (Figs 1a and 1b). This represents a 5.6% increase per degree of global warming, which is consistent with the 4% per degree Celsius reported by Labat et al. (2004) based on historical runoff and temperature observations, though not with Dai and Trenberth (2009) who did not detect a significant trend in global runoff over the past 50 years. Over the specified northern latitude land region, runoff increased by 9.3 mm yr⁻¹ (Fig. 2a), corresponding to a total discharge increase of 0.006 Sv. By way of comparison, Dai and Trenberth (2009) reported a significant increase in continental discharge to the Arctic Ocean of 0.014 Sv between 1948 and 2004, with no significant trend in the Atlantic as a whole; this is broadly consistent with the simulated results in our northern latitude region, which covers a portion of both the Atlantic and Arctic oceans (see area marked on Fig. 3). Over the twenty-first century, changes in global runoff per degree of global warming were slightly larger (6.3% per degree Celsius), which is generally consistent with the range of 5–15% per degree Celsius reported from a range of different models by Solomon et al. (2010). In our selected northern latitude region (Fig. 2a), runoff increased by 32% over the twenty-first century alongside a 4.1°C temperature rise—a slightly higher regional sensitivity of 7.8% per degree Celsius.

Fig. 1 Simulated global trend in a) continental runoff, b) land temperature, c) precipitation over land, and d) evaporation over land, in response to the drivers ALL, GG, V-CO₂, LCC and AER.

Fig. 2 Simulated northern latitude trend in a) continental runoff, b) land temperature, c) precipitation over land, and d) evaporation over land, in response to the drivers ALL, GG, V-CO₂, LCC and AER. The region shown here covers all river basins that drain into the North Atlantic and Arctic oceans north of 45°N latitude and between 105°W and 90°E longitude, as indicated by the shaded box in Fig. 3.

Fig. 3 Spatial distribution of precipitation–evapotranspiration (P–E) change between 1800 and 2100, in response to each driver: a) ALL simulation; b) greenhouse gas forcing; c) vegetation physiological forcing; d) aerosol forcing; and e) land cover change. The grey bounding box represents our selected northern latitude region.

As can also be seen in Fig. 1, greenhouse gas increases (line GG) were the primary driver of global runoff changes as a result of increased precipitation over all land areas at a rate of 1.2% per degree Celsius of land temperature increase. In the northern latitude region, warming from greenhouse gas increases led to both an increase in precipitation and a decrease in evaporation resulting in a relatively larger increase in continental runoff (Fig. 2). Vegetation physiological forcing was the second largest contributor to simulated runoff changes. At the global scale, the effect of vegetation physiological forcing was only slightly smaller than the effect of greenhouse gases; although it was closer to half the effect of greenhouse gases in the northern latitude region.

The mechanism for runoff changes caused by physiological forcing was fundamentally different from that of greenhouse gases. In the case of V-CO₂, physiological forcing resulted in a large decrease in precipitation, although unlike the case of GG, this was not a function of temperature changes (which were small); rather, precipitation in V-CO₂ was driven by a large decrease in evapotranspiration (Figs 1d and 2d) as a result of increased water-use efficiency at higher CO₂ levels. Consequently, there was a clear division between the two mechanisms of influencing changes in continental runoff. Climate warming tended to increase precipitation and drive a parallel increase in runoff (line GG), whereas elevated CO₂ tended to induce decreased transpiration, leading to decreased water recirculation to the atmosphere, increased soil moisture and a consequent further increase in runoff. The runoff change in the ALL simulation was driven in approximately equal measure by each of these two processes

Aerosols (AER) and land cover changes (LCC) were clearly of secondary importance in driving both historical and future runoff changes. Aerosols tended to have the opposite effect to greenhouse gases, leading to a cooling of land temperatures and an associated decrease in precipitation and runoff (with little change in evaporation). The effect of aerosols was relatively larger in the northern latitude region, because of the higher concentration of anthropogenic aerosols over northern continental areas; even here, however, aerosols had less than half the effect of vegetation physiological forcing by the year 2100. Land cover changes had an almost negligible effect on historical runoff changes and were not relevant to future changes given that areas of anthropogenic land use did not change in the simulations after the year 2000.

Land temperatures (Figs 1b and 2b) increased in both the GG and ALL simulations, though warming in the GG simulation was greater (4.8°C globally and 5.8°C in the northern region at 2100) than in the ALL simulation (4.2°C globally and 4.8°C in the northern region) as a result of the absence of the effect of reflective aerosols in the GG simulation. Globally, aerosols accounted for about 0.9°C cooling by the year 2000, with a larger cooling of 1.3°C in the northern latitude region. Temperatures continued to cool slightly for the first half of the twenty-first century, then warmed in the latter half of the century because of decreased aerosol forcing after approximately 2040. Vegetation physiological forcing and land cover changes did not play a prominent role in affecting air temperature change over the course of the two centuries. Vegetation structural feedbacks did result in a small amount of warming (approximately 0.25°C globally and in the northern region) as a result of increased leaf area and spatial coverage of vegetation in response to elevated CO₂ concentrations, leading to decreased surface albedo. Land-use change led to an even smaller amount of cooling (approximately 0.1°C) because of the higher surface albedo associated with agricultural vegetation types compared to forest vegetation.

Figure 3 shows the spatial distribution of precipitation minus evaporation (P–E) changes from 1800 to 2100 for each simulation. The spatial patterns seen here reflect the contrasting mechanisms behind simulated runoff changes shown in Fig. 2. For instance, in the GG simulation, a greater P–E increase is seen along coastal regions and in areas of higher elevation; this reflects an increase in precipitation under climate warming that occurs predominantly over the oceans and over orography. Similarly, cooling in the AER simulation generated a similar pattern of P–E changes reflecting decreased precipitation. By contrast, P–E changes in the V-CO₂ simulation are more representative of global vegetation distributions, which reflects the pattern of decreased evapotranspiration driven by vegetation responses to elevated CO₂ (CO₂ physiological forcing). In general, the spatial pattern of P–E in the ALL simulation reflects the combined P–E changes in the GG and V-CO₂ simulations, with smaller contributions from aerosols and land cover changes.

Finally, we assessed to what extent the simulated runoff increases shown here may have affected deep-water formation in the North Atlantic. Figure 4 shows the simulated change in the Atlantic Meridional Overturning Circulation (AMOC) (Fig. 4a) in response to changes in surface water temperature (Fig. 4b) and salinity (Fig. 4c; practical salinity scale is used). For all simulations, AMOC changes appear to correlate with both salinity and temperature variations, with a consistent pattern of decreased (increased) strength of AMOC associated with increased (decreased) temperature and decreased (increased) salinity. However, it is clear that North Atlantic salinity changes were not driven by changes in continental runoff but rather by changes in P–E over this ocean region. In particular, the V-CO₂ simulation, which showed a large increase in continental runoff, did not simulate decreased North Atlantic salinity nor a change in the strength of the AMOC. Instead, ocean circulation appears to be governed primarily by changes in high-latitude temperatures (and the associated change in the latitudinal temperature gradient); climate warming in the ALL and GG simulations led to a slowing of the AMOC, whereas climate cooling in the AER simulation led to a slightly stronger AMOC. Changes in continental runoff did not have a noticeable effect.

Fig. 4 Simulated trends in a) the Atlantic Meridional Overturning Circulation (AMOC), b) surface ocean temperature changes and c) surface ocean salinity changes in response to drivers ALL, GG, V-CO₂, LCC and AER. Surface ocean temperature and salinity values are averaged over the North Atlantic and Arctic oceans, north of 45°N latitude and between 105°W and 90°E longitude.

4 Discussion

In this study we investigated the variation in and explanation for future continental runoff changes using transient climate simulations driven by a suite of anthropogenic forcings. Climate change (including climate warming and changes in precipitation) and vegetation physiological responses to elevated CO₂ are projected to be the predominant drivers of runoff change for the twenty-first century, contributing in approximately equal measure to simulated global runoff increases. By contrast, land cover change and anthropogenic aerosols had secondary effects on continental runoff trends. We further assessed the extent to which simulated changes in continental runoff could affect the strength of North Atlantic deep water formation; while deep-water formation did decrease in response to climate change, this occurred as a result of temperature and P–E changes rather than runoff-induced salinity changes.

The effect of V-CO₂ interactions that we have shown here represents the net effect of both physiological and structural responses of vegetation to CO₂, although we have also identified the specific climatic outcome of each of these two vegetation processes. The effect of the vegetative structural feedback was primarily to generate a small increase in surface temperature; this occurred as a result of CO₂ fertilization, in which forest expansion over grasslands led to decreased local surface albedo, consequently increasing absorption of shortwave radiation (see also Matthews, 2007). In contrast, the effect of vegetation-CO₂ physiological forcing was seen predominantly in hydrological cycle changes; elevated CO₂ led to increased vegetation water-use efficiency, decreased evapotranspiration and a consequent increase in continental runoff.

Our model results contribute to the growing number of studies suggesting that vegetation physiological forcing from increasing atmospheric CO₂ is an important contributor to both runoff changes as well as to the overall climate response to CO₂ emissions (Betts et al., 2007; Boucher et al., 2009; Cao et al., 2010). The effect of vegetation physiological forcing shown here supports the findings of Betts et al. (2007) and Boucher et al. (2009), who use a similar version of the vegetation model employed here, although coupled in their case to the Hadley Centre global climate model. In our study, the effect of physiological forcing on the hydrological cycle was also substantially larger than the direct effect of vegetation changes on surface temperature; we simulated a relatively small temperature increase caused by CO₂ fertilization of vegetation (less than 0.25°C increase in the northern latitude region and only a 0.1°C increase averaged over the globe). By contrast, other studies have found a somewhat larger effect; for example, Cao et al. (2010) simulated an approximately 0.42°C increase in global surface temperature caused by vegetation physiological responses to elevated CO₂. This difference reflects inter-model uncertainty, as well as differences in methodology between our studies; in particular, we have carried out transient model simulations here, whereas Cao et al. (2010) used equilibrium time-slice simulations. This has implications for the time scale of vegetation responses: while hydrological responses to elevated CO₂ occur quickly, the structural response and particularly the lateral expansion of vegetation is limited by the fairly slow growth of vegetation in the model. Consequently, we suggest that the transient methodology we have adopted provides a more realistic representation of the relative balance of hydrological and temperature responses to CO₂-induced vegetation changes.

Historical land cover change did not contribute in any large measure to simulated runoff changes in our study. This is consistent with the conclusions of Gedney et al. (2006), who found that land cover change did not have a significant influence on observed continental runoff changes. There is some evidence that deforestation can lead to local-scale changes in runoff because of decreased evapotranspiration and soil water retention (e.g. Costa et al., 2003; Bradshaw et al., 2007); however, these studies have focused on tropical forests, which have experienced a much larger extent of land cover change compared to northern latitudes. There is also the potential for interactions between future land-use change and the response of vegetation to increased CO₂. For example, agricultural vegetation is likely to respond differently to CO₂ changes compared to natural vegetation types—a phenomenon which we have not explicitly considered here. In addition, we have considered changes in cropland areas only as a representation of land cover change and have not included pastures as an additional source of land conversion, which would probably slightly increase the total effect of land-use change shown here.

In the case of aerosols, we did simulate decreased precipitation in response to the cooling induced by increased anthropogenic aerosols, which is consistent with other models, as well as the climate response to aerosols resulting from recent volcanic eruptions (Trenberth and Dai, 2009). This change in precipitation led to a small decrease in runoff, both globally and in the northern latitude region considered here. We note, however, that we have considered only the direct effect of sulphate aerosols and not a full representation of the possible direct and indirect effect of a range of anthropogenic aerosols; this could also lead to an underestimate of the effect of aerosols on the global hydrological cycle.

We note also, that none of the models which have been used thus far to assess the effect of vegetation physiological forcing on continental runoff dynamics include the impacts of permafrost and changes in frozen soils on high-latitude soil water retention in response to climate warming. This is potentially an important additional driver of runoff changes (see e.g., Rawlins et al., 2003; Slater et al., 2007) which would alter the response of high-latitude runoff to continued climate warming. The response of permafrost to climate warming remains a critical open question regarding both the potential for greenhouse gas feedbacks to climate warming, as well as the possible implications for continental runoff changes.

We did not find that simulated runoff changes were sufficient to induce a reduction in the strength of North Atlantic deep-water formation. Our model did simulate a decrease in the MOC of 4.5 Sv between 1800 and 2100, from a pre-industrial state of 21.5 Sv (about a 20% decrease). This general trend agrees with previous research; for example, Bryan et al. (2006) showed a 22–26% circulation decline per century in response to a 1% per year increase in CO₂ forcing. This simulated decrease was driven by temperature increases in the North Atlantic (leading to associated changes in precipitation and sea-surface salinity), as well as by a decreased equator to pole temperature gradient brought about by climate warming. By contrast, simulated runoff changes did not affect the strength of the overturning simulation, as evidenced by the lack of AMOC response to runoff changes from vegetation physiological forcing. Our results suggest that vegetation responses to elevated CO₂, while important for projections of hydrological cycle changes, are not likely to influence projections of the AMOC response to anthropogenic climate warming. However, we note that a significant potential source of freshwater input to the North Atlantic from melting land ice is not included in this study; this has been identified previously as a more important factor influencing salinity in the North Atlantic than stream discharge (Bryan et al., 2006). Clark et al. (2002) found that only an approximately 0.1 Sv change to the hydrological cycle was sufficient to induce a thermohaline circulation response during the last glaciation, although we note that recent research has questioned the ability of coarse-resolution ocean models to simulate AMOC responses to continental freshwater discharge (Condron and Winsor, 2011). Nevertheless, our findings represent an approximately 0.04 Sv increase between 1800 and 2100 in freshwater discharge to the North Atlantic, which if combined with significantly increased land ice melt, may be sufficient to affect a more dramatic deep-water formation response than simulated here.

Considering our findings more broadly, they do suggest the potential for major regional alterations in the hydrological cycle, which could have significant implications for both human and ecological communities. One possible positive implication may be that increased runoff could result in increased water availability for agriculture and other human uses. However, such increases will likely be concentrated in areas that already have an ample water supply, whereas more arid regions may be negatively affected by regional declines in precipitation. Ultimately, the combination of greater soil moisture availability and a warmer climate will require adaptation in how humans make use of the available water resources in a changing climate.

5 Conclusions

Projecting runoff change for the twenty-first century is an important step in understanding potential future climate change as anthropogenic activities continue to be the dominant driver of changes in the Earth system. Understanding the complexities of the climate system components will help us prevent harmful climate impacts that have begun as a result of unmitigated temperature increases and changes in the spatial patterns of precipitation. This study concludes that plant responses to atmospheric CO₂ concentration will play an important role, in addition to the direct effect of climate warming, in driving future changes in the global hydrological cycle. Vegetation changes will likely also influence regional patterns of temperature changes, although this effect is secondary to the hydrological effects of vegetation physiological responses to elevated CO₂. The relationship between increased freshwater discharge and ocean circulation response was shown to be less significant, with changes in North Atlantic deep-water formation responding primarily to temperature changes rather than changes in continental runoff. The use of transient simulations with our coupled climate-carbon cycle model has been a useful approach as it has enabled us to investigate potential links between land surface processes and ocean circulation, while advancing previous runoff change research that, thus far, has focused largely on equilibrium-only climate simulations.

Acknowledgements

Funding for this research was provided by the Natural Sciences and Engineering Research Council of Canada (Undergraduate Student Research Award and Discovery Grant) as well as the Canadian Foundation for Climate and Atmospheric Sciences. We would also like to acknowledge the excellent assistance provided by D. Seto and to thank the three anonymous reviewers for their helpful comments and suggestions.