On the influence of shrub height and expansion on northern high latitude climate

There is a growing body of empirical evidence documenting the expansion of shrub vegetation in the circumpolar Arctic in response to climate change. Here, we conduct a series of idealized experiments with the Community Climate System Model to analyze the potential impact on boreal climate of a large-scale tundra-to-shrub conversion. The model responds to an increase in shrub abundance with substantial atmospheric heating arising from two seasonal land–atmosphere feedbacks: a decrease in surface albedo and an evapotranspiration-induced increase in atmospheric moisture content. We demonstrate that the strength and timing of these feedbacks are sensitive to shrub height and the time at which branches and leaves protrude above the snow. Taller and aerodynamically rougher shrubs lower the albedo earlier in the spring and transpire more efficiently than shorter shrubs. These mechanisms increase, in turn, the strength of the indirect sea-ice albedo and ocean evaporation feedbacks contributing to additional regional warming. Finally, we find that an invasion of tall shrubs tends to systematically warm the soil, deepen the active layer, and destabilize the permafrost (with increased formation of taliks under a future scenario) more substantially than an invasion of short shrubs.


Introduction
In the northern high latitudes (NHL), rapid environmental changes are under way in response to climate change, and are anticipated to continue in the future (Overpeck et al 1997, Hinzman et al 2005. These changes include a rapid warming rate (Serreze et al 2000) attributed to human activities (Gillett et al 2008), a sharp decline in sea-ice and terrestrial snow (Parkinson et al 1999, Min et al 2008, and a progressive thawing of permafrost (Jorgenson et al 2006). Initially triggered by the radiative effects of increasing greenhouse gas (GHG) emissions, the warming is amplified by several regional positive feedback mechanisms (e.g., Chapin et al 2005Chapin et al , 2008, such as the darkening of the surface caused by the earlier seasonal melting of snow exposing soil, plants, and sea (e.g. Foley 2005, Dye andTucker 2003).
Changes expected under a warming climate also include the expansion of trees and shrubs into regions currently occupied by tundra (e.g., Chapin et al 2005). In anticipation of such changes, a number of global climate model studies have evaluated the feedbacks induced by a tundra-to-forest conversion. In general, these studies indicate that an expansion of the boreal evergreen forest will amplify regional warming through an albedo feedback (Bonan et al 1992, Foley et al 1994, Levis et al 1999, while an expanding temperate deciduous forest (another possible outcome) will additionally cause a positive transpiration-induced greenhouse feedback (Swann et al 2010). However, the invasion of tundra by a tall forest is unlikely to occur in the current century (Chapin et al 2005) and is mainly predicted by equilibrium vegetation models (e.g., Kaplan and New 2006) which typically do not account for processes associated with the migration and the succession of plants (Epstein et al 2007).
In contrast, no global climate experiments have been conducted thus far to estimate the role of shrub expansion on NHL climate, despite a growing body of empirical evidence (photographs, transect studies, satellite-derived indices, field experiments, and local testimonies) documenting the increase in shrub abundance and size in various locations of the circumpolar arctic (e.g., Tape et al 2006, ACIA 2005. For instance, Sturm et al (2001b) showed that shrub area has increased by 1.2% decade −1 since 1950 in Northern Alaska and satellite data are now used to document the presence of short and tall shrubs across this region . Forbes et al (2010), using remote sensing evidence, detected an enhanced greening signal likely associated with shrub expansion in the tundra since the early 1980s. This corroborates earlier studies that contrasted consistent greening trends in pan-Arctic tundra areas (with a longer growing season length; e.g. Goetz et al 2005, Bunn andGoetz 2006) with the decline in photosynthetic activity found in forest areas (Angert et al 2005). Small shrubs are already present in most tundra areas (Chapin et al 2005), and therefore can grow in stature and expand further under increasingly favorable environmental conditions, as suggested by field measurements (Sturm et al 2005b, Tape et al 2006, Forbes et al 2010, Hallinger et al 2010, experimental warming treatments (Walker et al 2006), and warm periods of the past climate (Bret-Harte et al 2002). Such changes in ecosystem structure may affect the biogeochemical and biophysical processes regulating the exchanges of energy, water and carbon between the biosphere and atmosphere . Shrubs can also promote their own development by favoring snow accumulation and thereby winter soil warming, which stimulates soil microbial activity and mineralization of nitrogen that could support further shrub growth (Sturm et al 2005b). Finally, a plant dynamics model (ArcVeg; Epstein et al 2007) estimated that a 2 • C warming can lead to a rapid increase in tall shrubs in regions occupied by low shrub, or increase their abundance where they are already dominant.
Here, we investigate the possible climate feedbacks arising from the biophysical aspects of a tundra-to-shrub conversion. Our analysis is guided by two questions: (1) What are the biophysically induced effects of shrub expansion on the NHL climate? (2) Is this climate response sensitive to the stature of shrubs? As in a forest expansion scenario (Swann et al 2010), we expect feedbacks driven primarily by altered albedo and evapotranspiration (ET). However, an increase in shrub height can also affect the timing of those feedbacks, as tall shrubs protruding above the snow and short shrubs buried under the snow affect the surface energy budget during the snow-covered period differently (Sturm et al 2001a, 2005a. Through shading of the surface, shrubs can keep soil temperatures cooler than in surrounding tundra, thereby protecting permafrost (Yi et al 2007, Blok et al 2010, though a strong climate warming signal associated with the aforementioned land biophysical feedbacks could overwhelm the ground shading cooling effect of shrubs (Lawrence et al 2011b).

Methods
To investigate the role of a shrub expansion on NHL climate, we performed nine 30 yr steady-state simulations with the . Currently, only one type of shrub is represented in the arctic and sub-arctic regions in CLM4 (the half-meter tall broadleaf deciduous shrub), while some other vegetation models (e.g. Kaplan and New 2006) also represent prostate dwarf shrub, erect dwarf shrub, and low and high shrubs. The simulations were performed using three model configurations (see below) and three idealized representations of the NHL vegetation north of 60 • N: (1) the standard present-day vegetation distribution (denoted as CTL); (2) greener NHL where half-meter tall broadleaf deciduous shrubs replace bare ground (denoted as S for short shrubs; figure 1); and (3) greener NHL where 2 m tall shrubs replace bare ground and present-day shrubs (denoted as T for tall shrubs) 5 . In the latter two scenarios, the total shrub fraction was increased from 32 to 51% of the area north of 60 • N (figure 1), while the other land cover types were left unchanged (32% for other vegetation types; 13% for glaciers; 3% for wetland; and 1% for lakes). The prescribed leaf and stem area indices (LAI and SAI) were the same for the T and S scenarios.
Following the design strategy of Swann et al (2010), we simulated the climate response to short and tall shrub invasions under present-day greenhouse gas conditions in both a fixed ocean (i.e., prescribed climatological monthly sea surface temperatures and sea-ice extents-experiments; FO), and an interactive ocean (sea-ice and slab ocean models-experiments; IO) 6 , with each ocean configuration having its own control experiment (CTL FO and CTL IO ). It is important to note that unless otherwise specified, we will discuss all the results in terms of changes relative to these control runs (table 1). As a first approximation, the direct atmospheric climate response to the shrub cover change can be estimated from the anomalies (S-CTL) FO and (T-CTL) FO , 5 These shrub heights, mainly chosen for studying the sensitivity of the boreal climate change to shrub stature, compare closely with field measurements (e.g. see table 1 of Sturm et al 2005a andDe Groot et al 1997 for more detailed descriptions on shrub characteristics). 6 These configurations are respectively referred to as 'E 2000' and 'F 2000' cases in the CCSM nomenclature. CCSM4 was run with the finite volume dynamical core at 1.9 • × 2.5 • resolution using year 2000 initial conditions for all components and prescribed aerosol and GHG concentrations. We integrated each simulation for 30 yr but discarded the first 17 yr that were required to reach climate equilibrium. The prescribed land cover changes are designed to study the sensitivity of the boreal climate to a large-scale shrub expansion and are not meant to be realistic representations of the future spatial distribution and/or stature of shrubs in response to global warming.
while the climate response to both shrub cover change and indirect ocean and sea-ice feedbacks can be estimated from the anomalies (S-CTL) IO and (T-CTL) IO (Swann et al 2010).
We also included a pair of 2×CO 2 simulations to estimate the climate response to simultaneous shrub invasion, interactive ocean, and radiative doubling CO 2 ((S 2×CO 2 -CTL) IO and (T 2×CO 2 -CTL) IO ) that can be compared to the changes arising from a 2 × CO 2 control run (CTL 2×CO 2 -CTL) IO performed in the absence of shrub cover change. We analyzed the atmospheric climate and land feedbacks associated with these changes, with an emphasis on local and regional surface and sub-surface energy and moisture fluxes. We employed a two-tailed Student's t-test with an estimated total of 13 degrees of freedom for both 13 yr periods (see footnote 6) to test the statistical significance of the mean difference between two experiments at the 99% confidence level (p value < 0.01, critical t value = 3.012).

Exposed vegetation
Because the exchanges of moisture and energy between the vegetation and the atmosphere are regulated by plant physiology and morphology (Farquhar and Sharkey 1982), it is important to assess the timing and amplitude of exposed vegetation area. In all of our simulations, the snow height averaged over land north of 60 • N reaches an annual maximum of about 0.75 m (figure 2). This depth implies that, on average, short shrubs are buried by snow in winter, but protrude above the residual snow layer from spring to fall. The length of time during which the shrubs protrude above the snow varies across our simulations. In contrast, the 2 m tall shrubs remain, on average, above the snow all year long in all our scenarios (for pictures of typical shrub and snow interactions, see figure 3 of Sturm et al 2005a). Note that snowfall and snow depths tend to be biased high across much of the northern high latitudes in CCSM4 (Lawrence et al 2011a). If these biases were not present the short shrubs would protrude above the snow more frequently than we predict at the end of winter ( (b)). The amount of exposed leaves and branches is highest in the tall shrub experiments (T-CTL) FO/IO , with no additional changes occurring when CO 2 is doubled. In those runs, leaves are exposed from April to November and branches are visible all year. In contrast, the leaves of short shrubs (S-CTL) FO/IO only begin to be exposed in May, while a comparatively smaller number of branches protrude from May through December. These changes in ELAI and ESAI substantially exceed those induced by doubling CO 2 alone (CTL 2×CO 2 -CTL) IO , but the effects are approximately additive when the two forcings are combined (S 2×CO 2 -CTL) IO .

Tall shrub invasion (current climate; fixed ocean)
In the case of tall shrub expansion (and relative to the control run) under the current climate and with a fixed ocean (T-CTL) FO , the annually averaged land temperature poleward of 60N increases by 1.84 • C, with a peak at 4.75 • C in May ( figure 3(d), green area). In winter and early spring, only the shrub branches mask the snow (figure 3(c)) and reduce the albedo. The decline in albedo amplifies further as snow melts (exposing more branches), and plants leaf-out 7 , and exhibit two troughs in May and November associated with plant phenology (leaf-out and leaf senescence) and the seasonal evolution of snow depth and fractional cover. The decrease in albedo is however the weakest in mid-summer and is a simple function of the difference between the darker shrubs and the brighter tundra. This change in the seasonal evolution of albedo affects the energy balance and temperature anomalies accordingly when sunlight returns after the polar night, with more absorbed solar radiation (ASR) at the surface from February to September (and with a peak in May coinciding with maximum albedo decline between the two simulations; figures 3(e) and (f)). The increase in ASR is then available for increasing the surface sensible heat flux (albedo feedback), evaporation, and transpiration (ET feedback).
Regarding the ET feedback, before leaf-out, the rise in ASR induced by the exposed branches mainly increases ground evaporation (figure 3(h)). As leaves emerge, ground evaporation is progressively replaced by the transpiration process (figure 3(g)), since plants can more readily access soil water and meet atmospheric moisture demand through transpiration. The top soil layer becomes drier (despite more plentiful precipitation owing to local recycling, but accentuated by transpiration and the deeper active layers that increase drainage; figure 3(j)), and ground evaporation decreases. The combined soil evaporation and transpiration changes produce a large increase in latent heat flux in spring and early summer (figure 3(i)), which amplifies the atmospheric water vapor content, and thus the greenhouse 7 In this version of the model, LAI and SAI are prescribed with an annual cycle, which can lead to unrealistic situations where plants can leaf-out before snow melt is complete, with a reduction of albedo too early in the season. The seasonality of LAI and SAI have been improved in Lawrence et al (2011b) using data derived from MODIS, but these modifications do not prevent an unrealistic length of the snow season in CAM, with leaf-out possibly occurring prior to the snow-free season. This model limitation could be alleviated by running the land model with inclusion of a carbon-nitrogen (CN) cycle and prognostic phenology; but this will not be a viable option until the model representation of Arctic vegetation is improved. warming of the lower atmosphere during those months (figures 3(m) and (n)). The ratio of rain to snow precipitations increases, as does the ground liquid/ice water fraction near the surface (not shown).

Short shrub invasion
In this section, we discuss the effects of a short shrub invasion relative to the results presented in section 3.2 for the tall shrub case. In the short shrubs runs (S-CTL) FO , both albedo and ET feedbacks are weaker and delayed until the stems protrude above the snow and leaves appear. The resulting temperature change is much smaller (+0.66 • C) than in the tall shrub case. From January to April, the snow masking mechanism is ineffective because the snow buries both bare and shrub-covered ground. Later, the decline in albedo is weaker and peaks later in the spring when snow has melted down to depths less than the shrub height. The resulting increase in ASR is smaller and occurs later, peaking in June and substantially limiting the increase in sensible heat and the albedo-induced warming response. The ET feedback is also weaker than the tall shrub case throughout the year. Before leaf-out, the small ASR change stimulates less surface evaporation than in the tall shrub case. After leaf-out, the increase in transpiration is limited by the number of exposed leaves, and is primarily offset by a large decline in soil evaporation due to higher shading of the surface. This decline is larger than in the tall shrub case, which is probably related to the effect of vegetation height on surface roughness. Taller shrubs are aerodynamically rougher and therefore favor higher turbulent transfer of latent heat from surface to atmosphere, which limits the impact of soil surface shading on ground evaporation. These results also indicate that the land albedo and ET feedbacks are not independent processes, since the timing and magnitude of the former directly affect the timing and magnitude of the latter.

Interactive ocean effects
In experiments where the ocean and sea-ice are interactive, the warming and atmospheric moistening over land lead to a significant decrease in sea-ice extent and an associated increase in ocean latent heat flux presumably because the anomalous heat and moisture is advected over the Arctic Ocean. These changes over the Arctic Ocean contribute to further increases in atmospheric water vapor and associated greenhouse warming (Swann et al 2010). There are only slight additional feedbacks related to this sea-ice loss over land for fields such as albedo, ASR, transpiration and soil evaporation ( figure 3, cyan areas). The sea-ice signal is most distinct from June to December, being larger, and peaking earlier, when tall shrubs are present. The inclusion of interactive sea-ice extends the impact of increasing shrub abundance throughout the year, with the direct shrub-induced feedback dominating during the warm season and instigating the indirect ocean and sea-ice-driven feedbacks that dominate during the cold season.

Doubled CO 2 effects
As predicted in future climate projections (IPCC), the climate will be much warmer under a doubled CO 2 scenario, accompanied by a large sea-ice loss and a substantial increase in ocean latent heat flux and atmospheric moisture content (Santer et al 2007). We find that independently of shrub height, the climate response to doubling CO 2 and shrub invasion is generally similar to the sum of the climate responses arising from the individual forcings (not shown). We also find that the sensitivity of the NHL climate to shrub height is slightly reduced under a doubled CO 2 scenario ((T 2×CO 2 -S 2×CO 2 ) IO ) because the GHG-induced reduction of snow depth weakens the distinction between snow masking by short or tall shrubs. One implication of this interaction is that the decline in albedo in the short shrub case (S 2×CO 2 -CTL) IO peaks a month earlier (in May) than in the current climate, substantially enhancing the ASR, sensible heat flux, spring ground evaporation and summer transpiration. In these experiments, soil moisture increases and the wetter root-zone soil does not limit ground evaporation during the growing season. Further, the change in surface temperature arising from the tall shrub feedbacks alone (T 2×CO 2 -CTL) FO is as large as that imposed by doubling CO 2 (CTL CO 2 -CTL) IO without additional shrubs. We show in section 3.6 that in contrast to the surface temperature case, this statement does not hold for atmospheric or soil column temperatures.

Permafrost vulnerability
Here we extend the analysis of the effects of shrub expansion to the sub-surface climate regime. The representation of soil thermodynamics in the CCSM model has been greatly improved in recent years due to an explicit representation of the thermal and hydrological properties of organic soil as well as a deeper soil column 8 (Nicolsky et al 2007); the annual cycle of simulated soil temperature compares reasonably well with observations (Lawrence et al 2008a(Lawrence et al , 2011c. Figure 4 represents the annual cycle in temperature as a function of atmospheric height and soil depth (down to 5 m), for each individual simulation. The active layer thickness (ALT) deepens with the invasion of shrubs (compare figures 4(b) and (a), and figures 4(c) and (a)), in response to the associated changing climate. When the shrub stature increases and an interactive ocean is used, ALT deepens further, and its below-freezing season shortens (compare figures 4(e) and (a)). In experiments where the shrub expansions are prescribed under a scenario with doubled CO 2 (figures 4(g) and (h)), the refreezing of the soil only occurs in the top meter of soil. Below this level the soil no longer refreezes and this is more likely to favor the formation of taliks (i.e., layers of year-round unfrozen ground that lie in the permafrost zone), which can very efficiently accumulate soil heat content (O'Connor et al 2010, Lawrence et al 2008b).
While we are aware that the model snow bias (section 3.1) tends to favor warm soil temperature and the formation of taliks in the simulations, we also found that the regions where deep layers of permanent unfrozen ground are found underneath a frozen surface (i.e. the regions of potential talik formation) appear to move poleward in response to introducing shrub and greenhouse warming (not shown).
The difference plots reveal information about the seasonality of such changes. The presence of shrub cover (S-CTL) FO induces a seasonal increase in atmospheric heating which favors summer thawing and reduces winter refreezing (figures 4(i) and (j)). The heat penetrates the ground during May-June for the short shrub case, and April-May in the tall shrub case (figure 3(r)), accumulates greatly (figure 3(p)), and is not fully returned to the atmosphere during the cold months. However, the vertical heat flux to the ground is reduced during June-July, especially in the tall shrub experiment due to the increase in vegetation cover (which prevents the solar radiation from reaching the surface and reduces the air-soil temperature gradient) and the transpiration-induced desiccation of the organic soil (O'Donnell et al 2009 have shown that an decrease in moisture in organic soil lowers its thermal conductivity, and therefore the heat transfer). These results suggest that a seasonal atmospheric warming can modify the ground thermal regime all year round.
When tall shrubs and an interactive ocean are imposed simultaneously, the atmosphere becomes warmer (figure 4(l)) and wetter all year (figure 4(s)), which extends the period of ground-atmosphere thermal coupling. With the wetting of the soil, more energy can penetrate the ground at greater depths in spring, while the penetration of summer heating is limited by the increase in dryness of the top soil layer. When the atmospheric CO 2 is doubled, the atmosphere again becomes warmer (figures 4(m)-(o)) and its humidity increases (figures 4(t)-(v)). Meanwhile, as soil ice melts and precipitation increases (figure 3(t)), the ground becomes moister in early spring (figures 4(t)-(v)) and increases its heat conductivity, making its thermal regime more responsive to the atmospheric heating 9 . The increased total liquid groundwater storage (aquifer water and saturated soil water; figure 3(q)), which induces a higher heat content of the soil 10 , and the formation of taliks both favor the accumulation of heat in the ground. Overall, the hydrological response due to permafrost thaw remains complex and poorly represented in large-scale models. . Climatological annual cycle in air and soil temperature (in • C) as a function of height and depth for each simulation ((a)-(h)) and the difference relative to the CTL for each simulation ((i)-(o)). Annual cycle in soil water content (mm 3 mm −3 ) and atmospheric specific humidity (in 10 −3 kg kg −1 ) for different anomalies ((p)-(v)). The black line outlines statistically significant anomalies at the 99% confidence level.

Discussion and conclusion
This study is the first to investigate the biophysical effects of deciduous shrub expansion and height on NHL climate. The idealized simulations have been designed to separate the effects of shrub expansion (short or tall) from the effects arising from an active ocean and sea-ice, or from doubling CO 2 (table 1).
As in a previous analysis of increased tree cover at high latitudes, we found that the large-scale shrub expansion triggered substantial regional atmospheric warming in spring and summer through direct albedo and ET feedbacks. We also found that the strength and timing of these two feedbacks are very sensitive to the stature of the expanding shrubs: when tall shrubs (instead of short shrubs) are prescribed, the sensible heat flux, pre-leaf-out soil evaporation and post-leaf-out transpiration increase sooner. These effects, in turn, impact the strength of the contributions of the indirect sea-ice albedo and ocean evaporation feedbacks to additional regional warming during the winter months. Comparatively, the imposed shrub expansion has a stronger albedo feedback, but a weaker ET feedback than doubling CO 2 (cyan and brown lines of figures 3(f) and (m)). Finally, we found that the invasion of tall shrubs systematically increases sub-surface soil temperature, deepens the active layer, shortens the length of the sub-freezing season, and triggers permafrost degradation more substantially than do short shrubs. The active soil layer deepens further with inclusion of an interactive ocean. In summary, we found that the amplitudes of the warming and other climatic changes are highly dependent on the stature of the invading shrubs, with all feedbacks being more intense for taller shrubs.
We caution that several pertinent mechanisms are not fully represented in this modeling study. For example, LAI and SAI are prescribed, so the effect of CO 2 fertilization of plants, which can alter plant photosynthesis rates and therefore ET through LAI and SAI (if they were prognostic), was not taken into account with respect to vegetation structure (e.g., canopy height or LAI) or phenology (i.e., timing of leaf-out and senescence). Furthermore, bending of thin and supple stems has been observed for some short shrubs and in this situation could keep those shrubs buried and protected by snow (Sturm et al 2001a(Sturm et al , 2005a, thereby reducing their estimated impact on climate. We also did not consider the 'snow fence effect'. Some previous observational studies have found that snow preferentially accumulates around shrubs, thus producing a more effective insulating snow layer which can potentially affect the permafrost and stimulate soil microbial activity (Sturm et al 2005b).
This study, and that of Lawrence et al (2011b), prescribed substantially different distributions of hypothetical shrub area increase. Despite the differences, both studies indicate that the warming response to a large-scale expansion of shrubs in the NHL could overwhelm the soil cooling effect due to shading (Blok et al 2010, Yi et al 2007, leading to increased rather than decreased vulnerability of permafrost. Although not analyzed here, these physical feedbacks would likely impact ecosystem biogeochemical processes (e.g., soil decomposition, plant growth) controlling the exchanges of carbon between soil and atmosphere.
To summarize, our study highlights the need to account for the expansion, stature, and phenology of invading vegetation to improve climate prediction in the NHL regions. This study also highlights the need of augmenting the number of CCSM arctic and sub-arctic shrub types, improving their representation of fractional cover and height, and treating the shrub expansion more realistically with a dynamic vegetation component in the model. Dye